The Notions of Literal and Non-literal Meaning in Semantics and Pragmatics Dissertation zur Erlangung des akademischen Grades doctor philosophiae (Dr. phil) Eingereicht an der Philologischen Fakultät der Universität Leipzig von Kristin Börjesson Verleihungsbeschluss 10. Oktober 2011 Gutachter: PD Dr. Johannes Dölling Institut für Linguistik, Universität Leipzig Prof. em. Dr. Anita Steube Institut für Linguistik, Universität Leipzig Prof. Robyn Carston, PhD Research Department of Linguistics, University College London ii Contents 1 Introduction 1.1 Standard Notions . . . . . . . . . . . 1.2 Are the Standard Notions Adequate? 1.3 Interpretation and Levels of Meaning 1.4 Aims of the Thesis . . . . . . . . . . 1.5 Thesis Plan . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Against the Standard Notions of Literal Meaning and Non-literal Meaning 2.1 Literal Meaning and Context-Independence . . . . . . . . . . . . . . . 2.1.1 Literal Meaning as Compositional Meaning? . . . . . . . . . . 2.1.2 Literal Meaning as Context-Independent? . . . . . . . . . . . . 2.1.3 Literal Meaning as Primary to Non-literal Meaning? . . . . . . 2.2 Non-literal Meaning and Conventionality . . . . . . . . . . . . . . . . 2.2.1 Empirical Evidence . . . . . . . . . . . . . . . . . . . . . . . . 2.2.2 Theoretical Considerations . . . . . . . . . . . . . . . . . . . . 2.3 Consequences for Lexical Meaning . . . . . . . . . . . . . . . . . . . . 2.3.1 Problematic Data . . . . . . . . . . . . . . . . . . . . . . . . . 2.3.2 Approaches to Meaning in the Lexicon . . . . . . . . . . . . . 2.3.3 Semantic Underspecification in the Lexicon . . . . . . . . . . . 2.4 Empirical Investigations of Aspects of Semantics . . . . . . . . . . . . 2.4.1 Polysemy vs. Underspecification in the Lexicon . . . . . . . . 2.4.2 Empirical Evidence for Semantic vs. Pragmatic Processing . . 2.5 Why the Standard Notions? . . . . . . . . . . . . . . . . . . . . . . . 2.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Utterance Meaning and the Literal/Non-literal Distinction 3.1 Levels of Meaning . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1.1 Grice’s Four Types of Meaning . . . . . . . . . . . . . . 3.1.2 Bierwisch’s Three Levels of Meaning . . . . . . . . . . . 3.1.3 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2 The Problem of Characterising the Level of Utterance Meaning 3.2.1 Explicit/Implicit Meaning . . . . . . . . . . . . . . . . . 3.2.2 Unarticulated Constituents vs. Hidden Indexicals . . . . 3.2.3 Minimal Semantic Content and Full Propositionality . . 3.2.4 Minimal Proposition vs. Proposition Expressed . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 1 3 5 6 6 9 9 10 14 19 23 23 29 35 36 41 51 67 67 71 75 78 83 84 84 88 92 94 95 106 115 123 iv Contents 3.3 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129 4 Utterance Meaning and Communicative Sense – Two Levels or One? 133 4.1 Problematic Phenomena . . . . . . . . . . . . . . . . . . . . . . . . . 134 4.1.1 Metaphor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134 4.1.2 Irony . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151 4.1.3 Conversational Implicatures . . . . . . . . . . . . . . . . . . . 162 4.1.4 Speech Acts . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180 4.2 Differentiating What is Said from What is Meant . . . . . . . . . . . 184 4.2.1 What is Said/What is Meant and Indirect Speech Reports . . 186 4.2.2 Primary vs. Secondary Pragmatic Processes . . . . . . . . . . 189 4.2.3 What is Said/What is Meant and Distinct Knowledge Systems 194 4.3 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200 5 Varieties of Meaning, Context and the Semantics/Pragmatics Distinction 203 5.1 Towards an Alternative Characterisation of (Non-)Literal Meaning . . 203 5.1.1 Literal Meaning and Types of Non-literal Meaning . . . . . . . 205 5.1.2 Literal Meaning as ‘Minimal Meaning’ . . . . . . . . . . . . . 211 5.1.3 Nature of the Processes Determining (Non)-Literal Meaning . 215 5.1.4 (Non-)Literal Meaning as (Non-)Basic Meaning . . . . . . . . 221 5.2 The Nature of Context in Utterance Interpretation . . . . . . . . . . 227 5.2.1 Context and the Interpretation of Implicit Meaning Aspects . 228 5.2.2 Context, Semantic Interpretation and the Semantics/ Pragmatics Distinction . . . . . . . . . . . . . . . . . . . . . . . . 242 5.3 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 251 6 Summary 255 Bibliography 259 Chapter 1 Introduction 1.1 Standard Notions One of the major issues in investigating the relation of language and meaning is the question of how to characterise and draw the line between what traditionally are called semantics and pragmatics. In describing what they take to be the characteristics of one or the other system, linguists often make use of the terms literal meaning and non-literal meaning. For example, Lyons (1987) lists a number of propositions used in the differentiation of semantics from pragmatics, amongst which is the following: ‘. . . that semantics deals with literal, and pragmatics with non-literal, meaning . . .’ (ibid., p. 157). Similarly, Cole (1981, p. xi) states that semantics ‘. . . is involved in the determination of conventional (or literal) meaning . . . ’, whereas pragmatics is concerned with ‘. . . the determination of nonconventional (or nonliteral) meaning . . . ’ and Kadmon (2001, p. 3) writes ‘. . . I think that roughly, semantics only covers “literal meaning.” Pragmatics has to do with language use, and with “going beyond the literal meaning.”’. More recently, Recanati (2004) summarised (and criticised) the standard view on the division of labour between semantics and pragmatics, starting with the following sentence. ‘Semantics deals with the literal meaning of words and sentences as determined by the rules of the language, while pragmatics deals with what the users of the language mean by their utterances of words and sentences.’ (Recanati 2004, p. 3). Actually, the pair of terms literal meaning/non-literal meaning is only one of quite a number of dichotomies used in the characterisation of semantics and pragmatics. Thus, the two systems are often characterised in terms of the differentiation between conventional vs. non-conventional meaning, as, e.g. in the quote from Cole (1981) above. Similarly, in Lyons (1987) one finds the proposition ‘. . . that semantics has to do with conventional, and pragmatics with the non-conventional, aspects of meaning . . .’ (ibid., p. 157). Another important pair of terms traditionally used is context-independent vs. context-dependent meaning. Thus, Lyons (1987, p. 157) states ‘. . . that semantics deals with context-independent, and pragmatics with context-dependent, meaning’. More specifically, Katz (1977) introduces the notion of the ‘anonymous letter situation’ to characterise the kind of meaning captured 2 Introduction by semantics. ‘[I] draw the theoretical line between semantic interpretation and pragmatic interpretation by taking the semantic component to properly represent only those aspects of the meaning of the sentence that an ideal speaker-hearer of the language would know in an anonymous letter situation, . . . [where there is] no clue whatever about the motive, circumstances of transmission, or any other factor relevant to understanding the sentence on the basis of its context of utterance.’ (Ibid., p. 14) To summarise, whereas semantics is characterised as dealing with literal, conventional and context-independent meaning, pragmatics deals with non-literal, nonconventional and context-dependent meaning. More generally, the standard notions of semantics and pragmatics may be described as follows. Semantics deals with those aspects of meaning that both simple and complex expressions have, independent of their use. In contrast, pragmatics deals with those aspects of meaning that are determined by the actual use of language.1 That is, semantics is concerned with meaning that is independent of any specific context, whereas pragmatics specifically draws on contextual information for the interpretation of some expression. Assuming that the borderline between semantics and pragmatics is fixed and stable, using the dichotomies mentioned above in the characterisation of the respective systems suggests that there is a correspondence between literal, conventional and context-independent meaning, on the one hand, and non-literal, non-conventional and context-dependent meaning on the other.2 However, one usually does not find actual characterisations of the kind of meaning picked out by the terms literal meaning and non-literal meaning other than such that again (more or less explicitly) relate the two notions back to the semantics/pragmatics distinction. For instance, Bach (2001a) writes ‘Words do not have nonliteral meanings [. . . ], but they can be used in nonliteral ways. [. . . ] In familiar cases, such as metaphor and metonymy, particular expressions are used nonliterally. [. . . ] But there is a different phenomenon which I call “sentence nonliterality,” [. . . ] Here a whole sentence is used nonliterally, without any of its constituent expressions being so used.’(Ibid., p. 249, my emphasis). Thus, whereas literal meaning is a feature that expressions are said to have, the nonliteral meaning of an expression results from the particular use of that expression. Moreover, not only simple but also complex expressions can be used non-literally. In both cases, the non-literal meaning results from the actual use of a certain expression in a specific context. In addition to the circumscription ‘non-literal use of language’, one finds descriptions such as ‘non-literal utterances’ (cf. Carston 2005) or ‘nonliteral interpretation’ (cf. Papafragou 1996, Ariel 2002, Carston 2005). 1 Cp., again, Lyons (1987) who mentions the idea that ‘. . . semantics has to do with meaning, and pragmatics with use . . .’ (Ibid., p. 157). 2 From the quotes given above, this is especially apparent in Cole’s, who uses the terms literal and non-literal as synonymous to conventional and non-conventional, respectively. Are the Standard Notions Adequate? 3 Generally, it seems that the two terms literal meaning and non-literal meaning are treated as denoting basic kinds of meaning that are intuitively clear and as such need no further description. However, although there are no clear definitions of the terms literal meaning and non-literal meaning given in the literature concerned, the fact that the two terms are used amongst others in a dichotomous characterisation of semantics and pragmatics suggests that these terms may also be used in characterising literal meaning and non-literal meaning as such (or the other way around as done by Cole). This latter characterisation has led to what might be called the standard notions of literal meaning and non-literal meaning which are summarised in what follows. Literal meaning, on the one hand, is assumed to be conventionalised, that is, it does not take any special interpretation effort to arrive at it. The literal meaning of simple expressions is listed in their lexical entries; the literal meaning of complex expressions is the result of a principled combination of the literal meanings of their parts. Thus, both the literal meaning of simple as well as complex expressions is characterised by the fact that it is context-independent. Non-literal meaning, on the other hand, is assumed to be non-conventionalised, thus, it does take a special interpretation effort to arrive at it. Intuitively, it is considered as deviating from some more basic (literal) meaning in a fairly special way. Overall, the term nonliteral meaning is used to differentiate from literal meaning a kind of meaning that is derived from the latter and, in a sense, has a secondary status. Therefore, it is traditionally assumed that in terms of the enfolding of the interpretation process, the literal meaning of an expression is processed first, whereas any potential nonliteral meanings are processed afterwards and only if the literal interpretation does not fit the given context. Finally, in accordance with the standard characterisations of semantics and pragmatics given above, literal meaning is traditionally assumed to belong to the domain of semantics (as it can apparently be determined without reference to any particular context of utterance), whereas non-literal meaning is assumed to be properly investigated by pragmatics (as it seems it can only be determined by taking into account the context of an utterance). Such a classification presumes that the definitions of the terms literal meaning and non-literal meaning, as well as the characterisation of what semantics and pragmatics are concerned with are indisputably clear. However, the ongoing debate about the proper demarcation of semantics from pragmatics show that this is not at all the case (Carston 1999, Levinson 2000, Bezuidenhout and Cutting 2002, Bianchi 2004, Carston 2004a, Recanati 2004, Cappelen and Lepore 2005, Dölling 2005, Carston 2007, etc.). 1.2 Are the Standard Notions Adequate? Note that although there is something like a standard characterisation for the terms literal meaning and non-literal meaning, they are not consistently used in accord with these characterisations. For instance, the term literal meaning is not only used in connection with such phenomena as are considered totally context-independent. Thus, Carston (2007) refers to the ‘. . . literal meaning of [a speaker’s] utterance’ (ibid., p. 21). Similarly, Recanati (1995) speaks of ‘. . . the literal interpretation of 4 Introduction an utterance (the proposition literally expressed by that utterance). . . ’ (ibid, p. 2) and Sag (1981) speaks of the ‘. . . propositional content of an utterance (i.e., its literal meaning) . . . ’ (ibid, p. 274-5). What makes such descriptions interesting in the present context is that what the authors refer to by the description ‘literal meaning of an utterance’ does not necessarily correspond to the context-independent meaning of the expression used to make the particular utterance, since, as we will see, this need not actually be a full proposition. Rather, they use the term literal here to differentiate the meaning of some expression in a particular context from the meaning that the speaker actually intended to express (or at least is taken by the hearer to have expressed). Bezuidenhout and Cutting (2002, p. 435) make a similar observation when they say that ‘[t]he phrases “literal meaning” or “literal interpretation” have been used to cover both the literal meaning of a sentence and what is said by the utterance of a sentence in a context.’ This is also noted by Korta and Perry (2008), who write that ‘[w]hat is said has been widely identified with the literal content of the utterance . . .’. Generally, the term literal meaning is used with respect to a level of meaning which can no longer be said to be context-independent. It should be noted that it is authors who reject the standard characterisation of semantics and pragmatics that use the terms literal meaning and non-literal meaning in a non-standard understanding. The problem is that they do so implicitly. That is, these authors do not explicitly say anything new concerning the understanding of the terms literal meaning and non-literal meaning. As Wilson and Sperber (2000, p. 250) put it, ‘[t]he notion of literal meaning, which plays such a central role in most theories of language use, is unclear in many respects’. Note also that it is not only the pair of terms literal (meaning)/non-literal (meaning) that is employed differently in the various approaches. Such notions as (non)-conventionality and context-(in)dependence are problematic as well. Thus, the use of the pair of terms conventional vs. non-conventional as exemplified above suggests that conventionality is an all-or-nothing property. However, as is suggested by the results of various experiments investigating the nature of the interpretation process on the one hand (cf. Giora 1997, 1999, Gibbs 2002), as well as theoretical considerations within the field of historical semantics on the other (cf. Busse 1991), this view is an oversimplification of the facts. Similarly, not all approaches that are characterised as essentially semantic by their proponents necessarily share the view that what semantics deals with is context-independent meaning only (cf. Sag 1981, Borg 2004b, Cappelen and Lepore 2005). The problem that arises is that if the terms in which the notions literal meaning and non-literal meaning are defined are put under closer scrutiny, it might turn out that they do not capture the aspects that we intuitively and pre-theoretically associate with the notions of literal meaning and non-literal meaning. Another problem is that with only the standard notions of literal meaning and non-literal meaning to rely on, it is no trivial question to ask how these two meaning aspects are related to other kinds of meaning aspects identified in the individual approaches, such as e.g., explicit/implicit meaning aspects of an utterance, so-called ad-hoc concepts or conversational implicatures. For instance, metaphor – one particular type of non-literal meaning – has actually been treated in terms of conversational implicature and so has irony. In fact, conversational implicatures in general have been characterised as Interpretation and Levels of Meaning 5 non-literal (cf. Recanati 2004), although, intuitively at least – and especially if one actually does not count metaphor and irony amongst them – this is a questionable claim. 1.3 Interpretation and Levels of Meaning A further point to note is that the assumption that two systems, namely semantics and pragmatics, are involved in the interpretation of natural language utterances suggests that there is also only a differentiation of two levels of meaning necessary. Thus, the semantic component takes the lexical meanings of the individual expressions used in an utterance and combines them in a principled manner, resulting in the truth-conditional or semantic meaning of the complex expression used in an utterance, traditionally taken to be a proposition. This semantic meaning, then, is the input to the pragmatic component which determines what the speaker who uttered the expression in question actually meant in a particular situation. This is the view usually taken to underlie Grice’s differentiation of the two levels of meaning what is said and what is meant.3 In Grice’s terminology then, literal meaning is part of the level of what is said, whereas non-literal meaning is to be found only at the level of what is meant. However, it has been pointed out both in Neo- as well as Post-Gricean approaches that Grice’s notion of what is said is problematic. More specifically, it has been argued that what is said cannot both be the semantic meaning of an utterance as well as the basis for drawing conversational implicatures (cf. Sperber and Wilson 1995, Levinson 2000, Carston 2002c, Recanati 2004). As a consequence of such problems of Grice’s approach, more than two levels of meaning have been assumed: a level that captures the semantic meaning of an expression, a context-dependent level that can function as the basis for further pragmatic inferences to be drawn (call this what is said for now) and the level of what is meant. What is interesting to note with respect to the differentiation of three levels of meaning is the various possibilities this opens up for the classification of literal meaning and non-literal meaning. Thus, with the differentiation of semantic meaning from what is said, the question arises where ‘to put’ literal meaning: is it part of the level of semantic meaning or part of the level of what is said? The quotations above suggest that different linguists hold different views concerning this question. Note also that a similar question can be asked concerning non-literal meaning. That is, with the differentiation of the two distinct levels of contextsensitive meaning what is said and what is meant, the question arises whether nonliteral meaning aspects exclusively belong to the latter level or whether they might not already arise at the former. Note that if both literal meaning and non-literal meaning are assumed to arise at the level of what is said already, then this would suggest that the two types of meaning are not as different from one another as traditionally assumed. However, the various approaches to the interpretation of natural language differ in the characterisations they assume for the individual levels of meaning identified. 3 It should be noted that in his efforts to explicate the notions of saying vs. meaning, Grice actually contemplated four different types of meaning. For more on this issue see chapter 3. 6 Introduction Again, the different characterisations are dependent on the particular views held regarding the nature of the semantic and pragmatic components, the roles they play in the overall interpretation process as well as the type of information they are assumed to have access to. 1.4 Aims of the Thesis I want to show that the standard notions of literal meaning and non-literal meaning are not consistent and that, as a result, the two terms cannot be used to differentiate between the types of meaning aspects semantics and pragmatics are usually assumed to deal with. In contrast, I want to develop a characterisation of literal meaning and non-literal meaning that is consistent and captures the various uses the two terms are put to. In addition, I want to defend the view that differentiating between two context-dependent levels of meaning what is said/utterance meaning and what is meant/communicative sense is possible and necessary. In particular, I will propose a modelling of the process that provides implicit meaning aspects to the level of what is said/utterance meaning that allows viewing this process – as the processes contributing to this level of meaning in general – as operating independently of considerations of speaker intentions. Finally, I want to show that a differentiation of semantics from pragmatics that implicitly relies on a logical or temporal ordering of the two systems is inappropriate. Rather, the processes constituting these two systems should be seen as operating in tandem. 1.5 Thesis Plan The thesis is structured as follows. In chapter 2, I will argue against the standard notions of literal meaning and non-literal meaning. In particular, I will argue against the traditional characterisation of literal meaning and non-literal meaning, according to which the former is taken to be context-independent and the latter non-conventional. Having established that literal meaning does not necessarily have to be taken to be context-independent and as such semantic in nature, I will discuss the consequences this view has for the nature of lexical meaning. After reviewing a number of different types of approaches to lexical meaning, I will argue for a view that assumes a high degree of underspecification of lexical meaning. Generally, in the discussions in chapter 1, I will consider both theoretical viewpoints as well as empirical data. In particular, one section is dedicated to empirical studies on aspects of the semantics component, namely that lexical meaning is characterised by underspecification and that, generally, semantic processes of meaning construction should be differentiated from pragmatically based plausibility checks. In the last part of chapter 1, I will try to answer the question of why the standard notions of literal meaning and non-literal meaning came to be assumed in the first place. Here, the idea of stereotypical interpretations of linguistic expressions presented ‘out of context’ will be considered. Having argued against the standard notions in chapter 2, and more specifically, having argued for viewing literal meaning, similarly to non-literal meaning as essen- Thesis Plan 7 tially context-dependent as well, chapter 3 is dedicated to looking in detail at the first context-dependent level of meaning called what is said by Grice, to see how this has been characterised subsequently and to identify the processes potentially involved in determining literal meaning at this level of meaning. I will start with Grice’s differentiation of four different types of meaning and relate them to the two levels of meaning Grice introduced: what is said and what is meant. Following that, I will present Bierwisch’s threefold differentiation of levels of meaning, based on the different knowledge systems made use of in their determination. In the second part of chapter 3, I will discuss a range of approaches that give alternative characterisations for Grice’s level of what is said. The overall aim is to identify the different processes at work in determining what is said, how these processes are characterised and which types of meaning aspects can be found at this level of meaning (appart from potentially literal or non-literal meaning). At the same time, the various approaches discussed also all offer slightly different views on the nature of the semantics and pragmatics components and how they interact in the process of utterance interpretation. While the greater part of chapter 3 is taken up by theoretical considerations, towards the end of that chapter a few empirical results will also be discussed. Chapter 4, then, is concerned, on the one hand, with phenomena traditionally assumed to arise at Grice’s level of meaning what is meant, and, on the other hand, with the more basic question of whether a differentiation of two context-dependent levels of meaning what is said and what is meant actually is necessary/possible. Thus, in the first part of chapter 4, alternative approaches to the phenomena of metaphor, irony, (primarily generalised) conversational implicature and (primarily indirect) speech acts will be reviewed as well as empirical results considered that test the predictions following from the individual approaches. Here, the aim is to establish, on the one hand, how these different meaning aspects are determined and, on the other hand, which of the phenomena actually can be usefully considered as non-literal. More generally, the question is addressed at which level of meaning (i.e. what is said or what is meant) the individual phenomena should be taken to arise. In the second part of chapter 4, various arguments will be presented for and against differentiating the two levels what is said and what is meant from one another. I hope to make clear that such a differentiation is useful and necessary, although it might be difficult to decide on the criteria to be used in this differentiation. Chapter 5, finally, turns back to the basic question that chapter 2 ends with, namely how literal meaning and non-literal meaning actually should be characterised if one wants to capture the various uses the two terms are put to. I will start out with two alternative characterisations of what literal meaning and non-literal meaning should be taken to be, before presenting my own characterisation, based on the discussion in the preceding chapters. As a preliminary for my characterisation, I will review the various processes identified in the preceding chapters as involved in the overall interpretation of utterances. The main consequence drawn from my characterisation of literal meaning and non-literal meaning will be that these two notions actually cannot be used in the characterisation of the distinction between semantics and pragmatics, if the former, in contrast to the latter, is essentially taken to be context-independent. The last part of chapter 5 will take up exactly this point, namely the nature of contextual information in utterance interpretation and 8 Introduction whether the notion of context-(in)dependence actually is useful in differentiating between semantics and pragmatics. Thus, I will first offer a proposal concerning the nature of the contextual information the process of free enrichment makes use of. Free enrichment is one of the processes assumed to contribute to the level of utterance meaning and crucially is taken to depend on a consideration of potential speaker intentions for its operation. I will show that this assumption is not necessary. In the final section of chapter 5, I will turn back to the characterisation of the semantics/pragmatics distinction and after discussing a number of views on that characterisation present my own. Chapter 2 Against the Standard Notions of Literal Meaning and Non-literal Meaning In this chapter, I will argue against the standard notions of literal meaning and non-literal meaning described in chapter 1 (sections 2.1 and 2.2, respectively). The main aim of this chapter is to show that the dichotomies traditionally used to differentiate literal meaning from non-literal meaning either cannot in fact differentiate the two meanings (as is the case with the feature of context-(in)dependence) or are not such ‘all-or-nothing’ concepts as traditionally implied (as is the case with the property of conventionality). Generally, the arguments presented point to the crucial conclusion that literal meaning and non-literal meaning are in fact not so different from one another as traditionally assumed. Having argued against viewing literal meaning as essentially context-independent and non-literal meaning as essentially non-conventional, I will consider the consequences this has for the nature of lexical meaning (section 2.3). Moreover, I will consider empirical evidence supporting the assumption of underspecification of lexical meaning and, more generally, a distinction between distinctly semantic and pragmatic processes in interpretation (2.4). In addition, I will address the question of why the standard assumptions came into existence in the first place (section 2.5). 2.1 Literal Meaning and Context-Independence Traditionally, complex expressions are assumed to have literal meaning in the form of what formal semantics1 calls sentence meaning, which results from the process of semantic composition which combines the literal meanings of the simple expressions that together constitute the complex expression and which captures the proposition expressed by that sentence.2 Moreover, during interpretation, the literal meaning of 1 Note that in what follows, on the semantics side, I am primarily interested in assumptions made in the programme of formal semantics. 2 In this section, I will primarely be concerned with two of the three properties literal meaning is standardly claimed to exhibit, namely that it is context-independent and primary to non-literal meaning. Thus, I will concentrate on complex expressions, leaving the discussion of the nature of 10 Against the Standard Notions of Literal Meaning and Non-literal Meaning a complex expression is computed first, whereas its potential non-literal meaning is computed afterwards and only if the literal meaning does not fit the given context (cf. Grice 1975; Searle 1979). Intuitively, these characterisations seem to be sound. They give a fairly general description of what we take to be literal meaning with respect to complex expressions. However, looking at each of the characteristics in more detail reveals that they are not unproblematic. Thus, it is questionable whether what we usually take to be a complex expression’s literal meaning does in fact correspond to its contextindependent, compositional meaning. Put differently, the question is whether the formal semantic notion of sentence meaning can be assumed to both be the sum of the lexical meanings of the simple expressions involved as well as having a fully propositional form. Furthermore, in computing the ‘speaker-intended’ non-literal meaning of an expression, it may not actually be necessary to first compute the literal meaning of the expression the speaker used as an intermediate step. 2.1.1 Literal Meaning as Compositional Meaning? Concentrating on the traditional characterisation of the programme of formal semantics and the role of literal meaning therein reveals that, in a sense, the characterisation of literal meaning and non-literal meaning is interdependent on the characterisations of semantics and pragmatics. Thus, basically, formal semantics can be characterised as dealing with the context-independent meaning of simple and complex expressions.3 More specifically, it aims at formulating truth conditions for sentences. That is, it takes as a starting point for analysis the level of sentence meaning, mainly for two reasons. First, it seems that sentences express propositions, that is, complete thoughts, something of which it makes sense to ask whether it is true or not. Second, intuitively at least, the meaning of a sentence can be grasped without any reference to an actual utterance of that sentence and is thus context-independent. It contrasts with interpretations of a sentence that can only be derived by considering the actual context in which that sentence is uttered (e.g., cases of irony or particularised conversational implicature). Thus, sentence meaning is considered literal in the sense that its derivation is independent of contextual information. Moreover, sentence meaning also is the level from which the meanings of the individual expressions involved are derived, following the principle of compositionality. And since sentence meaning is context-independent, the meanings of the simple expressions derived from it are context-independent too. They are the lexical meanings of the expressions concerned. Thus, primarily, what the term literal meaning refers to is a certain type of meaning that, intuitively, seems to differ from other types of meaning mainly by virtue of the fact that it is context-independent and fully propositional (sentence meaning). Derivatively, the term also refers to types of meaning which are not propositional, but crucially are context-independent and are derived from a full proposition via the principle of compositionality (lexical meaning). lexical meaning to section 2.3. 3 See below, however, for formal semantic approaches that also take into account contextual information. Literal Meaning and Context-Independence 11 meaning context-independent lexical m. sentence m. literal context-dependent ... ... ... non-literal Figure 2.1: Traditional differentiation of types of meaning So far so good. However, the characterisation of formal semantics as stated above has proven to be problematic. And, as we will see, these problems also extend to the characterisation given to the notion of literal meaning. Thus, to summarise: in its traditional form, three of formal semantics’ main assumptions are the following: a) semantics is concerned with the context-independent meaning of natural language expressions, b) for sentences, what is determined by the semantic component of a natural language grammar is the proposition expressed by that sentence and c) for simple expressions their semantics (or lexical meaning) is whatever aspects of their meaning remain constant across different uses of that expression. However, as Sag (1981) points out: a formal semantic theory which does not allow for any contextual information to be made use of in determining the proposition expressed by a sentence ‘. . . appears to be falsified by the mere existence of sentences containing tense morphemes or other indexical expressions.’ (Ibid, p. 274). Thus, consider the sentence in (1). (1) He went to the bank yesterday. For the sentence in example (1), it is clearly not the case, that semantics determines a truth-evaluable proposition, due to the occurrences of the context-dependent expressions he and yesterday as well as the homonymous noun bank. As is the case for all indexical expressions, the exact reference of he and yesterday differs with the contexts in which they are uttered. Thus, for such expressions semantics only gives rules for ’where to look’ in the search for potential referents. In the case of the occurrence in a sentence of homonymous expressions such as bank, the assumption is that the process of semantic composition has to build up as many different structures for the sentence, as there are ambiguous expressions in it. Thus, for (1) to express a full proposition, the references of the occurring indexical expressions have to be fixed, that is, recourse has to be taken to the context of the utterance. Moreover, the sentence has to be disambiguated, which, again, is only possible with the help of contextual information. Even then, the sentence does not express a proposition until the reference of the NP the bank to some unique location has been fixed. What this shows is that the proposition expressed by some sentence can only actually be 12 Against the Standard Notions of Literal Meaning and Non-literal Meaning determined once the context in which it is uttered is taken into consideration. Thus, it seems that the semanticist cannot uphold both assumptions a) and b). If he wants to rescue assumption a), it seems he has to concede that, in fact, the semantic component does not determine the truth-evaluable proposition expressed by a sentence; if he wants to rescue the assumption in b), he has to allow for context-sensitive processes to take place during the determination of the proposition expressed by a sentence. However, formal semantic approaches exist which attempt to capture the difference between context-sensitive and context-insensitive expressions and at the same time uphold assumptions a) and b). One such approach is Kaplan (1989b). Thus, Kaplan proposes to differentiate between, in a sense, two meanings of expressions: their character and their content.4 Consider example (2). (2) a. Mary: I am hungry. b. John: I am hungry. On the one hand, the notion of character captures the intuition that Mary and John in a way have said the same thing: both used the same sentence. The notion of content, on the other hand, captures the intuition that, at the same time, Mary and John have not expressed the same idea. Kaplan’s suggestion is that a sentence’s character is a function that takes a context in order to deliver a proposition or the content of that sentence in that context. Thus, although Mary and John use the same sentence, they express different propositions: Mary says that she is hungry, whereas John says that he is hungry. This difference is due to the character of I, which can be glossed as ‘referring to the speaker or writer’. Applying I’s character to a particular context determines the actual speaker in that context, i.e., the content (or intension) of I in that particular context. Having determined the proposition expressed by a sentence in a particular context, the proposition can then be evaluated with respect to a circumstance of use or possible world. Thus, the content (or intension) of a sentence in a particular context is a function from possible worlds to truth values. In Kaplan’s approach, then, context-sensitive expressions are such that their character applied to different contexts yields different contents. However, a contextsensitive expression’s content in turn is a constant function from possible worlds to extensions since regardless of the world at which the content of the expression is evaluated, it will always have the same extension. For example, the content of an expression such as I varies depending on the context in which it is used. However, once the content is determined, it stays the same for all possible worlds. In contrast, the content of hungry does not depend on the context in which it is used. It always is the property ‘being-hungry’. However, the actual extension of this predicate depends on the possible world that is assumed. That is, the set of individuals to which the predicate applies may differ across different worlds. Thus, contextinsensitive expressions have varying extensions, while their characters are such that regardless of the context the respective character is applied to, the same content will be determined. In a way, for context-insensitive expressions their character and content fall together. 4 Cf. Chierchia and McConnell-Ginet (2000), Braun (2010) for accessible introductions to these notions. Literal Meaning and Context-Independence 13 Kaplan’s approach, thus, allows a differentiation of three levels of meaning: character, content or intension and extension. For sentences this means one can differentiate between the context-independent sentence meaning, the proposition expressed by a sentence in a context and the truth value of a sentence in a context with respect to a possible world. Thus, implementing these ideas in a model-theoretic semantic apparatus leads to the truth of a sentence not only being determined with respect to a world and time, but also a context of utterance (cf. Sag 1981). In a way, within such an approach, both assumptions a) and b) can be maintained. That is, what is determined by the semantic component is the contextindependent meaning of a sentence and the conditions under which that sentence is true. Using the indices w, i and c, the instruction of how to determine the proposition expressed by a sentence is also given. However, it should be noted that the proposition expressed only actually is determined, once the functions are applied to a particular world, time and context. In other words, although it is possible within such an approach to formulate conditions under which a particular sentence is true, due to the indices used the sentence’s meaning thus given may be compatible with quite a number of different situations. Thus, it cannot be taken to represent the proposition expressed in a particular utterance situation. Traditionally, formal semanticists have assumed that the semantic component of the language faculty determines the meaning both of simple and complex expressions and then there are a restricted number of processes (namely, resolving of reference, fixing of indexicals and disambiguation) that lead to the proposition expressed by a sentence. However, these processes are not explicitly referred to as being of a pragmatic nature. This is quite obvious in the works of Grice, who mentions the processes that lead to what he called what is said, but does not seem to consider them as pragmatic in the same sense as the processes that result in conversational implicatures (Grice 1975) (cf. figure 2.2). However, if pragmatic processes are characterised by the fact that they take into account contextual information then, surely, the processes of fixing indexicals, resolving references and disambiguation are of pragmatic nature. Semantics lexical meaning semantic composition sentence meaning ? reference resolution fixing indexicals disambiguation what is said Pragmatics basis for further pragmatic inferences conversational implicature ... what is meant Figure 2.2: Grice’s distinction of what is said and what is meant 14 Against the Standard Notions of Literal Meaning and Non-literal Meaning Thus, as Strawson (1950) noted, it is not sentences which express something of which it makes sense to ask whether it is true or false but rather the utterances of those sentences. Thus, one and the same sentence can be used to express something true at one point and something false at another. That is, regardless of whether sentences include indexical or ambiguous expressions, it is not a general property of sentences, but rather of utterances that they express propositions. If it is not sentences per se that express propositions and are truth-evaluable but rather their utterances, what exactly, then, does the concept of sentence meaning capture? This is an important question considering that formal semantics takes sentence meaning as the starting point from which to deduce the meanings of simple expressions, which presupposes that the notion of sentence meaning is clearly defined. A possible answer is to still regard both the meaning of simple as well as complex expressions, in particular sentences, as essentially context-independent. That is, as traditionally assumed, semantics deals with the meaning of both simple and complex expressions, where the meaning of simple expressions forms part of their lexical entries and the meaning of complex expressions is a function of the meanings of their parts and their syntactic combination. However, such a view does not claim that sentence meaning necessarily is propositional; it simply assumes that sentence meaning is context-independent. 2.1.2 Literal Meaning as Context-Independent? But what about the correlation between sentence meaning and literal meaning suggested above? There it was stated that, apparently, literal meaning refers to a level of meaning identified as sentence meaning by traditional formal semantics and characterised as being context-independent and fully propositional. The assumption was that the notion of literal meaning mainly captures the fact that sentence meaning is context-independent, thus, with the revised characterisation of sentence meaning as ‘only’ context-independent but not necessarily fully propositional, the term literal meaning should still be applicable to that level of meaning. There are a number of considerations that go against this characterisation. Thus, recall the uses of the term literal meaning mentioned in chapter 1, where the term, on the one hand, is used to refer to a kind of context-dependent but at the same time in some sense ‘basic’ meaning and, on the other hand, is contrasted with a kind of meaning that is not only context-dependent but crucially in some sense ‘derived’ or non-basic (cf. Sag 1981, Recanati 1995, Carston 2007). As mentioned before, such a use calls into question the adequacy of characterising literal meaning as context-independent meaning. In fact, already in his (1978) paper, Searle criticised this characterisation of literal meaning. He argues that there is no such thing as a solely linguistically determined literal meaning of a complex expression. As regards sentence meaning, one cannot speak of the literal meaning of a sentence in the standard sense. As Recanati (2004) puts it, Searle holds the view of contextualism, according to which ‘. . . there is no level of meaning which is both (i) propositional (truth-evaluable) and (ii) minimalist, that is, unaffected by top-down factors.’ (Ibid., p. 90). Thus, Searle assumes that the expression of a determinate proposition takes place against a set of background assumptions. To illustrate his point of view, Searle Literal Meaning and Context-Independence 15 uses the sentence in (3), which, taken out of context, seems to have a quite obvious literal meaning, which Searle (1978) depicts as in 2.3. (3) The cat is on the mat. Figure 2.3: The typical cat-on-the-mat configuration The problem with this ‘literal sentence meaning’ is that although speakers or hearers are not necessarily aware of the fact, a number of preconditions are assumed to hold5 . To show this, Searle constructs a context of utterance for the sentence in (3), where it is questionable whether one would want to say that the sentence correctly describes the state of affairs at hand. ‘. . . suppose the cat and the mat are in exactly the relations depicted only they are floating freely in outer space, perhaps the Milky Way galaxy altogether. In such a situation the scene would be just as well depicted if we turn the paper on edge or upside down since there is no gravitational field relative to which one is above the other. Is the cat still on the mat?’ (Searle 1978, cited from Searle 1979, p. 122) Thus, if what the meaning of a sentence does is determine a set of truth-conditions, Searle argues that for most sentences this determination can only take place against specific background assumptions. These background assumptions are not part of the semantic structure of the sentence, that is, they are unarticulated. Moreover, due to possible variations in the background assumptions, the same sentence might have varying truth-conditions. For any sentence, there is no fixed set of background assumptions of which it could be said that it determines that sentence’s literal meaning. To illustrate this fact, Searle construes a context of utterance for (3), in which it could be used to truthfully describe a situation such as depicted in figure 2.46 . A further example for the fact that the literal meaning of a sentence depends on background assumptions can be found in Searle (1980). Searle gives a number of sentences containing the verb to cut; here are the first five. 5 Note that Searle is not referring to the fact that the sentence in (3) additionally contains indexical elements. That is another matter. 6 This is Searle’s context: ‘The mat is in its stiff angled position, as in [figure 2.4], and it is part of a row of objects similarly sticking up at odd angles - a board, a fence post, an iron rod, etc. These facts are known to both speaker and hearer. The cat jumps from one of these objects to another. It is pretty obvious what the correct answer to the question “Where is the cat?” should be when the cat is in the attitude depicted in [figure 2.4]: The cat is on the mat.’ (Searle 1978, cited from Searle 1979, p. 125). 16 Against the Standard Notions of Literal Meaning and Non-literal Meaning Figure 2.4: A rather unusual cat-on-the-mat configuration (4) a. b. c. d. e. Bill cut the grass. The barber cut Tom’s hair. Sally cut the cake. I just cut my skin. The tailor cut the cloth. As Searle notes, in each of the example sentences in (4) cut occurs in its literal meaning. There is nothing in these sentences as such that would lead one to interpret them as metaphorical or figurative. However, although cut occurs in its literal meaning, the situations that it is used to describe differ conceptually. Thus, although cut is used in its literal meaning, for the different sentences in (4), it determines different truth conditions. This can be seen if one considers what it would mean to obey an order of cutting something. Searle puts it as follows. If someone tells me to cut the grass and I rush out and stab it with a knife, or if I am ordered to cut the cake and I run over it with a lawnmower, in each case I will have failed to obey the order. That is not what the speaker meant by his literal and serious utterance of the sentence. (Searle 1980, p. 223). Thus, again, in the examples in (4), the literal meaning of the individual sentences (and of the word cut) is determined against a set of background assumptions, namely what we know about lawns and cakes and so on and what are usual actions in which we involve with regard to those ‘things’. Furthermore, in his discussion on the cut examples, Searle points out that it is not sufficient to assume that the different readings of cut – its different literal meanings are due to some intrasentential interaction between the verb and its internal argument. That is, he argues against the view according to which cut together with the respective argument determines that cut in ‘cut the grass’ will receive a different interpretation from the one it receives in ‘cut the cake’. His reasoning is that it is possible to ‘. . . imagine circumstances in which “cut” in “cut the grass” would have the same interpretation it has in “cut the cake”. . . ’ (Searle 1980, p. 224). ‘Suppose you and I run a sod farm where we sell strips of grass turf to people who want a lawn in a hurry. [. . . ] Suppose I say to you, “Cut half an acre of grass for this customer”; I might mean not that you should mow it, but that you should slice it into strips as you could cut a cake or a loaf of bread.’ (Searle 1980, p. 224–5). Literal Meaning and Context-Independence 17 Moreover, he points out that there is a difference to be drawn between what he calls background assumptions and the special context of utterance for a given utterance. While background assumptions are involved in determining a sentence’s literal meaning or truth-conditions, the context in which a sentence is uttered helps the hearer to decide on whether a speaker intended her utterance to be taken literally or non-literally. However, since Searle does not explicitly define what constitutes background assumptions and what is part of the context of an utterance, the question arises whether this differentiation really is necessary. From the examples Searle uses to defend his view of what constitutes literal meaning, it could be argued that the background assumptions necessary for determining the literal meaning of a sentence are in fact part of the specific context in which an utterance takes place. What Searle obviously means by background assumptions are certain aspects of knowledge that we have, namely those aspects which are relevant in the particular utterance situation. Thus, one could also assume that depending on the situation speakers and hearers find themselves in that situation will make certain aspects of knowledge they have more prominent (or salient). Those aspects, then, constitute what Searle calls background assumptions in the sense that speakers and hearers are presumably normally not aware of basing their utterances and interpretations on such assumptions. As their name implies, background assumptions are in the background; they form the basis from which speakers formulate their utterances and hearers intepret them. Thus, background assumptions depend on the particular context of utterance and therefore can be said to form part of the contextual information used in interpreting. Thus, in order to disambiguate whether the expression cut is used with the meaning as in ‘cut the grass’ or with the meaning as in ‘cut the cake’, the hearer needs to take into account contextual information. That is, even if the background of the utterance is such as Searle gives it, the hearer would still have to decide that the reading of cut as in ‘cut the cake’ is the one the speaker intended in that situation. From what has been said about salience of meaning above, it is of course very likely that the particular utterance situation will speed up the hearer’s unconscious decision. Furthermore, Searle’s argument that there are possible circumstances in which cut in ‘cut the grass’ may be interpreted as in ‘cut the cake’ actually does not constitute an argument against the assumption that the interpretation of cut is influenced by the intrasentential context. That is, one could assume that the co-occurrence of particular lexical items does help the hearer to narrow down the possible sentence meaning. However, this influence on the interpretation might have a default character. Thus, it only applies where the particular contextual conditions do not prevent it from applying. In the context Searle supplies, the interpretation of cut as in ‘cut the grass’ is rendered less likely as being the intended reading, than the reading as in ‘cut the cake’. As Searle argues, this reading of cut in the given context does not seem to be a non-literal reading of cut, since, intuitively at least, it does not seem to be derived from some clear basic, underlying meaning. However, assuming that the reading only comes about, or is interpretable as intended, in a particular context of utterance, suggests once again that literal meaning should not be taken to be a phenomenon of context-free sentence meaning. If that is the case, then the concept of literal meaning is not applicable at Searle’s level of sentence meaning, but 18 Against the Standard Notions of Literal Meaning and Non-literal Meaning rather at some context-dependent level of meaning. Be that as it may, Searle still assumes that literal meaning is the basis for any non-literal meaning. It is speakers who may use some expression or other non-literally. Thus, non-literal meanings have to be intended and should be expected to be consciously recognisable as such. That is, speakers should have no difficulty identifying some reading as being nonliteral, as they have to intentionally use some expression ‘deviantly’ in order for that expression to get interpreted non-literally. Although Searle thus argues against the view according to which literal meaning is determined by the linguistic system alone, he does not want to deny that sentences in fact do have literal meanings. ‘Literal meaning, though relative, is still literal meaning.’ (Searle 1978, cited from Searle 1979, p. 132). However, he applies the concept of literal meaning to those cases in which the speaker means what she says, contrasting them with those cases in which the speaker means more, or something different from what she said (e.g., cases of irony, conversational implicature or indirect speech acts). Thus, although Searle identified literal meaning as belonging to the level of sentence meaning, actually the differentiation between literal and nonliteral utterances seems to hold at the level of utterances. However, he also argues that literal meaning is a relative notion. That is, it is rather likely that what we take to be the literal meaning of the utterance of some sentence will depend heavily on the specific circumstances in which the utterance of that sentence takes place. This is in agreement with uses of the term literal meaning where it refers to an utterance’s meaning, in contradistinction to the meaning intended by the speaker of that utterance. Similarly to Searle, Bierwisch (1979, 1983) assumes that what is called the literal meaning of an utterance of some (simple or complex) expression is not identical to the linguistically determined meaning of that expression. Thus, Bierwisch also places literal meaning at a level of meaning that is no longer independent of context, namely the level of utterance meaning. Utterance meaning is the meaning an utterance token of an expression has when it is used in a context. The utterance meaning can be equivalent to the utterance token’s literal meaning, but it does not necessarily have to be. Therefore, Bierwisch differentiates the literal meaning of an expression from its utterance meaning. Crucially, an utterance token of an expression can only have literal (or, for that matter, non-literal) meaning in a context. With respect to this assumption, Bierwisch and Searle hold similar views. That is, what is called the literal meaning of an expression is not determined language internally, rather, it is dependent on certain background assumptions (Searle) or a particular context of utterance (Bierwisch). Thus, literal meaning is a special case of utterance meaning. A consequence of such a view is the assumption that the lexical semantic representations of simple expressions do not encode what we take to be their literal meaning. What has been said sofar, corroborates a suspicion expressed at the beginning of this section. That is, one has to ask whether the particular standard characterisation of literal meaning might not be very much influenced by our characterisation of the field of semantics. Thus, consider again that, traditionally, semantics takes as its starting point sentence meaning of which it assumes that it is both contextindependent as well as propositional. Because it is taken to be context-independent, Literal Meaning and Context-Independence 19 it seems to be what the sentence literally expresses. However, we saw that, for a large number of sentences, it cannot be said that they express propositions, unless contextual information is first taken into account (e.g., cases of reference resolution, fixing of indexicals and disambiguation). That is, sentence meaning and ‘propositional content of an utterance’ are not equivalent. Furthermore, as we saw from Searle’s remarks, actually, what we take to be the literal meaning of a complex expression, is dependent on the context in which that expression is uttered. That is, essentially, literal meaning seems to be context-dependent after all. Thus, it seems reasonable to posit a partial new characterisation of the terms sentence meaning and literal meaning (with respect to sentences). Whereas the former is the meaning of a certain type of complex expression and characterised by the facts that it is compositional, context-independent and (more often than not) sub-propositional, the latter is a certain type of meaning a sentence may have when used in a context that allows a literal interpretation and in which that sentence expresses a full proposition. What this characterisation of literal meaning suggests is that whatever differentiates between literal meaning and non-literal meaning cannot be the criterion of context-(in)dependence. Moreover, this characterisation of literal meaning (with respect to sentences) makes it equivalent to a particular type of proposition. Thus, a new question arises, namely, how this particular type of proposition is characterised. That is, which conditions does a proposition have to fulfil for it to be literal in meaning? This is a question which is very close to the core of the discussion around the semantics/pragmatics distinction and we will come back to it in the following chapters. meaning context-independent context-dependent lexical m. sentence m. ... underspecified literal ... ... non-literal Figure 2.5: Revised differentiation of types of meaning 2.1.3 Literal Meaning as Primary to Non-literal Meaning? The traditional view of the interpretation process assumes that an utterance’s literal meaning is always activated first. Potential non-literal meanings only get activated as a result of the literal meaning’s not fitting in the respective context. This view, of course, is based on the traditional assumption about literal meaning being context-independent. Since this allows an interpretation of the literal meaning of an 20 Against the Standard Notions of Literal Meaning and Non-literal Meaning utterance without taking recourse to contextual information, it will be computed automatically. If afterwards it becomes apparent that the literal meaning does not fit the contextual circumstances, a reinterpretation will take place, resulting in a non-literal interpretation of the utterance (call this the standard pragmatic view). This is essentially how Grice must have viewed the relation of literal meaning and non-literal meaning, since he described different kinds of non-literal meaning, such as irony and metapor, as being conversational implicatures, that is, inferences that require a prior recovery of what is said by an utterance. Of this latter level of meaning, Grice said that it is very closely connected to the conventional meaning of the words or the sentence uttered by the speaker. Specifying on this characterisation, what is said has been taken to be the fully propositional semantic form of the utterance resulting from the processes of disambiguation, reference resolution and fixing of indexicals, where these processes already are of pragmatic nature since they involve contextual information (see figure 2.6).7 Semantics lexical meaning semantic composition sentence meaning Pragmatics does not involve any non-literal meaning reference resolution fixing indexicals disambiguation what is said basis for further pragmatic inferences conversational implicature speech acts ... may involve non-literal meaning what is meant Figure 2.6: (Non-)Literal meaning and Grice’s levels of what is said/what is meant Thus, what is said, being a full proposition, provides the basis for further inferences about what the speaker actually meant with his utterance. That is, conversational implicatures are derived from the fact that the speaker said what he said in the particular way he did and with respect to a number of conversational maxims, which are taken to underlie human communication. Since what is said includes the conventional meanings of expressions, it is traditionally assumed to be the level at which literal meaning is expressed. And since Grice, similarly to Searle, viewed nonliteral meaning such as irony or metapher, as an aspect of speaker meaning, what is meant is the level of meaning at which such non-literal meaning aspects come into play. Thus, since for the recovery of what the speaker meant we first have to know what the speaker said, interpretation of the literal meaning of an utterance is prior to the interpretation of a potential non-literal meaning. Moreover, since non-literal 7 Note that this fact is independent of the assumption that what is said constitutes the level of meaning from which conversational implicatures are determined. That is, even if what is said is not viewed as wholly semantic, the standard pragmatic view could still hold in that what is said still forms the basis for drawing further inferences and thus has to be determined first. Literal Meaning and Context-Independence 21 meaning aspects such as irony or metaphor rest on the violation of a conversational maxim, they will only get derived if the literal interpretation, that is, what is said, cannot possibly be construed as the meaning intended by the speaker. Interestingly, although arguing against viewing literal meaning as solely linguistically determined, Searle shares the view according to which literal meaning is prior to non-literal meaning. Thus, in his (1979) paper, Searle says of metaphorical and ironical utterances, that their respective interpretations are arrived at by going through the literal meaning of the sentences used to make the utterances. Thus, he seems to assume that although a sentence’s literal meaning can only be determined against particular background assumptions, in terms of the temporal progression of the interpretation process, literal meaning is a necessary intermediate step in the interpretation of non-literal meaning. Fortunately, with the methods developed in psycholinguistics, the assumption of the primacy of literal meaning as an assumption about the operational sequence of the interpretation process has become empirically testable. And in fact, results of experiments employing different methods in examining the understanding of various types of non-literal meaning in comparison to literal meaning show that the standard pragmatic view makes the wrong predictions. That is, in terms of cognitive effort, the standard pragmatic view predicts that interpreting non-literal meaning should be cognitively more exacting than the interpretation of literal meaning. Given that reaction or reading times mirror the relative cognitive effort involved in interpreting an utterance, results such as the following suggest that the interpretation of non-literal meaning does not necessarily differ from that of literal meaning. Thus, Gibbs (1994) mentions an experiment (Ortony et al. 1978), where subjects were presented with sentences in a context that was either literal or metaphoric. The hypothesis that was tested in the experiment was that people may not have to analyse the literal interpretation of a metaphorical utterance before actually deriving the intended metaphorical reading. The hypothesis was confirmed. Thus, although subjects took longer to read metaphorical targets than literal ones in short contexts, in long contexts, there was no difference in reading times for the metaphorical and the literal target sentences. These results suggest that the richness of contextual information available during the interpretation of an utterance has an effect on how difficult it will be to give that utterance a non-literal interpretation. Another experiment showed that utterances may be interpreted non-literally although there are no conditions that trigger the failure of a literal interpretation, suggesting that people automatically apprehend the metaphorical meaning of an utterance (Glucksberg et al. 1982). The task was to judge sentences such as Some jobs are jails as to their literal truth. Thus, it was not necessary to seek a non-literal interpretation for the sentences, subjects only were asked for the literal truth of each sentence. Now, if for a non-literal interpretation of a sentence a pragmatic triggering condition is required, sentences such as Some jobs are jails should simply be considered as false. If, however, people automatically interpret the metaphorical meaning of such sentences, then the ‘false’ judgement for the literal reading of the sentence should be in conflict with the ‘true’ judgement for the non-literal reading of the sentence. And in fact, although subjects correctly judged sentences such as Some jobs are jails as literally false, if a metaphorical interpretation for the sentence in question was 22 Against the Standard Notions of Literal Meaning and Non-literal Meaning available, subjects took much longer to make that judgement. Thus, apparently, the metaphorical meanings were automatically interpreted, without the need for some pragmatic triggering condition (i.e. maxim violation). This suggests that the interpretation of non-literal meaning does not rely on the violation of some conversational maxims or principles. Moreover, it shows that, although context may facilitate the interpretation of an utterance as non-literal, it is not absolutely necessary. Note, however, that more can be said with respect to such examples as Some jobs are jails. Thus, although it is true that there does not seem to be a pragmatic triggering condition such as maxim-violation for a metaphoric interpretation, nevertheless, it can be argued that in cases such as this, there is some triggering condition, either semantically or pragmatically induced. To repeat, the standard pragmatic view assumes the literal meaning of an utterance is interpreted first and only if this does not fit the contextual circumstances a non-literal interpretation is determined. The problem with sentences such as Some jobs are jails or, for that matter, The ham sandwich is sitting at table 7 is that it is not clear what their literal meaning should be (cp. Stern 2006). Thus, the longer reaction times measured in the experiment mentioned above might be due to the fact that whatever component is responsible for this stage in the interpretation process is having problems determining a literal interpretation for the sentences in question. This, in turn may be sufficient to trigger an alternative, non-literal interpretation (cp. Dascal 1987). Keysar (1989) takes up this criticism and shows that even in contexts were a particular sentence is understood as literally true, that sentence’s potential but false metaphorical meaning interfers nevertheless. Thus, subjects were asked to judge test sentences as true or strongly implied to be true after having read small texts. The texts consisted of two parts, were one part related to the literal interpretation (L) of the target sentence, rendering it either true (L+) or false (L-), and one part related to the metaphorical interpretation (M), again rendering this either true (M+) or false (M-). Thus, texts were (L+M+), (L-M+), (L+M-) or (L-M-). For example, one of the test sentences was Bob Jones is a magician. An example text for which that sentence is interpreted as literally true but metaphorically false (L+M-) is given below. Bob Jones is an expert at such stunts as sawing a woman in half and pulling rabbits out of hats. He earns his living travelling around the world with an expensive entourage of equipment and assistants. Although Bob tries to budget carefully, it seems to him that money just disappears into thin air. With such huge audiences, why doesn’t he ever break even? (Keysar 1989, p. 378) The results show that subjects are quickest in responding after texts that rendered the target sentence both literally and metaphorically true (L+M+). Generally, they are quicker in responding in congruent contexts (i.e. L+M+ and L-M-) than in incongruent ones. Thus, as in (Glucksberg et al. 1982)’s experiment, subjects take longer judging literally false but metaphorically true sentences. Crucially, however, they also take longer judging literally true but metaphorically false sentences, indicating that even in such a situation the potential metaphorical interpretation of the sentence is computed. This result shows that the longer reaction time measured by Non-literal Meaning and Conventionality 23 (Glucksberg et al. 1982) for literally false but metaphorically true sentences is not due to the difficulty of determining a literal interpretation for those sentences in the first place, as similar results are achieved in situations in which the target sentence actually is literally true but metaphorically false. Summing this section up, the conclusion one can draw from examining different empirical studies is that it is not generally necessary to compute the complete literal meaning of an utterance before deriving that utterance’s intended non-literal meaning (cf. Gibbs 2002). Thus, it seems that the possible but non-intended literal interpretation of an utterance does not constitute a necessary intermediate step during the processing of that utterance’s intended non-literal interpretation. Moreover, contextual information seems to be an important factor in the non-literal interpretation of utterances. Thus, if the context in which an utterance takes place points toward a literal interpretation, the utterance will be interpreted literally. If, however, the context of an utterance points towards a non-literal interpretation, the utterance will be interpreted non-literally. However, it seems that context is not always necessary for the non-literal interpretation of an utterance. Thus, Glucksberg et al. (1982) found that sentences such as Some jobs are jails are interpreted nonliterally, even if there is no context given with which the literal meaning could be said to be conflicting. The criticism that in such cases what may underlie the longer time needed for judgements concerning the literal truth/falsity of such sentences is not a maxim-violation problem, but rather the problem of determining a literal meaning at all was refuted by Keysar (1989). He found that even in contexts were a particular sentence gets a straightforward literal interpretation, a potential but non-fitting metaphorical interpretation interfers leading to longer reaction times in subjects. Thus, non-literal interpretations seem to get considered without the need of a triggering condition. With such results as the above, it becomes questionable whether it is warranted to assume a difference in conventionality for literal meaning and non-literal meaning. 2.2 Non-literal Meaning and Conventionality Whereas, traditionally, literal meaning was taken to be semantic, that is, contextindependent, conventional and primary, non-literal meaning was characterised as pragmatic, that is, context-dependent, non-conventional, a feature of the intended speaker meaning arrived at by a reinterpretation process and thus secondary to literal meaning. Having in the last section mainly argued against the traditional characterisation of literal meaning as context-independent and thus primary, in this section, I will argue that non-literal meaning should not be viewed as necessarily nonconventional and thus secondary. Generally, I want to argue that conventionality should not be viewed as an ‘all-or-nothing’ property, but rather as being gradual. 2.2.1 Empirical Evidence Since non-literal meaning usually is considered a feature of language use, it is not surprising that the term non-conventional often features in its characterisation, where its opposite conventional is mostly used in the sense of ‘stored in the lexicon’. How- 24 Against the Standard Notions of Literal Meaning and Non-literal Meaning ever, it is not necessarily so that a particular instance of non-literal meaning may not be considered as conventional. Rather, given that conventionality is reflected by familiarity and relative ease of interpretation, cases of conventionalised non-literal meanings of expressions are quite numerous.8 As a consequence, context may not actually play such an important role when it comes to the interpretation of an expression as literal or non-literal. Moreover, and as already suggested in the last section, the claim that non-literal meaning is always the result of a reinterpretation of some earlier composed literal meaning can no longer be sustained. Thus, consider the claims Rachel Giora (1997, 1999, 2002) makes in the context of her graded salience hypothesis. Although she does not question the general importance of contextual information for utterance interpretation, she argues that regardless of the richness of the context, it can be shown that in particular cases both the literal as well as the non-literal meaning of an expression may be initially activated, suggesting that in such cases, the two types of meaning have the same status. More specifically she argues that it is the notions of salient vs. non-salient meaning that play a decisive role for the process of interpreting utterances. The crucial idea is that salient meanings are always activated, whether they are actually intended or not. Thus, in her (1999) paper, Giora mentions a number of experiments which, again, show that the non-literal meaning of an expression is not necessarily activated after its literal meaning has been. For example, Giora and Fein (1999) showed that reading times for familiar metaphors9 were equal for literally as well as metaphorically biasing contexts. That is, for familiar metaphors, both their possible literal and non-literal meanings are activated during interpretation, regardless of whether they are uttered in literally or metaphorically biasing contexts. This is because both meanings are familiar (salient) as for instance for to break ones head. (5) a. In order to solve the math problem, the student broke her head. b. Because she was so careless when she jumped into the pool, the student broke her head. Thus, the experiments show that in cases of familiar metaphors literal meaning and non-literal meaning of an expression are computed in parallel. In contrast, for less familiar metaphors only their literal meaning is salient, thus, it should be activated prior to the less familiar metaphoric meaning. This hypothesis was indeed confirmed by longer reading times found for less familiar metaphors in metaphorically biasing contexts as in contrast to literally biasing contexts. Giora (1999) cites the following sample contexts and target sentences for a less or unfamiliar metaphor.10 8 Having said this, one might wonder whether under such circumstances it is actually still possible and useful to differentiate literal from non-literal meaning. It is if one wants to capture the fact that non-literal meaning, even if conventionalised, is perceived as in some sense non-basic or derived. 9 It should be noted that this is a somewhat sloppy use of the term metaphor. What is meant is that expressions where investigated that could either be interpreted metaphorically or literally and for which the metaphorical interpretation is rather common. 10 It should be mentioned that the original experiment was carried out in Hebrew. Unfortunately, neither Giora (1999) nor Giora and Fein (1999) give examples for any of the original material. Thus, the critical points I will make below concerning the material and the differentiation of idiom from metaphor could just be an artefact of the English translation. However, as an Hebrew informant Non-literal Meaning and Conventionality 25 (6) Mary: My husband is terribly annoyed by his new boss. Every day he comes home even more depressed than he had been the day before. Somehow, he cannot adjust himself to the new situation. Jane: Their bone density is not like ours. (7) Our granny had a fracture just from falling off a chair and was rushed to the hospital. I told my sister that I never had fractions falling off a chair. She explained to me about elders. She said: Their bone density is not like ours. Note, however, that at least for the example cited, the longer reading time may also at least partly result from a further difficulty encountered during the interpretation process. Thus, whereas, arguably, in (5a) and (5b) the binding of the anaphoric expression her is equally difficult, the case is different for their in (6) and (7). Whereas in (7) their can be bound to the overt antecedent elders in the preceding sentence, there is no overt antecedent for their in the preceding sentence in (6). Thus, the longer reading time may reflect the greater difficulty in binding the anaphoric expression their to an appropriate, not explicitly expressed antecedent. A similar example of metaphoric material taken from Giora and Fein (1999) suggests that this is a general problem of the testing material used to investigate differences in the interpretation of metaphoric vs. literal interpretations. (8) The Saturday night party went on for hours. Drinks were poured, and we danced all night. We were probably less than considerate when, the next evening, we called on our friends who had been partying with us. When they opened the door, we realised: Only then did they wake up. (9) A bloody war has been going on in central Europe for a few years. Thousands of innocent women, men and children got massacred, and no one budged or lifted a finger. At last, a decision was made to intervene in the fights. Only then did they wake up As in the example before, in (8) the anaphoric expression they in the last sentence can be bound to an appropriate antecedent in the preceding sentence, namely our friends. This, however is not the case for they in (9), where there, once again, does not seem to be an appropriate, overt antecedent given. An attendant, general problem is the fact that the context preceding the target sentence may also prime the meaning intended in the target sentence. Thus, the fact that in (7) in the sentences preceding the target sentence, such expression as fracture and fraction are used, facilitates a literal interpretation of bone density in the target sentence. There is no such priming effect for the metaphoric interpretation of bone density in (6). The results of another experiment employing a word fragment completion task (Giora and Fein 1999) also suggests that the interpretation process varies with the familiarity of the metaphor processed. In this experiment, subjects were instructed to read little texts, the last sentence of which was the target sentence, involving a metaphor. Depending on the nature of the preceding context, subjects were intented to be biased either towards a literal interpretation or a metaphorical interpretation of the target sentence. Their task then was to complete two fragmented words, one of tells me, the criticism also holds when considering the Hebrew equivalents. 26 Against the Standard Notions of Literal Meaning and Non-literal Meaning which was related to the literal (or rather compositional) interpretation of the target sentence, and one which related to the metaphorical interpretation. The results showed that in the case of familiar metaphors - regardless of the bias induced by the preceding context - both the compositional as well as the metaphorical meaning were activated. In contrast, for less familiar metaphors, there was near to none activation of the metaphorical meaning in the literally biasing context, and more activation of the literal meaning than the metaphoric meaning in the metaphorically biasing context. Giora and Fein (1999) also tested the activation of meaning during the interpretation of familiar and less or unfamiliar idioms. The assumption was that for familiar idioms, their idiomatic meaning is more salient than the possible literal meaning. Thus, the idiomatic meaning should get activated both in idiomatically biasing as well as literally biasing contexts, whereas the literal meaning should only get activated in literally biasing contexts. As before, the task was to complete fragmented words with the first word that came to mind after reading an either literally or idiomatically biasing context with the last sentence being the target (see example below). This time, however, subjects were asked to complete only one of the two given fragmentations. Again, the results of the experiment confirmed this hypothesis. (10) a. He told me the whole story on one leg. b. In the zoo, I saw a stork standing on one leg. Thus, what the results of the different experiments mentioned in this and the preceding section show is that literal meaning is not necessarily processed first in interpretation. Moreover, once a non-literal meaning of an expression has grown familiar, it can be activated in parallel to the literal meaning of that expression, suggesting that non-literal meaning should not be seen as a feature of the level of speaker meaning, exclusively. Thus, the results from psycholinguistic experiments reinforce the need to rethink the view of the relation between literal meaning and non-literal meaning. That is, although historically and with respect to new or ad hoc instances of figurative meaning, the latter is derived from literal meaning by certain processes, such as metaphorical use, synchronically speaking this fact may no longer play a role. More specifically, in terms of the operational sequence of the interpretation process, the fact that the non-literal meaning of an expression may be activated in parallel to its literal meaning rather suggests that they are more similar in nature than traditionally assumed. However, the results of experiments such as the above cannot be used as evidence for one view or the other of literal meaning and non-literal meaning without caution, as there are a number of issues that need careful consideration. Generally, a lot of psycholinguistic studies are interested in phenomena such as metaphor, idioms and irony, none of which are uncontroversial. That is, there is a lot of debate concerning the appropriate characterisation of these phenomena. Particularly with respect to metaphor and idioms, it does not seem to be trivial differentiating the one from the other, especially since one way for idioms to get the particular meaning they have is through a metaphorical use of the phrase in question. Thus, looking again at the examples for a metaphor and an idiom from Giora given above, I would content that Non-literal Meaning and Conventionality 27 it is not intuitively clear by which criteria one is a metaphor and the other an idiom. (11) to break ones head (12) on one leg Moreover, if metaphor is taken to differ from idiom by the fact that the metaphoric meaning is related to a single simple expression rather than to a phrase, the question arises which expression in to break ones head carries this metaphoric meaning.11 However, judging from the examples Giora gives for metaphors, she does not seem to assume that metaphoric meanings are related to a single expression, which makes the question, what, then in her view differentiates between idiom and metaphor even more pressing. Furthermore, Giora’s use of the terms literal meaning and non-literal meaning with respect to idioms is rather confusing. It is generally assumed that idioms are coded in the lexicon as a whole form associated with the respective (more or less) idiomatic (or in Giora’s sense non-literal) meaning. This coded meaning is usually taken to be the literal meaning of idioms, simply because the idiom is not taken to be a decomposable phrase, but as a whole. Idioms are usually not conceived of as having literal meaning in Giora’s sense. It is the phrasal, compositional counterparts of idioms that can be said to have literal meaning. The problem is, that it is not totally clear how selection or activation in cases where idioms are involved works. That is, when a sentence such as He told me the whole story on one leg is read or heard, the question is at which point the idiomatic meaning is activated. That is, how many segments must the addressee have heard in order to activate the idiomatic meaning? It seems rather plausible then, that an addressee will initially assume that he is dealing with a compositional phrase and then revise this assumption as soon as it becomes clear that the phrase in question is one that is coded in the lexicon as such. That is, from the phonological surface form only it is not obvious whether a specific phrase is used with a compositional meaning or with an idiomatic meaning. In this sense idioms and their compositional counterparts are similar to homonyms. They have the same surface graphical or phonological form, but differ in their meaning and, additionally, in their complexity. Given this assumption, of course, one would expect that both meanings should be equally activated, which, according to Giora and Fein (1999), they are not. The results in that study show that for familiar idioms, in an idiomatically biasing context, the compositional meaning is only marginally activated. However, since it does not seem clear on which basis Giora differentiates idioms from metaphors, the results cannot really be compared, since some of the example expressions that were classified as metaphors may quite as well be classified as idioms. Moreover, one cannot rule out that the results for (familiar) metaphors and idioms in the two experiments only differ due to the difference in testing procedure. Thus, recall that in the experiment on metaphors, subjects were asked to complete both of the two given fragmented words, whereas in the experiment on idioms, subjects only had to complete one of the two given words. The fact that for familiar idioms, the number of activated 11 One would probably be inclined to say that break carries the metaphoric meaning, however, then the question arises of whether this reading of break can be described without referring to the expression head. 28 Against the Standard Notions of Literal Meaning and Non-literal Meaning idiomatic meanings is so much higher than the number of activated literal meanings probably is due to the fact that the idiomatic meaning is the most salient, however, for all we know the possible literal meanings might be activated to some degree as well, but this simply isn’t captured by the design. In other words, had subjects been asked to complete both given words, the difference between the extent of activation of the idiomatic and literal meanings might have been smaller and thus more similar to the difference recorded for familiar metaphors. In addition, since the experiment involved a task that measures the activation of concepts after the actual interpretation process has already taken place, the results do not tell us anything about early, initial processes. In other words, for all we know, an idiom’s literal meaning (i.e., the compositional meaning of the idiom’s compositional counterpart) may still be activated first during the interpretation process, but may be suppressed as soon as it is clear that it does not fit the context. The degree to which such suppression takes place may have a significant impact on whether the initially activated literal meaning is still active enough to get accessed by the time the task is being solved.12 Generally, Giora (and Feins)’s results do not tell us what exactly is going on during the interpretation process.13 14 Thus, Giora and Fein (1999) used a word fragment completion task, which, it can be argued, is only sensitive to the state of activation at a time when the interpretation process has already finished. Although reading times may be better reflective of difficulties encountered during the interpretation process, it still is not clear what, or which steps, exactly are reflected by that longer reading time. Vice versa, similar reading times do not tell us anything either of what exactly is going on during the interpretation process and whether what we assume to be literal meaning and non-literal meaning are processed similarly. However, although the results of such experiments do not clearly contribute to disentangling the notions of literal meaning and non-literal meaning, at least they do not provide evidence that the traditional view of the operational sequence of the interpretation process is accurate. On a more positive note, ignoring the problematic aspects of such experiments for the moment, it nevertheless seems significant that all these studies point towards the same conclusion, namely that literal meaning and non-literal meaning are not so different in kind as traditionally assumed. In fact, Mashal et al. (2008)’s study provides evidence for differential processing of salient vs. non-salient meanings, rather than literal vs. non-literal meanings. Although this study, using fMRI, cannot make any claims concerning the time-course of activation of salient vs. non-salient meanings during interpretation, it does show a difference in the brain areas involved in interpretation related to the salience of the interpreted meaning. Thus, Mashal et al. (2008) investigated the interpretation of highly familiar idioms as compared to their possible compositional interpretation and the interpretation of matched literal sentences. Recall that according to 12 Cp. Cacciari and Tabossi (1988). Similarly, Dascal (1989) questions the appropriateness of assuming that reaction times reflect what happens early on during interpretation. 14 Although Laurent et al. (2006) used ERPs in their testing of the graded salience hypothesis concerning the activation of meaning for idioms, they only recorded EEGs during presentation of the last word of the respective idioms and the target words following that. Thus, their results do not reflect initial processes either. 13 Non-literal Meaning and Conventionality 29 the graded salience hypothesis the (idiomatic) meaning of highly familiar idioms is more salient than the (‘literal’) meaning of their compositional counterparts. The results show that while for the interpretation of the idiomatic meaning of idioms and the interpretation of literal sentences – both assumed to be salient – there is no involvement of the right hemisphere (RH), interpreting the literal and less-salient compositional counterparts of highly salient idioms does involve the RH. Thus, in this study, involvement of the RH does not correlate with literal/non-literal meaning of the sentence interpreted, but rather with the degree of salience of the meaning interpreted. 2.2.2 Theoretical Considerations There are also some theoretical considerations supporting the view that literal meaning and non-literal meaning should not be seen as concepts definable in terms of dichotomies. In particular, literal meaning and non-literal meaning should not be differentiated on the grounds that the former is conventional meaning as opposed to the latter, which is non-conventional meaning. Generally, one has to ask, what property exactly the concept of conventionality is supposed to pick out. That is, is an expression’s conventional meaning that which is coded in the lexicon, or is conventional meaning the reading with which a certain simple or complex expression is frequently used, regardless of whether that meaning is pragmatically derived or coded in the lexicon? In this context, the studies mentioned above, which showed that there are cases in which a non-literal meaning cannot be supressed, are interesting. Thus, the fact that for highly familiar metaphors their metaphoric meaning is activated even in contexts, where a literal interpretation is intended, lends credit to the assumption that such metaphoric meaning is as conventional as literal meaning. In addition to such empirical evidence, there are tendencies in theoretical linguistics, in particular with respect to meaning change, to view conventionality as a gradual concept instead of as a binary notion.15 This actually seems to be necessary, if one wants to explain how meaning change can take place at all, since it does not take place abruptly, but gradually. Thus, following Morgan (1978), Busse (1991) differentiates between conventions of language and conventions of language use, which he defines from a functional-pragmatic perspective.16 Thus, conventions of language are manifest in conventional language use, that is, a use of linguistic items that is generally and consistently conceived of as an intuitively and thoroughly ‘literal use’ by the members of a language community (Busse 1991, p. 55). In contrast, conventions of language use are defined as a costumary and established procedure of use of a particular linguistic expression for a consistent communicative purpose, where the use of that expression presupposes a still recognisable literal meaning as a necessary means of arriving at that purpose (Ibid.). Busse (1991) argues that different degrees of conventionality can be identified on the basis of a threefold distinction of levels 15 But see also Nunberg (1978, esp. ch. 4) for a characterisation of conventionality that is gradient. Note that Busse also argues against an absolute differentiation between literal meaning and non-literal meaning, which conceals the fact that in most concrete cases, the dividing line cannot be drawn according to general, objective features. However, he also acknowledges that the differentiation is one that is made by normal speakers on an everyday basis. That is, it corresponds to some intuition about language and thus should not be ignored by theory. 16 30 Against the Standard Notions of Literal Meaning and Non-literal Meaning of communicative use of language. At the first level, language is used ‘literally’, in that the members of the language community cannot detect any sort of ‘underlying meaning’ (i.e. a use of the conventions of language). The second level is the so-called ‘indirect’ or ‘deviating’ use, where that use might also already be conventionalised again in some sense (e.g. idioms or routine formulas), however, where most speakers may still recognise a ‘literal’ use of the expression in question. Thus, Busse acknowledges the fact that a particular ‘indirect’ meaning may also be conventionalised to some extent by differentiating between conventions of language and conventions of language use. On the third level of use, speakers use items from the first and second level or both, however, this use constitutes a communicative sense that is only comprehendable within a particular situation or context of utterance. Referring to the levels identified above, one can differentiate degrees of conventionality, according to whether speakers can still differentiate literal or ‘underlying’ meanings from a given communicative sense. conventions of language unconventional language use conventionalised conventions of language use literal meaning fully more idioms less non-literal meaning not particularised CIs Figure 2.7: Busse’s view of degrees of conventionality Again following Morgan (1978), Busse (1991) recognises three elements of the conventions of language use: situation (here: context), purpose and means. Thus, the clearer the connection between communicative purpose and linguistic means to the communication partners, that is, the more obvious it is to the participants that the literal meaning differs from the intended meaning, the further removed from a convention of language a particular use is. It then depends on the epistemic status of the purpose-means-relation whether what we are dealing with is a contextdependent pragmatic inference or a form of language use that is already consolidated to a convention (of use). At the point where the relation between purpose and means becomes totally obscure, it may be reinterpreted as completely arbitrary, that is, at this point the convention of language use is reinterpreted as a convention of language (Busse 1991, p. 57). Thus, Busse (1991) shows that Grice’s theory of conversational implicature together with his theory of meaning can be utilised for the explanation of how meaning change proceeds. A conversational implicature is the result of a chain of inferences, which goes from the literal meaning, context, Non-literal Meaning and Conventionality 31 purpose and further assumptions to the actually intended meaning. The shorter this chain of inferences becomes, that is, the fewer necessary ‘links’ it has, the more conventionalised the expression in question is. This may happen, e.g., if certain inferences are no longer necessary, because a particular use has already turned into an established procedure of use (Busse 1991, p. 58). Busse (1991) argues that meaning change starts at the point where speakers are no longer able to reconstruct all links within a chain of inferences. The becoming increasingly unnecessary of links within an inference chain then corresponds to increasing degrees of conventionality of the expression under consideration (cf. König 1988, Rolf 1995). However, as long as there is some transparency of purpose and linguistic means, we are still in the realm of what Busse calls convention of language use. Only at the point where the relation between purpose and means becomes absolutely ‘dark’, does the convention underlying the use of the expression in question become a convention of language. Thus, on Busse’s view conventionality and non-conventionality still go along with the differentiation of literal and non-literal meaning, however, he acknowledges the fact that there is a ‘grey area’, where meanings are to some extent conventionalised (as conventions of language use), but are not yet perceived of as literal meanings (thus, are not conventions of language). Thus, whereas Busse would still characterise literal meaning as fully conventional (being part of the conventions of language), nonliteral meaning can be characterised as either being non-conventional or as being conventionalised to a lesser or greater extent. These differences in conventionality can be captured by assuming that the process of conventionalisation is a gradual one, leading to different grades of conventionality at different stages of that process. Looking, once again, at idioms shows that this characterisation still is problematic. As mentioned above, idioms are usually characterised as being coded in the lexicon as a whole, since their meaning is not (totally – cf. Nunberg et al. (1994)) compositionally determined. Thus, idioms may have compositional counterparts with meanings built up regularly from the meanings of their parts. Crucially, there are cases of idioms, where (at least synchronically) there is no transparent connection between the compositionally determined meaning and the meaning of the idiom, thus, making idioms and their compositional counterparts rather similar to homonyms (e.g. kick the bucket). As we saw, such idioms may still be characterised as having non-literal meaning17 (albeit their meaning is considered conventional), although this seems a rather irritating use of the term non-literal meaning. However, the reason for calling the meaning of an idiom non-literal may simply be that speakers are, in principle, able to recognise that these expressions have counterparts that have compositional and thus ‘regular’ meanings. In that case, the term nonliteral is not only used for a meaning of some expression that is somehow ‘derived’ or based on some ‘underlying’ meaning, but also for a meaning of what ‘looks like’ a complex expression but is characterised by the fact that it does not have a (totally) compositional meaning. Thus, it is questionable whether normal speakers of English recognise a connection between the idiomatic and the compositional meaning of kick the bucket, in the sense that they take the idiomatic meaning to be derived from the compositional. One could argue, therefore, that idioms like kick the bucket are to be found on Busse’s first level of language use, the level where no ‘underlying’ meaning 17 Cf. e.g. Giora (1999, 2002), Giora et al. (1997) 32 Against the Standard Notions of Literal Meaning and Non-literal Meaning can be identified. conventions of language unconventional language use conventionalised conventions of language use literal meaning fully idioms more less non-literal meaning not particularised CIs Figure 2.8: Idioms in Busse’s framework In this feature, idioms differ from metaphors in that with metaphors, the ‘underlying’ meaning always has to be transparent or accessible for us to be able to call the phenomenon in question metaphor. Thus, at the point when the underlying meaning of a metaphor is no longer transparent, the metaphor turns into a dead metaphor, effectively seizing to be a metaphor at all. However, that does not mean that certain types of metaphor may not be conventional. It simply means that in order to classify some phenomenon as a metaphor, recognising an ‘underlying’ meaning is necessary, whereas for idioms it is not.18 19 Thus, it seems that whether we categorise a meaning of an expression as non-literal or literal, does not only have to do with whether we are consciously aware of some connection between the meaning in question and some underlying, basic meaning (as in metaphor), but also whether we recognise that a certain expression form can be used with a different, compositional meaning (as with idioms). Thus, idioms are usually taken to be coded in the lexicon, since their meanings have to be learned and since they only superficially look like complex expressions with complex meanings associated with them. That is, in Busse’s terms, they are considered as being ‘conventions of language’, although according to Busse they should not be, since their meaning is perceived to be a so-called non-literal meaning – hence his classification of idioms belonging to the second level, the level of conventions of language use. However, there does not seem to be any independent reason why at least some idioms should not be treated similarly to expressions that are perceived as having no ‘underlying’ meaning. On the contrary, as mentioned above, from a processing perspective, the interpretation of idioms may not be any more ‘costly’ than the interpretation of ‘literal language’, where it seems that the level of conventionality of the idiomatic meaning plays a decisive role here. Thus, Giora recognises a type of idiom that she explicitly calls conventional idiom, where 18 Cp. Stern (2006, p. 250), who makes a similar point. This is, of course, not to say that there are no idioms which are clearly perceived as ‘derived’ in their meaning. 19 Non-literal Meaning and Conventionality 33 the crucial point is that in terms of activation their meanings are similar, if not privileged, to the literal (i.e., compositional) meanings of their compositional counterparts. That is, the meanings of highly salient idioms will be activated as early or even earlier than the meanings of their compositional counterparts, the latter being the case if the compositional meaning of some complex expression is actually nonconventional (in the sense of ‘not normally or often used’). Once again, if ‘earliness of activation’ reflects the level of conventionality of the meaning concerned, then there is no reason why one should assume that the meaning of such conventional idioms is less conventional than the meaning of their compositional counterparts. conventions of language use unconventional language use conventionalised conventions of language fully highly, less & non-salient literal meaning more less not highly & less-salient non-literal meaning non-salient non-literal meaning Figure 2.9: Salient meanings according to Busse’s characterisation Generally, with respect to different types of what traditionally is called nonliteral meanings such as metaphor or idioms, the experimental results differ with the grade of salience that these meanings have. However, according to Busse’s characterisation of levels of conventionality, Giora’s salient and less-salient non-literal meanings belong to Busse’s conventions of language use, whereas the non-salient non-literal meanings are the result of what Busse calls an unconventional language use. As regards highly, less and non-salient literal meanings, according to Busse’s classification criteria, they are all part of the level of conventions of language. In contrast, Giora does not actually make use of the traditional differentiation between literal and non-literal meaning, precisely because the conventional/non-conventional ‘divide’ seems to cut across that distinction. Thus, she assumes that regardless of whether some meaning would traditionally be classified as literal or non-literal, if that meaning is highly salient, it is reasonable to assume that this meaning is coded in the lexicon and thus, is part of the conventions of language. Moreover, Giora assumes not only of highly salient meanings, but also of less-salient meaning that they are part of the lexical semantics of the relevant expressions. That is, an expression’s salient and less-salient meanings form part of its lexical entry, whereas possible, but non-salient readings are pragmatically derived. Thus, Giora’s differentiation between salient, less- and non-salient meanings differs from Busse’s assumptions in an interesting way. Busse assumes that only fully 34 Against the Standard Notions of Literal Meaning and Non-literal Meaning conventionalised meaning is lexical meaning (or a convention of language), whereas less conventional or non-conventional meanings are not lexical. The former may be considered as a convention of language use, but not of language. Giora, on the other hand, assumes that both highly and less-salient meanings are coded in the mental lexicon, and thus are part of the conventions of language, whereas non-salient meanings are not. In terms of the differentation of what is traditionally called literal and non-literal meaning this means that whereas Busse only allows for genuine literal, basic, underived meanings to be entered in the lexicon, Giora allows both literal as well as non-literal meanings in the lexicon, provided they are salient, where salience correlates strongly with conventionality.20 Moreover, she does not only assume of highly salient meanings, but also of more and less-salient meanings that they are entered in the lexicon. What is problematic about this assumption is that one has to ask, how the grade or level of conventionality of an expression’s meaning can actually be measured and thus allows one to identify and differentiate non-salient (and thus non-coded) from less or highly salient (and thus coded) meanings.21 unconventional language use conventionalised conventions of language fully more highly and less-salient meaning less not non-salient meaning literal and non-literal meaning Figure 2.10: Giora’s characterisation within Busse’s framework Returning to the argument that literal meaning and non-literal meaning should not be characterised in terms of dichotomies: the discussion in the section on literal meaning above already suggested that one dichotomy between literal meaning and non-literal meaning cannot be sustained, namely that of context-independent vs. context-dependent meaning. Instead it was suggested that what traditionally is called literal meaning actually is not a complex expression’s context-independent, compositionally derived meaning, but rather a level of meaning that, similarly to 20 Giora takes conventionality to be only one graded property among others that influence an expression’s meaning’s salience, such as frequency of use. Nevertheless, conventionality seems to play an important role. That is, other things being equal, the more conventional an expression’s meaning is, the more salient it will be. Moreover, from the kind of conventionalisation process that Busse suggests, conventionality seems to be closely connected to the frequency of use of an expression with a particular reading. Thus, it can be argued that it is increased frequency of use that leads to a shortening of the inference chain ending in a conversational implicature. 21 This point will be taken up in section 2.3.3. Consequences for Lexical Meaning 35 non-literal meaning is context-dependent. From what has been said above concerning conventionality, it should have become clear that this property cannot be used either in the differentiation of literal from non-literal meaning, since, on the one hand, conventionality seems to be gradual and not binary, and on the other hand, there are instances of what is traditionally called non-literal meaning, which may be characterised as conventional to some greater or lesser degree. Thus, neither the dichotomy of conventional vs. non-conventional, nor that of context-independent vs. context-dependent can be used to differentiate literal meaning from non-literal meaning. As concerns the characterisation of non-literal meaning as always being the result of a reinterpretation process, similar points can be made. If reinterpretation processes are characterised as requiring the trigger of ‘non-compatibility of literal meaning with context’ to get started, then the process by which non-literal meanings are generated should not be called a reinterpretation process. That is, if non-literal meanings only get generated if the literal meaning of an expression does not fit the context, we would expect that non-literal meanings are never activated in contexts where the literal meaning of an expression is intended. As we saw above, this is not the case. Whatever triggers the activation of non-literal meaning, it does not necessarily have to be the non-compatibility of an expression’s literal meaning with the context. 2.3 Consequences for Lexical Meaning As mentioned above, formal semantics traditionally assumed that an expression’s lexical meaning can be derived from the meaning of a sentence the expression is part of via the principle of compositionality. Since sentence meaning was assumed to be context-independent, propositional and literal, lexical meaning is also taken to be context-independent and literal. However, in section 2.1 arguments were given against characterising sentence meaning as both context-independent as well as fully propositional. Rather, sentence meaning was characterised as context-independent, but not necessarily fully propositional. Moreover, arguments were given for not assuming that sentence meaning as such is literal, but rather that it is at some context-dependent and fully propositional level of utterance meaning that we might speak of a literal meaning, where it is not clear as yet how exactly this particular level of meaning is characterised. The crucial point here is that if sentence meaning is no longer considered to be literal but one still wants to assume that lexical meaning can be derived from sentence meaning via the principle of compositionality, lexical meaning can no longer be characterised as being literal either. In fact, I will argue below that the lexical meaning(s) of an expression should not be taken to be full readings at all. First, however, I will present some of the problematic data the semantic component of a grammar may or may not be required to be able to handle (2.3.1). What is important to note here is that the stance one takes in this question has consequences for the amount of lexically given semantic information assumed. I will then go on and discuss two general types of approaches to lexical meaning which differ mainly in how rich they assume the lexical semantic component to be (2.3.2). As we will see, approaches that do not assume underspecification of lexical meaning have quite 36 Against the Standard Notions of Literal Meaning and Non-literal Meaning a number of problems to face concerning the literal/non-literal meaning distinction they assume, the idea of grades of conventionality of meanings and the roles of the principle of compositionality and context in semantic composition. Finally, I want to show that an approach that assumes underspecification for lexical meaning forms an appropriate basis for a view according to which both literal and non-literal meaning are context-dependent and more or less conventional. 2.3.1 Problematic Data Traditionally, it is assumed that the lexical meaning of an expression captures those readings that are ‘basic’ in the sense that they do not depend on an actual utterance context. A potential problem with such an approach is that it leads to a high degree of (non-idiosyncratic) polysemy in the lexicon. Consider example (13). (13) a. The newspaper was on the table. b. The newspaper was censored, yesterday. c. The newspaper was founded three years ago. It seems intuitively obvious that the expression newspaper is understood differently in the individual sentences. Thus, in (13a) it refers to a physical object, in (13b) to a mental object and in (13c) to an institution. The difference in meaning is not due to a difference in the utterance context, which is completely ignored. Traditionally, one might therefore assume that the different readings of newspaper are all coded in that expression’s lexical entry and that newspaper has to be considered as polysemous. However, this type of variation in meaning is not characteristic of newspaper alone. That is, it is not an idiosyncratic property of that expression. Compare example (14) (originally due to Bierwisch 1983). (14) a. Die Schule spendete einen größeren Betrag. ‘The school donated a larger sum.’ b. Die Schule hat ein Flachdach. ‘The school has a flat roof.’ c. Die Schule macht ihm Spaß. ‘He enjoys school.’ d. Die Schule ist eine der Grundlagen der Zivilisation. ‘The school is one of the bases of civilisation.’ Thus, both newspaper and school (or Schule) have in common the ‘physical object’ and ‘institution’ reading. The latter reading allows for a metonymical interpretation as ‘person associated with institution’ of the NPs in (15a) and (15b), respectively. In this respect newspaper differs from book which, although it does have the ‘physical object’ and ‘mental object’ readings, does not have an ‘institution’ reading and therefore does not allow the metonymical interpretation ‘institution-for-person’.22 22 This is not to say that there is no metonymical interpretation possible in principle. Thus, in a suitable context The book in (16d) could, e.g., be interpreted as ‘product-for-producer’. Consequences for Lexical Meaning 37 (15) a. The newspaper called. b. Die Schule hat angerufen. ‘The school called.’ (16) a. The book was on the table. b. The book was censored, yesterday. c. ??The book was founded three years ago. d. ??The book called. The different readings given for the individual nouns in (13), (14) and (16a-16b) are a good example of what is traditionally called systematic polysemy. The individual readings of the respective noun are clearly delimitable, yet at the same time relate to one another. Also, all the readings seem to have the same status: they all intuitively seem to be basic, or literal, in the sense that no underlying meaning from which they are derived is consciously detectable. Moreover, they are not characteristic for only one expression, but rather can be found across sets of expressions. Thus, with respect to the lexical entries of the nouns above, it is questionable, whether one should assume that all of the relevant readings are listed as possible literal meanings of those nouns (cf. Nunberg 1979). Note also that the variation in meaning exemplified above seems to arise from the intersentential context in which the respective expressions occur. Thus, the differences between the readings in (13a-13c), (14a-14d) and (16a-16b) are obvious simply from comparing the respective sentences with one another. It is arguable that sentential context, although being a special type of context in that it does not depend on a particular utterance situation, still is a type of context and that, thus, the readings of the expressions in question is not totally context-free, as it depends on the sentential contexts in which they occur. In fact, one actually needs to use fully sentential examples to make obvious that there are differences in the understanding of the expressions concerned. Moreover, even though the relevant examples are presented out of a particular context, as pointed out above, it is reasonable to assume that hearers or readers always interpret linguistic expressions against a number of background assumptions, which may be taken to have a default character. In contrast to the related readings in case of systematic polysemy, strictly speaking, the metonymical readings for (15a) and (15b) above are not ‘basic’. That is, our interpretation of such examples relies on general knowledge we have regarding institutions, namely, that people (e.g., as employees) are associated with them. One might wonder, whether this reading is also part of the lexical meaning of the respective nouns, as it can be understood in the absence of a particular utterance context. However, as suggested before, the metonymical interpretation ‘person associated with institution’ is generally available for all lexical items which have an ‘institution’ reading, rather than being an idiosyncratic lexical feature of individual lexical items. Thus, the NPs in example (17) with such nouns as university, hospital, police get a similar metonymical interpretation. (17) The university/hospital/police called. 38 Against the Standard Notions of Literal Meaning and Non-literal Meaning In addition, more often than not context does play an important role in determining the exact interpretation of a metonymically used expression.23 (18) a. The ham sandwich in the corner wants some more coffee. b. Plato is on the top shelf next to Russell. c. John got a dent in his left fender. Thus, in (18c), the proper noun John apparently has to be understood as referring to John’s car. In this case the interpretation of John as meaning ‘John’s car’ seems to be constrained by the sentential context, i.e., its co-occurrence with the expressions dent and fender. (18b) is a typical example for the metonymical interpretation of names of authors, which can then be used to refer to books written by those authors. However, again the actual context in which the sentence gets uttered plays an important role. Thus, in a suitable context the terms could be referring to the busts of the respective persons just as well. Similarly, depending on the actual context of utterance, ham sandwich in (18a) can be understood as, e.g. ‘hamsandwich-orderer’, or ‘person-who-prepared-the-ham-sandwiches’. More specifically, in examples like (18a), the sentential context already makes it clear that a shift in the meaning of the noun ham sandwich from ‘object’ to ‘person’ is necessary (especially due to the verb want). However, the exact interpretation or specification of the meaning (i.e., ‘person who prepared . . .’, ‘person who ordered . . ., ‘person who delivered . . . , etc.) depends on the actual utterance context (and, arguably, the speaker’s intentions). Similarly, to understand an utterance of the sentence in (19a), recourse has to be taken to the actual context of utterance of that sentence. In addition, general background knowledge is necessary to provide potential interpretations for an utterance of the sentence. Thus, depending on the actual context of utterance, (19b 19d) are possible readings for the sentence in (19a). (19) a. b. c. d. Joyce is difficult to understand. Joyce’s pronunciation is difficult to understand. Joyce’s behaviour is difficult to understand. Joyce’s books are difficult to understand. That is, in order to correctly interpret an utterance of (19a) in a given situation, the hearer needs to have some general knowledge concerning properties characteristic for the person named Joyce. Given that, it seems even less plausible that all the possible readings for a proper name such as Joyce are listed in that expressions’ lexical entry. Note that this example (as did Searle’s cut-example) shows that verbs have a variation potential with respect to their possible meanings as well. In this particular 23 The same, of course, holds true for particular cases of systematic polysemy. That is, for examples such as in (i) and (ii), the sentential context alone does not make it clear which particular reading of newspaper and school, respectively, is intended. (i) He is not interested in the newspaper. (ii) She paid particular attention to the school. Consequences for Lexical Meaning 39 example, the meaning of understand seems to co-vary with the meaning assigned to the proper name Joyce. That is, understanding someone’s pronunciation, behaviour or books differs conceptually. Another example for such a variability of verb meaning is given in (20). (20) a. b. c. d. e. John John John John John opened opened opened opened opened the the the the the door. exhibition. file. parcel. store. Again, the processes which are named by the verb to open in (20) are intuitively conceived of as being conceptually different. In contrast to the readings of the term newspaper, which are clearly delimitable, this does not seem to be the case for the readings of to open or to understand. Although there is some sort of similarity between the readings for each expression, the relation between them does not seem to be of the kind that one would call systematic. Thus, Dölling and Zybatow (2007) call this particular type of meaning variation of verbs non-systematic and characterise it as, in a sense, idiosyncratic for the respective verb that exhibits it. Nevertheless, similarly to the readings of systematic polysemous nouns, intuitively at least all the readings of open given above seem to be basic. Again the question arises, whether one should assume that all these readings are therefore listed in the verb’s lexical entry making open – and a large amount of verbs generally – polysemous. As was the case for systematic polysemous nouns, the fact that the relevant readings of open are all basic would be masked by assuming that it only has one primary reading from which all other readings are derived. Moreover, there is a further type of meaning variation with verbs that is more similar to the systematic polysemy exhibited by nouns. Thus, consider the examples in (21). (21) a. John opened the door. b. The door opened. In (21a) open has a causative reading, whereas in (21b) it has an inchoative reading. Thus, (21b) describes a change of state, wheras (21a) describes the bringing about of this change of state. This particular type of meaning variation is systematic (cf. Dölling and Zybatow 2007) because, on the one hand, the different readings are related to one another and, on the other hand, it can be found with a number of verbs and, thus, is not idiosyncratic (cf. 22). (22) a. John melted the ice. b. The ice melted. Again, the question is how this particular variation should be treated with respect to the lexical entries of the verbs concerned. On the one hand, one wants to capture the fact that the different readings of the respective form are related to one another, on the other hand, one wants to capture that this particular variation in meaning is not idiosyncratic for one verb. 40 Against the Standard Notions of Literal Meaning and Non-literal Meaning Note that the possible non-systematic readings of a verb also differ depending on the valency with which it occurs (cf. 23). (23) a. Karl hat das Geschirr/ die Wäsche/ den Mund/ . . . gespült. ‘Karl has washed the dishes/ the cloths/ his mouth . . . .’ b. Karl hat seiner Mutter das Geschirr/ die Wäsche/ . . . gespült. ‘Karl has washed the dishes/ the cloths/ . . . for his mother.’ c. (Mit Hinweis auf Geschirr:) Karl hat gerade gespült. ‘(With reference to dishes:) Karl has just washed (the dishes).’ Thus, in (23a) the verb spülen takes two NP arguments, whereas in (23b) it takes an additional benefactive NP argument. Although, in (23c) spülen only takes one overt argument, it is implicitly understood that there has to be something which Karl washed, that is, the NP complement has to be present covertly in the semantic structure of the sentence. In any case, the processes that spülen denotes in the examples in (23) are conceived of as being conceptually different depending on whether and if, what kind of complement NP occurs. Furthermore, the interpretation of intransitive spülen seems fairly restricted to specific contexts. The last set of phenomena I want to mention here are cases of so-called complement coercion (e.g., 24b) and aspectual coercion (e.g., 26b).24 Thus, verbs like to begin and to finish may occur in two different syntactic constructions: either with an event-denoting VP-complement as in (24a) and (25a) or with an entity-denoting object NP as in (24b) and (25b). (24) a. Max began reading a novel. b. Max began a novel. (25) a. Nigel finished eating a hamburger. b. Nigel finished a hamburger. Note that in the latter case the respective sentences are nevertheless understood as predicating of the respective subjects that they began/finished some event involving the entity denoted by the object NP. However, this meaning aspect is not overtly expressed by any of the constituents in the respective sentences. Thus, it seems that the two verbs’ selectional restrictions actually are such that they require eventdenoting complements. Since the sentences in (24b) and (25b) are grammatical, the usual assumption is that during the semantic composition of the meanings of these sentences a process intervenes that turns the entity-denoting object into an event description. Hence this process is called complement coercion. While in complement coercion the ontological sort of the complement is shifted (from entity to event), in aspectual coercion what is changed are the aspectual properties associated with a particular verb in order for it to fit with a given modification. Thus, consider the examples in (26). 24 Note that while coercion phenomena, on a first blush, may not seem to be relevant to the question of how much semantic information is lexically given, considering how the phenomena are explained in different approaches, it turns out that they are relevant after all. Consequences for Lexical Meaning 41 (26) a. The light flashed. b. The light flashed until dawn. Taken in isolation, the verb to flash is understood as denoting a punctual event. As such, a sentence including it as well as a durative modifier such as until dawn should be ungrammatical. However, as (26b) shows, such sentences are grammatical, when understood as involving a repetition of the event denoted by the verb. A similar contrast can be seen in example (27). In (27a), the VP play a sonata denotes an event that already has an intrinsic boundary. If such a bounded event is modified by an adjunct that has the function of introducing a boundary for the event expressed, a conflict should arise. However, example (27b) is grammatical but again only under the interpretation that the subject was involved in the event denoted by the VP repeatedly. (27) a. Amelie played a sonata. b. Amelie played a sonata for three weeks. Thus, it seems that complement coercion and aspectual coercion are variants of one and the same general phenomenon. In both cases a process intervenes during the semantic composition of the relevant sentences and adds some meaning aspect. Put differently, the expectation is that the semantic composition of such sentences should crash as it apparently involves semantic mismatches. However, what seems to happen is that such impending (semantic) mismatches are circumvented by the application of special operators that mediate between the potentially mismatching elements and at the same time introduce additional meaning aspects. All in all, although this is only a very small set of data presented here, it suffices to indicate the kinds of problems one faces when developing a theory of lexical meaning and the role it plays in the overall interpretation process. In the next section, we will start to look at possible approaches to meaning in the lexicon and see how the data given above is treated, as well as how these approaches reflect the revised notions of literal meaning and non-literal meaning as proposed so far. 2.3.2 Approaches to Meaning in the Lexicon Dölling and Zybatow (2007) identify three broad approaches to meaning in the lexicon: maximalist, intermediate and minimalist approaches. Generally, these approaches have in common the assumption that a lexical entry for an expression at least contain some phonetic (PHON), syntactic (SYN) and semantic (SEM) information about that item.25 However, the approaches differ in how much ‘semantic material’ they assume to be given in the lexical entry of an expression, how this material is structured, how it is made use of in the actual interpretation of an utterance containing the relevant expression and how it interacts with the syntactic information given in the entry. Specifically with respect to the lexical entries of verbs, there is the question, whether information concerning argument structure and event structure are part of the lexical information given for a linguistic expression. 25 Since here we are mainly interested in the interfaces between syntax, semantics and pragmatics, in the representation of lexical information the phonetic information will sometimes be neglected. 42 Against the Standard Notions of Literal Meaning and Non-literal Meaning In this section, I will be concerned with the maximalist and intermediate approaches, leaving the discussion of minimalist approaches to the next section. The Maximalist Approach Maximalist approaches to meaning in the lexicon can be characterised by the following features. All possible literal meaning variants of an expression are listed in the lexicon. In addition to differences in the valency of a verb, the sortal restrictions for the individual meaning variants are stated as well. Thus, differences in valency and/or sortal restrictions result in separate lexical entries. Furthermore, the semantic relations holding between the individual meaning variants are stated. In the lexical entries of verbs, information about their argument structure is given as well as the respective verb’s particular event structure. An example of a maximalist approach is Engelberg (2000). For example, he gives the lexical entry, or rather entries, for the verb spülen (see figure 2.11). As one can see from the entry, not only differences in the valency of the verb, but also different sortal restrictions of its arguments are taken into account. As a result, Engelberg can account not only for the different readings of spülen in (28a-28c), but also for those in (28d) and (28e). spülen1 spülen2 spülen3 spülen4 spülen5 syn: V, /npakk /npnom sem: λyλxλe. spül1 (x, y, e) bpspülen1 : ∀xye[spül1 (x, y, e) → geschirr(y) ∨ wäsche(y) ∨ . . . ] syn: V, /npakk /npdat /npnom sem: λyλzλxλe. spül2 (x, y, z, e) bpspülen2 : ∀xyze[spül2 (x, y, z, e) → belebt(x) ∨ belebt(z)] bpspülen3 : ∀xyze[spül2 (x, y, z, e) → spül1 (x, y, e)] syn: V, /npnom sem: λxλe. spül3 (x, y, e) bpspülen4 : ∀xye[spül3 (x, y, e) → geschirr(y)] bpspülen5 : ∀xye[spül3 (x, y, e) → spül1 (x, y, e)] syn: V, /npnom sem: λxλe. spül4 (x, y, e) bpspülen6 : ∀xye[spül4 (x, y, e) → mensch(x)] bpspülen7 : ∀xye[spül4 (x, y, e) → mund(y)] bpspülen8 : ∀xye[spül4 (x, y, e) → spül1 (x, y, e)] syn: V, /npnom sem: λxλe. spül5 (x, y, e) bpspülen9 : ∀xye[spül5 (x, y, e) → waschmaschine(x)] bpspülen10 : ∀xye[spül5 (x, y, e) → wäsche(y)] bpspülen11 : ∀xye[spül5 (x, y, e) → spül1 (x, y, e)] Figure 2.11: Lexical entry for spülen according to Engelberg (2000) (28) a. Karl hat das Geschirr/ die Wäsche/ den Mund/ . . . gespült. ‘Karl has washed the dishes/ the cloths/ his mouth . . . .’ Consequences for Lexical Meaning 43 b. Karl hat seiner Mutter das Geschirr/ die Wäsche/ . . . gespült. ‘Karl has washed the dishes/ the cloths . . . for his mother.’ c. (Mit Hinweis auf Geschirr:) Karl hat gerade gespült. ‘(With reference to dishes:) Karl has just cleaned (the dishes).’ d. (Beim Zahnarzt:) Karl hat gerade gespült4 . ‘(At the dentists:) Karl has just rinsed (his mouth).’ e. Die Waschmaschine hat gerade gespült5 . ‘The washing machine has just rinsed (the cloths).’ Note that for Engelberg, the sentences in (28c - 28d) are ambiguous. That is, only after having considered the context in which an utterance of the sentence Karl hat gerade gespült takes place, can the hearer disambiguate which literal reading of spülen was intended. However, if the context has to be taken into account in order to be able to decide which reading of spülen was intended, the principle of compositionality cannot be maintained. A possible counter-argument is to say that semantic composition actually generates all possible semantic forms for the sentence and only afterwards, during the process of disambiguation, the context is considered and those semantic forms that do not fit it are discarded. In this case the principle of compositionality would be adhered to. However, considerations of economy of the lexicon militate against a maximalist organisation. This is especially so considering that this kind of ambiguity is characteristic for the majority of lexical items in a language (cf. the newspaper, open and cut examples). Furthermore, Engelberg apparently has not reached his aim of listing all possible literal readings for spülen. Thus, Dölling (2005) argues that a reading for spülen as in (29) is yet another one that would have to be listed in the lexicon. Note also that this reading cannot be subsumed under Engelberg’s spülen1 . Probably even more ’literal’ readings could be listed, if one were to do a corpus analysis to find attested examples (cf. Ruhl 1989, on data won through intuition). (29) (Mit Hinweis auf die Toilette:) Karl hat gerade gespült. For maximalist approaches, the process of semantic composition is essentially as assumed by traditional formal semantics. That is, the process follows the principle of compositionality and therefore, does not take into account any contextual information. Thus, for an utterance such as Karl hat gerade gespült, three semantic forms are generated by the process of semantic composition, as there are three lexical entries for spülen in its intransitive form. Thus, although the lexicon is very rich in what is usually considered world knowledge, this never really comes into play, since the disambiguation process that follows semantic composition simply checks, whether the information given in the individual semantic forms fits the information provided by the contextual conditions. In a sense, then, the information stated in the individual lexical entries for intransitive spülen is redundant. Moreover, if the lexical entry for an expression only lists its literal meaning(s), one would have to assume a Gricean interpretation process, where necessarily a 44 Against the Standard Notions of Literal Meaning and Non-literal Meaning literal reading for an utterance has to be derived first via semantic composition, which, in case of a non-fitting context, has to be reinterpreted. Assuming that all literal readings are listed in the lexicon for each lexical item would make the disambiguation process a very important part of the overall interpretation process and arguably, a rather extended one in terms of the time needed for that process to finish. If literal meanings are context-independent and coded in the lexicon, the process of semantic composition will determine as many possible sentence meanings as there are possible combinations of the literal meanings of the expressions concerned. Only in a second step, when contextual information is considered and the sentence meanings are enriched to the point that they are actually propositional, does the disambiguation process filter out all those propositions that do not fit the given context. Even assuming that all of this takes place during the short period of time that addressees usually only require to understand some utterance, one would still assume that of the time span needed for interpretation, a relative large amount is spend disambiguating potential sentence meanings. However, the fact that potential non-literal meanings of some expression may be activated very early on during interpretation goes against such a view. Note also that an interesting question here is, which semantic form the reinterpretation process would be based on if during the first pragmatic interpretation, and due to a non-fitting context, none of the potential semantic forms derived during semantic composition could be chosen as fitting. Thus take once more the examples (15a) and (18), repeated below. (15a) The newspaper called. (18) a. The ham sandwich in the corner wants some more coffee. b. Plato is on the top shelf next to Russell. c. John got a dent in his left fender. Approaches that assume that semantic composition combines the literal meanings of simple expressions and thus produces the resulting complex expression’s literal meaning would be hard pressed to give the literal meaning of (18a) from which the metonymical reading is taken to be derived. This is especially so for approaches that assume that the sortal restrictions for lexical items form part of their lexical entry. In that case, arguably, the verb want would be specified as requiring an argument that has volition, which, literally, is not the case here. A possible way out would be to assume that ham sandwich is polysemous, one of its possible readings being ‘ham sandwich orderer’, and that this reading is listed in the lexicon. The question is whether one would like to assume such an entry and thus to make the process of metonymic shift a lexical one. As noted above, in view of the contextdependency of the results of an application of the process of metonymic shift, this is unlikely. Thus, as mentioned before, Plato in (18b) might be interpreted as ‘the person Plato’, ‘the book(s) written by Plato’ or possibly ’the bust of Plato’, depending on the particular context of an utterance of (18b). Regardless of how the question of which semantic form the re-interpretation process would be based on might be answered, the fact is that a maximalist approach that only allows literal readings of an expression to be listed in that expression’s Consequences for Lexical Meaning 45 lexical entry cannot account for cases of similar speed in, or even faster, interpretation of certain, conventionalised non-literal meanings. Thus, such a characterisation of meaning in the lexicon seems to assume the traditional view of a dichotomy of conventionalised vs. non-conventionalised meaning, for which, as we saw, there are arguments that it is not necessarily warranted. That is, a maximalist lexical entry does not reflect the fact that there are levels of conventionality rather than just a differentiation between meanings that are conventional and such that are not. The Intermediate Approach Intermediate approaches employ one of two general strategies. Either one specific meaning variant of an expression serves as the basis from which all other readings are deduced, or an abstract semantic representation provides the basis for all primary meaning variants of the expression. One example for an intermediate approach, using the first strategy is Pustejovsky (1991, 1995). With respect to the lexical entry for verbs, he states one specific meaning as their basic meaning, from which all others have to be deduced. Lexical entries are structured and involve a number of levels of representation: argument structure, event structure, qualia structure and lexical inheritance structure. For example, the lexical entry for the verb to bake looks as in figure 2.12. bake " # eventstr = e1 = e1 : process Head = e1 " # animate ind 1 arg1 = formal = physobj " # argstr = mass arg2 = 2 formal = physobj " # state change lcp qualia = agentive = bake act(e1 , 1 , 2 ) Figure 2.12: Lexical entry for to bake according to Pustejovsky (1995) Thus, not only the syntactic order of lexical items, but also the semantic content of their entries may be decisive for how lexical items are combined. That is, the lexicon is actively involved in the semantic composition of sentences. For example, with the help of the information given in the qualia structure of bake, together with information from the lexical entry of the respective complement bake takes, problematic cases as in (30) can be explained. (30) a. Mary baked a cake. b. Mary baked a potato. 46 Against the Standard Notions of Literal Meaning and Non-literal Meaning Whereas in (30a) the verb to bake denotes the act of creating something, in (30b) it denotes a change of state. As can be seen from the lexical entry of bake, its basic meaning is that of denoting a change of state, thus, the reading that needs to be explained is that in (30a). Information about the internal argument, in this case the noun cake, is used to shift the verb’s event type. Intuitively, an object such as a cake is brought about by the activity it is in composition with. This is expressed by the information stated in the qualia structure. Thus, the agentive quale of the noun cake makes reference to the process of baking. cake " = argstr = arg1 d-arg1 = const formal qualia = telic agentive # y x eat(e2 , z, x) bake act(e1 , w, y) x:food ind y:mass = = = = Figure 2.13: Lexical entry for cake according to Pustejovsky (1995) A number of operations take place in combining the lexical structures of bake and (a) cake. Amongst them is the process of co-composition, which results in a semantic structure for bake a cake that incorporates aspects of both the constituents. Thus, in combining the lexical information of bake with that of cake, the resulting VP is interpreted like a genuine creation verb such as build. The resulting semantic form of the VP bake a cake looks like in figure 2.14. Thus, systematic (or as Pustejovsky calls it: logical) polysemy is explained on the level of expression meaning with the help of lexical semantic processes that compose larger semantic structures out of the lexical semantic information of the individual expressions. However, it is not clear what leads Pustejovsky to assume that the ‘change of state’ reading is the underlying, basic meaning of bake, especially since in the examples in (30), neither the one nor the other seem to have a derived meaning. Generally, an advantage of Pustejovsky (1991, 1995)’s approach is the fact that it provides a framework to model the role of sentential context in interpretation. Thus, the process of co-composition might be used to explain the phenomenon that the reading of a particular expression may be determined before the context of utterance is actually considered. That is because the lexical entries of linguistic expressions are relatively rich, consisting of different types of information, where that information interacts with the information coded in the lexical entries of expressions which occur in the same sentence or phrase. Note also, that Pustejovsky (1991, 1995)’s approach allows for a relative straightforward modelling of complement coercion in terms of the process of co-composition. A sentence like (24b) – repeated below – for instance, is interpreted by using information stored in the qualia structure of the lexical entry for novel. Consequences for Lexical Meaning 47 bake a cake e1 = e1 :process e = e2 :state 2 = evenstr Restr = <∝ Head = e 1 " # animate ind = 1 arg1 formal = physobj artifact arg2 argstr 2 const = 3 = = formal = physobj " # material 3 d-arg1 = formal = mass create-lcp qualia 3 ) formal = exist(e , = 2 agentive = bake act(e1 , 1 , 3 ) Figure 2.14: Semantic representation for the VP bake a cake according to Pustejovsky (1995) (24) a. Max began reading a novel. b. Max began a novel. In case of an interpretation of (24b) as equivalent in meaning to (24a), novel’s telic role – namely that its purpose is to be read – is made use of. This information is integrated into the composition of the semantic form of the respective sentence by means of a special operator (QT ) which intervenes between the verb to begin and the NP a novel. Pustejovsky (1991, p. 430) gives the following derivation of the meaning of (24b).26 (31) a. b. c. d. e. f. Max began a novel. begin0 QT (a novel) (Max) begin0 λx, eT [read(a novel)(x)(eT )] (Max) Max{λx[begin0 λx, eT [read(a novel)(x)(eT )](x∗ ) (x∗ )]} Max{λx[begin0 λeT [read(a novel)(x∗ )(eT )] (eT )]} begin0 λeT [read(a novel)(Max)(eT )] (Max) Note, however, that several problematic issues can be identified for Pustejovsky (1991, 1995)’s approach. One such aspect is that the content and formality of the qualia structures are only loosely defined. There seem to be no convincing reasons for choosing a particular element over another as being part of the qualia. In fact, 26 Actually, his example sentence is John began a novel. 48 Against the Standard Notions of Literal Meaning and Non-literal Meaning Egg (2005) questions whether it is generally appropriate to include such parts of background knowledge in the lexicon as the coming about of some entity a lexical item can be used to refer to. Another problematic aspect is that if systematic reinterpretation becomes necessary during semantic composition, as we saw above, Pustejovsky argues for the application of special transfer functions (such as e.g., QT ), which are based on conceptual knowledge. It is not clear how the necessary access to conceptual knowledge is controlled. Moreover, if such transfer functions apply during semantic composition, the principle of compositionality is not adhered to (cf. Dölling 2001). Furthermore, transfer functions making use of material in the qualia structure of lexical items would have to be overruled when it comes to reinterpretation guided by contextual information. That is, the information that can be retrieved from the qualia structure can only have the status of default knowledge; otherwise the transfer functions would fail to generate the appropriate number of possible reinterpretations for an utterance. For instance, recall that the information retrieved from the lexical entry of novel and integrated into the semantic form of a sentence such as (24b) in Pustejovsky (1991, 1995)’s approach is taken to be semantic in nature. However, depending on the context in which (24b) is uttered, the process the speaker might implicitly refer to may not necessarily be a reading or writing process.27 Generally, the fact that a rich context may facilitate the interpretation of an expression with a less-conventional, but intended reading without any major noticable effects attributable to extensive re-interpretation processes suggests that the type of information used in cases of ‘out-of-context’ interpretation should be treated as having default character instead of being semantic in nature. Thus, this type of knowledge is applied only if there is no more specific (contextually provided) information that can be applied. Therefore, it should not be assumed to be semantic, if one wants to assume that semantic information always enters into the composition of the meaning of an utterance, regardless of context of or intentions behind some utterance. Thus, the type of information proposed by Pustejovsky in the form of qualia structure may actually play an interesting role when it comes to interpreting some linguistic utterance, however, this structured information is not available at the level of lexical entries but rather at the level of encyclopaedic knowledge. Moreover, one could argue that at this latter level, certain types of information are easier to access or privileged in the sense that they are the concepts certain expressions are taken to express in default situations. Also, it is at this level of information where it seems to make more sense to assume that we store facts such as for example how certain artefacts come about.28 Egg (2003, 2005) notes another problem for Pustejovsky (1991)’s particular formulation of the coercion operator. As can be seen in the analysis of (24b) in (31), 27 Egg (2005) mentions a context in which it is known that Max is a professor of literature, where (24b) might reasonably be understood as referring to Max’s beginning of a lecturing process on some novel. Crucially, this interpretation intuitively does not seem to require the revision of a previously computed reading or writing interpretation. 28 Incidentally, similar arguments can be made concerning such information as argument and event structure. That is, the question is whether the knowledge we seem to have about the number and type of obligatory or optional participants involved in some event needs to be reflected in our lexical knowledge. Consequences for Lexical Meaning 49 the verb to begin’s semantics ends up outscoping that of the quantifier in the objectNP. That is, since the coercion operator intervenes between the overt functor and argument, it fixes the scope of the quantifiers involved. However, this does not seem to be what is actually happening. Thus, consider (32). (32) a. Max began reading every novel. b. Max began every novel. Both (32a) and (32b) may be understood to describe a situation in which John read the initial part of each of a contextually specified range of novels. However, (32a) but not (32b) has another possible reading in which John is described as involved in the initial phase of an eventuality that is a reading of the contextually specified range of novels. This latter reading is compatible with a situation where John in fact is engaged in reading the initial part of just one of the contextually specified range of novels. Thus, whereas in (32a) either of to begin and every may take wide scope with respect to one another, in (32b) every always has wider scope. However, this state of affairs is exactly opposite to what is predicted by Pustejovsky (1991)’s analysis.29 Leaving these issues to the side for the moment, I will turn to an example of an intermediate approach which uses the strategy of providing an abstract semantic representation as the basis for all primary meaning variants of an expression: Bierwisch’s Two-Level-Semantics. As already noted, Bierwisch assumes that an expression’s literal or primary reading is only determined at the level of utterance meaning. In that sense, an expression’s semantic form has to be quite abstract and underspecified. In order to build up the linguistically determined meaning of a complex expression, the process of semantic composition combines the individual semantic forms, following the principle of compositionality. The main feature of this approach is the fact that the semantic information given in the lexical entry of an item is underspecified and only gets specified during the actual interpretation process of an utterance including that expression. Thus, on the one hand, the semantic form of an expression specifies the contribution of that expression to the semantic representation of utterances of which it is part. On the other hand, the semantic form of an utterance constitutes the interface with the conceptual-intentional system. That is, an interpretation function maps the linguistically determined semantic form of an utterance onto the utterance meaning, depending on the context in which it is used. Thus, only when an utterance gets actually interpreted, will its particular reading get fixed. For instance, for the noun Schule, which, similarly to the noun newspaper, can occur with different readings, Bierwisch (1983) gives the semantic form stated in (33), which is compatible with any of the readings Schule can be given in (14). (33) SEM: λx [ zweck x w ] w = lern- und lehrprozesse 29 Note that in Egg (2005)’s approach this problem does not arise since the process of semantic composition as such is characterised by the fact that it produces a semantic structure that is underspecified with respect to scope. For details of that approach, see below. 50 Against the Standard Notions of Literal Meaning and Non-literal Meaning (14) a. Die Schule spendete einen größeren Betrag. ‘The school donated a larger sum.’ b. Die Schule hat ein Flachdach. ‘The school has a flat roof.’ c. Die Schule macht ihm Spaß. ‘He enjoys school.’ d. Die Schule ist eine der Grundlagen der Zivilisation. ‘The school is one of the bases of civilisation.’ The actual ‘institution’ (14a), ‘physical object’ (14b), ‘assembly of processes’ (14c) or ‘institution as principle’ (14d) readings only get determined, or specified, during the interpretation of the respective utterances. Generally, no primary reading is given in the semantic form of an expression, however, the semantic form does determine as a possible interpretation a family of concepts for the lexical item it is part of.30 This differentiation reflects Bierwisch’s view of the lexicon as the repository of linguistically relevant information. Because they are not necessary for any linguistic process to function properly, an expression’s possible literal meanings need not all be listed in its lexical entry. Thus, for the noun Schule, the different meaning variants constitute a family of concepts, or more specifically, a set of conceptual schemata, in terms of which its semantic form can be interpreted, depending on the particular context. The interpretation function then chooses a member of that family as the actual interpretation in a given situation, depending on the particular context. Thus, in Bierwisch’s approach systematic polysemy no longer is treated as a phenomenon at the level of expression meaning, rather, it belongs to the context-dependent level of utterance meaning. However, although the semantic form of an expression is abstract and underspecified, similarly to Engelberg’s approach, information on the argument structure is part of a verb’s lexical entry. Thus, the lexical entry for a verb such as to enter looks as in (34) (adapted from Bierwisch and Schreuder 1992, 27). (34) PHON: /enter/ SYN: [+V, -N] SEM: λxλy [y DO [MOVE y] : FIN [y LOC IN x]] As in Engelberg’s approach, argument structure and semantic form of a particular lexical item are closely interrelated pieces of information. However, in contrast to Engelberg, differences in valency do not automatically lead to separate lexical entries. This can be seen from the underspecified lexical entry for open, which is intended to reflect that lexical item’s variable valency. That is, rather than assuming several lexical entries related by meaning postulates, Bierwisch tries to capture the relation between the different readings of open, by integrating the possibility of different valencies for a lexical item into its syntactic information, its argument 30 However, it is not clear what it means for the SEM of some expression to determine a family of concepts as possible readings for the expression at hand. Consequences for Lexical Meaning 51 structure as well as its semantic form. Thus, he even considers the possibility of including the adjective open in one and the same entry with the intransitive and transitive verb forms, in order to keep the lexically given information as minimal as possible (cf. 35) (adapted from Bierwisch 1997, 244). (35) PHON: /open/ SYN: [+V, -βN] SEM: λx (β (α λy) λe [ e INST [ (α y DO-CAUSE) [ BECOME) [ OPEN x ]]]] 2.3.3 Semantic Underspecification in the Lexicon The Minimalist Approach Although in Bierwisch’s approach already the semantic form of lexical items was no longer assumed to be maximally specified, an approach is possible which minimises the lexical information even more. Thus, minimalist approaches to lexical meaning are characterised by the following facts. The lexical meaning of an expression is maximally underspecified. Furthermore, the semantic form of an expression functions as a pointer31 to possible meaning variants of that expression. Thus, similarly to Bierwisch’s approach, the intended literal (or non-literal) reading is derived only during the interpretation process. Moreover, information on possible argument structures an expression might occur in are not part of its semantic information. An example for a minimalist approach is Dölling (1997, 2005), who gives the following lexical entry for the noun newspaper. (36) PHON: /newspaper/ SYN: N SEM: λx. x Kn newspaper Kn is an SF32 -Parameter which is part of the lexical meaning of all common nouns. The set of possible values for this parameter includes the predicates (‘instance of’) and ≤ (‘sub-kind of’). Thus, the semantic form in (36) can be specified either to a predicate which denotes the set of instances of newspapers or one which denotes the set of sub-kinds of the kind newspaper. Furthermore, the conceptual entity ‘newspaper’, which forms part of the lexical entry for newspaper, is an SF-Parameter as well. It identifies a certain kind, namely, the kind that is realised by each newspaper-asmental-object, newspaper-as-material-thing and newspaper-as-institution. The axioms in (37) make it clear that ‘newspaper’ is a specific combination of newspaper1 , newspaper2 and newspaper3 , which are sub-kinds of the categories of mental objects, material things and institutions, respectively. (37) a. newspaper = [newspaper1 ` newspaper2 ] b. newspaper1 ≤ ment.object c. newspaper2 ≤ mat.thing 31 ` newspaper3 This term was first introduced in Relevance Theory, more on which see below. SF is the abbreviation Dölling uses, following Bierwisch, for semantic form, here generally abbreviated to SEM. 32 52 Against the Standard Notions of Literal Meaning and Non-literal Meaning d. e. f. g. newspaper3 ≤ institution ∀x [x ≤ ment.object & x ≤ newspaper → x ≤ newspaper1 ] ∀x [x ≤ mat.thing & x ≤ newspaper → x ≤ newspaper2 ] ∀x [x ≤ institution & x ≤ newspaper → x ≤ newspaper3 ] The determination of a specific reading takes place only during the interpretation of an utterance that includes the noun newspaper. Thus, as is the case with Bierwisch’s approach, systematic polysemy is not a phenomenon of the level of expression meaning, as the intended reading for the noun is only determined during interpretation. The method of maximal underspecification is also used for verbs. Thus, Dölling (2005) states the following lexical entry for the German verb öffnen (to open). (38) SYN: V, [α /NPakk ] [β /NPdat ] /NPnom SEM: λe.ÖFFN(e) PRAG: If α = - and β = -, then <ÖFFN > = {ÖFFNk1 , ÖFFNk2 , . . . } If α = + and β = -, then <ÖFFN > = {ÖFFNl1 , ÖFFNl2 , . . . } If α = + and β = +, then <ÖFFN > = {ÖFFNm1 , ÖFFNm2 , . . . } In this entry, only the syntactic information SYN expresses the possible variation in valency for the verb öffnen. That is, depending on which value replaces α or β, one gets an inchoative or causative reading, or the causative reading with an additional benefactive complement. In the semantic form SEM, however, there is only one variable for the event argument. Thus, argument structure and semantic form are separate parts of information. SEM includes the SF-Parameter ÖFFN, which functions as a pointer to the set of possible primary meanings of the verb öffnen. The fact that specific valencies in which a verb can occur determine its interpretation is reflected in the information given in the PRAG component. However, this information is only accessed during the interpretation process and ensures that the relevant reading is picked, depending on the syntactic environment in which the item occurs, from the appropriate set of possible primary meanings. Thus, a lexical entry (and as a result the lexicon as a whole) actually consists of two parts: a grammatical one and a pragmatic one. The grammatical part is accessed during the semantic composition of the meaning of a complex expression, which, as in Bierwisch’s and Engelberg’s approaches, follows the principle of compositionality. The pragmatic part only comes into play once the result of semantic composition, the sub-propositional semantic form of an utterance is further processed by the pragmatic component which yields the actual propositional meaning of the utterance. Thus, the lexicon is not only a repository of information required by the individual linguistic subsystems, but it also constitutes an interface to the pragmatic system, providing information relevant for the correct interpretation of an utterance. Note that in Dölling (1997, 2005)’s approach, during the semantic composition of an utterance’s SF, operators may apply that are not the meaning of some syntactically provided functional head. However, such operators only provide variables that might be filled with values by pragmatic processes after the actual semantic composition process is finished. For example, using the operator in (39), Dölling (2005) proposes a treatment of coercion, where that operator is introduced during Consequences for Lexical Meaning 53 the process of semantic composition, providing parameters which have to get fixed or specified in a later step after semantic composition has been completed and recourse may be taken on conceptual knowledge. In fact, the same operator applies to all one-place predicates during semantic composition, regardless of whether there is an impending semantic mismatch. In other words, the operator applies by default.33 (39) metn : λPλx. Qn y [S n (y, x) C n P(y)]34 Using the example in (40), Dölling (2005) shows how different potential readings of that sentence – given in (41) – might be derived from one underlying semantic form (43l). The respective lexically given semantic forms for the expressions/functional heads used in (40) are given in (42).35 (40) Anna hat die Tür fünf Minuten lang geöffnet. (41) a. ∃p[AG(anna, p) & TH(def TÜR, p) & ∀e[CONST(e, p) → ÖFFN1 (e)] & τ (p) ≥ 5min] b. ∃p[AG(anna, p) & TH(def TÜR, p) & ∃e[COMPL(e, p) & ÖFFN1 (e)] & τ (p) ≥ 5min] c. ∃e[AG(anna, e) & TH(def TÜR, e) & ÖFFN1 (e) & ∃s [RESULT(s, e) & τ (s) ≥ 5min]] (42) a. b. c. d. e. f. g. öffnen: λx. ÖFFN(x) fünf Minuten lang: λx. τ (x) ≥ 5min MOD: λQλPλx. P(x) & Q(x)36 θakk : λPλyλx. θakk (y, x) & P(x) die Tür: def TÜR θnom : λPλyλx. θnom (y, x) $ P(x) Anna: anna (43) a. b. c. d. e. λPλx. Qn y [S n (y, x) C n P(y)] (λx. ÖFFN(x)) λx. Q1 y [S 1 (y, x) C 1 ÖFFN(y)] λPλx. Qn y [S n (y, x) C n P(y)] (λx. τ (x) ≥ 5min) λx. Q2 z [S 2 (z, x) C 2 τ (z) ≥ 5min] λQλPλx. P(x) & Q(x) (λx. Q2 z [S 2 (z, x) C 2 τ (z) ≥ 5min]) (λx. Q1 y [S 1 (y, x) C 1 ÖFFN(y)]) f. λx. Q1 y [S 1 (y, x) C 1 ÖFFN(y)] & Q2 z[S 2 (z, x) C 2 τ (z) ≥ 5min] 33 Note that Dölling (2000) uses the same operator (in a slightly different form) to compose metonymic or metaphoric readings of nouns (see section 4.1.1). 34 Qn and C n as well as S n are SF-Parameter which need to get fixed during interpretation, Qn by either ∀ or ∃, C n by → or =, respectively and S n by a predicate of relations between elements of two sorts of entity. 35 Note that due to the fact that in Dölling’s approach generally verbs are highly underspecified regarding their semantics, additional operators are necessary (i.e., θnom , θakk ) that introduce the arguments with which the respective verb occurs in a particular sentence. 36 This is the operator used to integrate the semantic contribution of potential adjuncts into the semantic form of the complex expression under consideration. Note that this operator is solely semantic, i.e., it is not the meaning of some overt syntactic element. 54 Against the Standard Notions of Literal Meaning and Non-literal Meaning g. λPλyλx. θakk (y, x) & P(x) (λx. Q1 y [S 1 (y, x) C 1 ÖFFN(y)] & Q2 z[S 2 (z, x) C 2 τ (z) ≥ 5min]) h. λyλx. θakk (y, x) & Q1 y [S 1 (y, x) C 1 ÖFFN(y)] & Q2 z [S 2 (z, x) C 2 τ (z) ≥ 5min] (def TÜR) i. λx. θakk (def TÜR, x) & Q1 y [S 1 (y, x) C 1 ÖFFN(y)] & Q2 z [S 2 (z, x) C 2 τ (z) ≥ 5min] j. λPλyλx. θnom (y, x) & P(x) (λx. θakk (def TÜR, x) & Q1 y [S 1 (y, x) C 1 ÖFFN(y)] & Q2 z [S 2 (z, x) C 2 τ (z) ≥ 5min]) k. λyλx. θnom (y, x) & θakk (def TÜR, x) & Q1 y [S 1 (y, x) C 1 ÖFFN(y) & Q2 z [S 2 (z, x) C 2 τ (z) ≥ 5min] (anna) l. λx. θnom (anna, x) & θakk (def TÜR, x) & Q1 y [S 1 (y, x) C 1 ÖFFN(y) & Q2 z [S 2 (z, x) C 2 τ (z) ≥ 5min] The result of the process of semantic composition for the sentence in (40) is the underspecified semantic form given in (43l). As we saw above, (40) might get any of the readings stated in (41) (and more). The different readings are derived by different fixations of the underspecified parameters in (43l) in correlation with a redefinition of the unbound variables x, y and z as sorted variables for eventualities.37 Thus, the different readings are not assumed to arise as a result of some co-composition process. Rather, the semantic contribution to the interpretation of an utterance of (40) in all three cases is the same. It is only once the compositionally determined SF of that sentence is ‘given over’ to the pragmatics component that the underspecified parameters and variables get fixed/sorted. Concentrating on Dölling (1997, 2005)’s assumptions concerning the nature of lexically stored information; a general point to note is that integrating pragmatically relevant information into the lexical entries is not unproblematic. Thus, one argument for underspecification of semantic information in the lexicon was that it reduces its size and makes it more economic. In the approach suggested by Dölling, however, the reduced semantic information is countervailed by the additional PRAG component. More specifically, the question arises what such labels as ÖFFNk2 or ÖFFNl2 refer to. From what Dölling says, they are actual conceptual constants which may be used to fix the parameter ÖFFN. If that is the case, then such a lexical entry, actually, would not differ much from those found in maximalist approaches, since the possible readings of an expression, although no longer listed in the semantic component of an item’s lexical entry would then be listed in its pragmatic component. Moreover, characterising the semantic form of an expression as a ‘pointer’ to the set of possible primary readings for that expression is thus superfluous, since semantic form as well as all possible primary readings are actually part of one and the same lexical entry. Also, with respect to the subsets identified in the PRAG component, the question arises, how they are determined and by which criteria their members are characterised. Dölling states that he takes the readings given in the PRAG component to be conventionalised and primary in that they are not derived from some underlying full reading. From this perspective, it seems that instances of non-literal readings that are highly salient in Giora’s terms are not part of the readings in PRAG. The question is, whether this is a desirable state of affairs. 37 For particulars, please refer to Dölling (2005). Consequences for Lexical Meaning 55 Having said all the above, the major advantage of such an approach that assumes maximal underspecification of lexical meaning is that it allows for strict semantic composition without that process having to generate numerous potential semantic forms for a particular utterance which, in a second step, as it were, have to be checked against the actual context of utterance for the actual semantic form to be disambiguated. Ruhl’s monosemic approach Another approach to lexical meaning that assumes a high degree of underspecification of semantic information in the lexicon is put forward by Ruhl (1989), who assumes that an expression’s lexical meaning, or semantic representation is very abstract and needs to be enriched by pragmatic processes in order to yield any full-fledged reading. He argues that the literal meaning we take an utterance to express actually results from an unconscious cognitive process, which makes use of language-external information. Thus, generally, Ruhl does not subscribe to the standard notion according to which the literal meaning(s) of a lexical item is/are what is stated in its lexical entry. Rather, he assumes an abstract semantic representation which is then specialised or generalised through what he calls pragmatic modulation, where modulation can take the form of pragmatic specialisation on the one hand, but also of pragmatic generalisation on the other38 . Thus, much of what has been assumed to be semantic aspects of a given word are actually pragmatically determined aspects of the meaning of that word. Moreover, Ruhl (1989) argues that one should not go by intuition concerning the possible readings an expression may have, since intuitive listing is very selective. Thus, looking at a dictionary entry for some expression, what is described are possible uses of an expression, however, dictionary entries by no means include all the possible readings in which an expression can occur, not even all its so-called literal meanings. Ruhl takes to bear as an example. Looking at the readings for to bear given in the Oxford English Dictionary, he argues that they do not exhaust to bear’s possible literal readings. Crucially, readings that cannot be described with the help of other expressions are ignored. He compares this to the results of a corpus analysis, where many more different, but still literal meanings can be differentiated. He then argues that the general problem with examples drawn from intuition is that they are the readings that come to mind first and they in turn are stereotypical for the expression under consideration, but they are by no means the only readings a particular expression may be used in. In order to avoid rampant lexical polysemy, Ruhl assumes of simple expressions that what is coded in their lexical entries is an abstract semantic representation which is worked on by pragmatic processes. As mentioned before, Ruhl uses the notion of stereotype to explain why, for a given example, we arrive at what we think of as the literal meaning of that expression. For example, if we were to give a description of the meaning of the verb hit, presumably the first thing that would come to our minds would be something along the lines of ‘strong physical contact’. 38 Note that these two processes are very similar in kind to those that in relevance-theoretic terms are called narrowing and loosening, respectively. 56 Against the Standard Notions of Literal Meaning and Non-literal Meaning We take this to be the ’basic’ sense of the expression. However, Ruhl argues that this supposed ‘basic’ sense is the expression’s stereotypical meaning. Like the other readings hit may have, the sense ‘strong physical contact’ is a pragmatic inference. It only differs from other possible readings in that it is a ‘“favored” pragmatic inference’ (Ruhl 1989, p. 122). Unfortunately, Ruhl (1989) does not explain what it means for a pragmatic inference to be ‘favoured’, that is, what it is that makes a reading a stereotypical one. He does state that stereotypes may be triggered by the immediate sentential context (as, for example, the interpretation of take as ‘steal’ in The thief took the jewels.). Very often, however, as in the case of hit, stereotypical meanings seem to not depend on any concrete context. But, as Ruhl argues, this is only an illusion since interpreters always make assumptions about contexts of utterances of expressions, although not consciously so. Thus, words never ‘mean’ in isolation. A Cognitive Approach A further view of the relation between lexical items and the meanings they encode – one not primarily aiming at formalisation, but rather cognitive plausibility – is offered by Relevance Theory. Originally, Relevance Theory (RT) assumes that (monomorphemic) expressions encode atomic concepts. On the one hand, these concepts form constituents of the logical form of an utterance the respective lexical item is part of. On the other hand, a concept is taken to be an address in memory, making available a variety of information. This information is structured under three distinct ‘entries’: a lexical, a logical and an encyclopaedic entry. For example, for a concept like robin, the lexical entry consists of specifically linguistic information (phonological and syntactic information), the logical entry consists of inference rules which apply to mental representations that include the concept robin. The encyclopaedic entry, then, includes all kinds of information about robins, that is, about the extension of the concept, in form of a set of assumptions and assumption schemas. Sperber and Wilson (1995) describe the relation of a concept as an address in memory and the information it makes available as follows. . . . the distinction between address and entry is a distinction between form and content, the address being what actually appears in logical forms, and the various entries spelling out its logical, lexical and encyclopaedic content. (Sperber and Wilson 1995, p. 92) However, if this is how concepts are generally characterised, then the question arises in which sense they are supposed to be atomic.39 Moreover, under such a characterisation of ‘concept’, we get the undesirable situation that a lexical item’s phonological form and syntactic information would be part of its meaning, as they are part of the lexical entry that forms part of the concept the lexical item in question is taken to encode (cp. Groefsema 2007). Thus, the concepts encoded by linguistic expressions 39 Incidentally, consider also the following quote from Sperber and Wilson (2008, p. 12) talking about the interpretation of temperature in I have a temperature: ‘This concept has a parameter that can take a range of values . . .’ (my emphasis). Consequences for Lexical Meaning 57 should not be taken to include lexical entries. Rather, as Groefsema (2007) suggests ‘. . . activation of linguistic information of a word (e.g. activation of its phonological form) gives access to a conceptual address, and vice versa.’ (Ibid., p. 138-9). Carston (2002c) actually proposes that what lexical items encode need not be full-fledged concepts but rather may be concept schemas, that is, very abstract ‘meanings’ in that they simply constitute ‘pointers’ which indicate the conceptual space that is the basis for inferring the actual reading that an expression will get in a particular utterance situation. This assumption goes well together with the general assumption that the semantic form of a lexical item is not of a different format than concepts in general but rather is a particular kind of conceptual structure. Thus, the lexical meaning of an expression only provides the basis from which pragmatic enrichment processes can take off, resulting in the more specific concepts the expression can be used to express.40 The information available in this ‘conceptual space’ is organised with respect to accessibility, where accessibility of certain elements of the various types of encyclopaedic information is a function of (amongst others) frequency and recency of use of that element in processing. Thus, for an expression like open, the concepts that this expression is used to communicate most frequently may be considered to be more or less ready-made retrievable. Generally, the characterisation of lexical meaning as underspecified is in line with RT’s overall assumption that literal meaning and non-literal meaning are not as different as traditionally assumed. Thus, an expression’s lexical meaning is neither literal nor non-literal; it is not even a full-fledged concept. That is, both the literal as well as potential non-literal meanings of an expression result from the application of pragmatic processes in interpretation. Thus, with both literal and non-literal meaning resulting from similar processes, there is no reason to assume that the former is necessarily prior to the latter in interpretation. Figure 2.15: Literal and non-literal meaning in Relevance Theory In this respect, the relevance-theoretic approach is similar to Giora’s approach, where it is not the traditional differentiation of literal and non-literal meaning, but rather the degree of salience that a certain meaning of a particular expression has that is responsible for the relative ease or difficulty of its interpretation. As already 40 See also Assimakopoulos (2008) for a discussion of RT’s original conception of encoded meaning and its problems. 58 Against the Standard Notions of Literal Meaning and Non-literal Meaning mentioned in section 2.2.2, Giora assumes that the salient and less-salient meanings of an expression are coded in that expression’s lexical entry. To repeat, the salience of a meaning is a function of its conventionality, familiarity, frequency, or givenness in a certain context (Giora 1997). Salient meanings are always processed first, regardless of the context in which the respective expression is used. In contrast, nonsalient meanings of an expression are not coded in that expression’s lexical entry, but rather are the result of a pragmatic interpretation process leading to the speaker intended meaning, which either first accesses the salient meaning, discards it as not being intended and then goes on to interpret the intended meaning (traditional view following Grice), or works in parallel to the process that accesses the salient meaning of that expression. open: PHON: . . . SYN: . . . SEM: {OPEN-DOOR[10], OPEN-WINDOW[9], OPEN-FILE[8], OPEN-EYES[8], . . .} degree of salience Figure 2.16: Lexical entry according to Giora As mentioned above, RT and Giora’s approach have in common that they consider literal meaning and non-literal meaning as more similar than traditionally assumed. However, whereas under the relevance-theoretic approach both literal meaning and non-literal meaning are phenomena that occur at a pragmatically determined level, in Giora’s approach, some meanings traditionally characterised as literal are of lexical and some of pragmatic nature. The same holds for non-literal meaning. Furthermore, Giora characterises salient meanings and less-salient meanings as being coded in the lexicon. However, there seems to be no important reason why this should be the case. Rather, assuming that salience is a gradual property, one could in fact ask, by which criteria less-salient meanings are differentiated from non-salient meanings. To put it differently, what combination of factors is it that makes a specific meaning pass ‘the threshold’ from being a pragmatically derived reading (i.e. a non-salient meaning) of a certain expression to being lexicalised (i.e. a less-salient to highly salient meaning)? Assuming, in contrast, that what Giora characterises as salient and less-salient meanings are actually on the same informational level as non-salient meanings avoids such problematic questions. Moreover, the feature of salience, on a first blush, seems to go well together with what RT calls accessibility. That is, different readings for an expression are more or less accessible, depending on their frequency and recency of use. Moreover, as Carston puts it, concepts that an expression is used to communicate most frequently may be more or less ready-made retrievable and maybe considered, in that sense, as conventional. Also, considering that the salience of a meaning may be influenced by such factors as whether it has been recently Consequences for Lexical Meaning 59 Figure 2.17: Salient and non-salient meaning in Giora’s approach used/activated makes it even more desirable to assume that it is meanings at some non-lexical level that are characterised as being salient, less-salient or non-salient. Otherwise one would have to assume that certain meanings are sometimes part of the lexical entry of a particular expression and sometimes are not, depending on their givenness in a context or recency of use. Integrating these ideas into a view of a lexical item’s entry where this is not part of the concept encoded by that item, one might assume the following constellation. Part of the lexical entry of a particular lexical item is its semantic form, which is highly abstract and underspecified and which ‘points’ to what Sperber and Wilson (1995) call the encyclopaedic entry of a concept.41 Thus, the semantic form of an expression makes available, in the sense of activates, a range of information that may potentially be important in the particular context of utterance. The information activated is structured in the sense that due to differences in the accessibility of the various pieces of information, they will be more or less strongly activated. As already mentioned, accessibility is taken to be a function of frequency and recency of use/activation of a particular piece of information in processing. Moreover, as the results of studies show that tested the role of context in interpretation, it is likely that accessibility is also sensitive to contextual information available at the time of utterance. Thus, on the one hand, pieces of information of the encyclopaedic entry activated by the utterance of a particular lexical item may be inhibited in their strength of activation, if they are incongruent with the contextual information available. On the other hand, contextual information congruent with particular pieces of information may strenghten their level of activation. However, since we saw that what Giora calls salient meanings are activated regardless of whether or not they fit the context, it seems that incongruent contextual information is never ‘strong’ enough to actually prevent salient meanings from being initially activated. Thus, the impact contextual information may have on the degree of accessibility of the readings of an expression seems to be constrained. 41 Remember that RT takes the content of a concept – ignoring the lexical entry, which has already been identified as problematic – to consist of both an encyclopaedic and a logical entry. In assuming that it actually is the encyclopaedic entry alone that the semantic form of an expression ‘points’ to, I follow a suggestion made by Groefsema (2007), who, in my view, gives a range of convincing arguments to this effect. 60 Against the Standard Notions of Literal Meaning and Non-literal Meaning open: PHON: . . . SYN: . . . SEM: <OPEN> points to {OPEN-DOOR[10], OPEN-WINDOW[9], OPEN-FILE[8], OPEN-EYES[8], . . .} degree of accessibility Figure 2.18: A lexical entry adapted from Relevance Theory Underspecification and Conventionality One other issue I would like to discuss is the relation of lexical meaning and the property of conventionality in approaches that assume a high degree of semantic underspecification. Thus, according to Saussure (1931), linguistic signs consist of two parts, a form and a meaning, where the relation between those two parts rests on a convention. That is, there is no natural connection between the meaning and the form of an expression. But, what do we mean by ‘convention’ ? (Lewis 1969) offers a formalised definition of convention. A regularity R in the behaviour of members of a poluation P when they are agents in a recurrent situation S is a convention iff it is true that, and it is common knowledge in P that, in almost any instance of S among members of P: 1. almost everyone conforms to R 2. almost everyone expects almost everyone else to conform to R 3. alomst everyone has approximately the same preference regarding all possible combinations of actions 4. almost everyone prefers that any one more conforms to R, on condition that almost everyone conforms to R 5. almost everyone would prefer that any one more conform to R’, on condition that almost everyone conform to R’ where R’ is some possible regularity in the behaviour of members of P in S, such that almost none in almost any instance of S among members of P could conform both to R’ and to R.’ Thus, conventions seem to have to do with ‘customariness’ within a group. That means that a speaker associates a certain expression form with a certain meaning, because a sufficiently large enough group of speakers associates that expression form with that meaning. Under such a view of convention, it seems inappropriate to characterise the abstract underspecified semantic form of an expression as being conventional in the intended sense, since SEMs are not actual readings of an expression. As mentioned above, there are a number of psycholinguistic studies which show that particular types of non-literal meaning may be interpreted as fast as literal Consequences for Lexical Meaning 61 meanings. In particular, it has been shown that certain non-literal meanings of expressions get activated even in contexts where it is obvious that a literal interpretation is intended. Moreover, the results differ with respect to whether the non-literal meanings are salient, less-salient or non-salient. According to Giora’s definition of salience, conventionality of a reading plays an important role. Moreover, I have already argued above, that conventionality should not be viewed as a binary feature and, thus, cannot be used to differentiate between literal and non-literal meaning. Thus, the fact that under the assumption of underspecification of lexical meaning, conventionality in the intended sense cannot be taken to be a feature of lexical meaning, does not pose a problem. As the results of the empirical studies mentioned suggest, conventionality is a graded notion which applies not only to what traditionally is called literal meaning, but also to non-literal meaning, in some cases leading to a non-literal meaning actually being conventionally associated with some phonological form to a greater degree than a possible literal reading. Thus, approaches assuming highly underspecified semantic forms come very close to what Recanati (2004) calls meaning eliminativism (ME) and which he characterises as follows. ‘. . . according to ME, we do not need linguistic meanings even to serve as input to the construction process [. . .] the senses that are the word’s contributions to contents [of utterances] are constructed, but the construction can proceed without the help of conventional, context-independent word meanings.’ (Recanati 2004, p. 147, my emphasis) Although underspecification-approaches do assume some SEM-component in an expression’s lexical entry that SEM only serves as a ‘pointer’ to potential readings of the expression.42 It does not constitute anything like the conventional meaning of an expression, as assumed under the traditional view of lexical semantics. Moreover, Recanati (2004) says of ME that it computes ‘directly the contextual sense which an expression assumes on a particular occasion of use on the basis of the contextual sense which that expression had on previous occasions of use . . .’ (Recanati 2004, p. 147). If, however, the contextually determined reading of an expression is computed on the basis of the readings of earlier uses of the same expression, there has to be some sort of connection between the expression and those readings: this is the radically underspecified SEM component. Underspecification and Semantic Relations Traditionally, one of the goals of semantics was to describe and explain the meaning relations that were found to hold between individual expressions. However, it has been noted that this task is not at all easy to accomplish. Thus, under the traditional perspective of linguistic signs as consisting of a form and a meaning, first, individual lexical units (combinations of a distinctive sound shape and a distinctive meaning) 42 For an actual ME-approach, one that denies that words actually consist of both sound form and meaning but rather assumes that phonological form merely represents conceptual content (hence the name representational hypothesis), cp. Burton-Roberts (1994, 2000, 2007a), Burton-Roberts and Poole (2006), Chng (1999). 62 Against the Standard Notions of Literal Meaning and Non-literal Meaning which share the same phonological form, have to be classified as belonging to the same or different lexemes. Second, a particular lexeme which is characterised as being polysemous, thus, incorporating more than one lexical unit, may have several distinct meaning relations to several, distinct lexemes. What is generally problematic about the notion of semantic relation is that there is the danger that specifically linguistic knowledge is mixed up with general conceptual knowledge (cp. Fodor and Lepore 1998). Thus, the underlying assumption for the specification of meaning relations between expressions is that their intensions share some feature that allows their extensions to be related in a certain way. That is, the fact that the expressions human and woman are taken to be in a hyponymy relation is due to the observation that the denotation of woman forms a subset of the denotation of human. In order to explain this fact about the denotations of these expressions, it is assumed that their intensions, or meanings, are in a particular relationship as well, namely, the meaning of woman includes the meaning of human. However, the question is whether, when thinking about such relationships between denotations, it is admissible to link this relation to a relation between their linguistic meanings. I think it is not. Rather, the relationship holding between the denotations of human and woman is due to the relation between the concepts human and woman that we have and which, crucially, we have independently of the fact that we can use the phonological forms /ju:mn/ and /w=mn/, respectively, to express them. Thus, semantic relations are much more closely related to our general cognitive ability of categorisation than to linguistically relevant aspects of meaning – whatever those may be. Moreover, a speaker may be able to use a particular expression correctly without being aware of the relationships that hold between the concept that the expression he uses expresses and other concepts. That is, I might use the expression whale successfully to refer to an instance of the kind ‘whale’ without knowing that the denotation of the concept whale is a subcategory of the denotation of the concept mammal. Thus, one might question Borg (2004b)’s claim that a semantic theory should make clear the relations that hold between individual complex expressions, for example, which inferential relations possibly hold between them. In fact, one consequence of assuming underspecification of the linguistic meanings of expressions is that semantics is no longer able to capture the various meaning relations an expression may enter into. In that case, how could semantics capture the inferential relations holding between elements of different domains that may be denoted by one and the same expression beyond only very general relations. More specifically, meaning relations may be taken to hold between full readings of expressions, but not between their abstract, underspecified semantic forms. However, from the arguments given above, it seems that one should not expect of semantics in general that it has anything to say concerning the relations holding between the concepts expressions may be used to express. More Underspecification in the Lexicon It should be mentioned that the idea that the lexical information for some linguistic item does not include information on that item’s argument structure has not only been pursued for the semantic, but also for the syntactic component of such entries. Consequences for Lexical Meaning 63 Thus, Kratzer (1996) argues against the assumption that the argument that will occupy the subject position in a sentence is projected by the verb, rather, it comes into the structure by means of a functional head. Thus, for the sentence in (44), (Kratzer 1996, p. 121) gives the partial syntactic structure as in (2.19). (44) Mittie feeds the dog. VoiceP HH H H DP Voice’ Mittie HHH H VP Voice Agent H HH DP V’ P PP Figure 2.19: Partial syntactic structure for Mittie feeds the dog the dog V Thus, Mittie is base-generated in SpecVoiceP, whereas the head of VoiceP introfeed duces the external argument. Note that Kratzer uses the ordered argument notation for argument structure, which does not clearly separate syntactic from semantic information. However, translating the facts into Dölling’s notation, the syntactic component of the lexical entry of feed would thus only specify that it is a verb and that it takes one NP (internal) argument. As Dölling (2005) notes, Borer (2004) goes even further in claiming that one should also separate the internal argument from its verb. Ignoring issues concerning the adequacy of such an approach for the moment, what this would result in is a syntactic component for feed, which would no longer include information on the number and kind of arguments that item takes. Thus, there are approaches in syntax also that reduce the amount of information required to be given in some linguistic expression’s lexical entry. In fact, there are attempts at radically reducing the role the traditional lexicon plays in the linguistic system. Thus, Marantz (1997) wants to dispense with the lexicon as traditionally conceived altogether. He assumes three distinct lists which to draw from in producing an utterance. The first of these, which he calls ‘narrow lexicon’, is the list of units that syntax operates with. This list contains a language’s atomic roots, as Marantz calls them, as well as the atomic bundles of grammatical features. What is interesting about the roots is that they are neutral with respect √ to word category. That is, a root like destroy is neither a noun nor a verb; depending on the kind of syntactic structure which it is part of it may function as a √ verb or as a noun. Thus, merging a root like destroy with ‘D’ (i.e., a nominalising environment) places it in a nominal context (cf. Figure 2.20). In contrast, merging the root with a verbalising functional head (v-1 in 2.21) results in destroy functioning as a verb. True nominalisations, then, contain both a verbalising and a nominalising head (cf. 2.22). What is interesting about this approach in light of the present concerns is the following. If lexical items (or rather roots) actually are characterised by the fact that they are category neutral, the syntactic information one would expect to find for them might be similarly underspecified as is the case for semantic information. 64 Against the Standard Notions of Literal Meaning and Non-literal Meaning D D HH H √ H destroy H HH √ destroy the city Figure 2.20: e.g. the destruction of the city, the city’s destruction v-1 HH HH v-1 √ destroy H H H √ destroy the city Figure 2.21: e.g. (John)destroyed the city Thus, the lexicon may not provide any distinct information on category membership of roots. However, and Marantz (1997) explicitly notes this, roots would need to include some type of information which ‘tells’ the syntax in which potential structures √ the root may occur. Thus, Marantz (1997) notes that a root like grow differs √ √ from a root like destroy in that it does not imply an agent. Thus, grow √ belongs to a different class (or possibly has different features) than destroy. √ Whereas grow semantically denotes a change of state that is internally caused, √ √ destroy denotes a change of state that is not. More specifically, with destroy an external cause or agent is implied. However, assuming such a characterisation may not be unproblematic from a semantic point of view. Generally, it is unclear in which form this type of information enters into the lexical entries of the items in question. One plausible possibility is to assume that it is part of an expression’s semantic form. However, this is precisely what a minimalist approach wants to avoid. Another possibility is to assume some abstract ‘group’ feature for different classes of roots. Leaving the particulars to the side, it is still interesting to note the parallels between the minimalist approach and the syntactic/morphological approaches mentioned above, all of which argue for some underspecification of lexically provided information. Underspecification of Semantic Composition Similarly to Dölling (1997, 2005), Egg (1994, 2003, 2005) also assumes that lexically given semantic information for expressions is highly underspecified. However, in his approach, the result of semantic composition is not only underspecified due to the underspecified nature of the elements that it composes, but in addition, the resulting semantic form only consists of fragments and an indication of the dominance relations between them. Crucially, the fragments are not yet totally fixed with respect to the scope they take over one another. For an exemplification of Egg (2003, 2005)’s approach, consider again (32), repeated below. (32) a. Max began reading every novel. b. Max began every novel. Consequences for Lexical Meaning 65 D H HH H D v-1 HHH √ H v-1 destroy H H H Figure 2.22: e.g.√(John’s) destroydestroying the city the city Recall that whereas (32a) has two readings depending on the scope the universal quantifier takes, (32b) only has one reading, where the universal quantifier outscopes the verb to begin. The respective semantic forms of (32a) and (32b) in Egg’s approach resulting from semantic composition are given in figures (2.23) and (2.24). Figure 2.23: The semantic form of Max began to read every novel. Figure 2.24: The semantic form of Max began every novel. The squares indicate ‘holes’ in the semantic representation. These holes have to yet be specified. The dotted lines indicate dominance relations between the individual fragments. Thus, in (2.23), both the universal quantifier and the verb to begin dominate the verb to read, whereas the dominance relation between them is not yet specified. However, either specification is compatible with the relations already fixed in the semantic representation of (32a). This is different for (32b), where in its semantic representation (2.24), the dominance relation between the universal 66 Against the Standard Notions of Literal Meaning and Non-literal Meaning quantifier and the verb to begin is already fixed such that the former dominates and thus has scope over, the latter. I do not want to go into too much detail of Egg (2003, 2005)’s analysis here, but note that the difference in the semantic representations for the sentences (32a) and (32b) is due to differences in the syntax-semantics interface rules that determine how the respective sentence meanings are composed. Moreover, whereas in the case of the semantic representation of (32a), there is a so-called constructive solution of the underspecification in the semantic representation, there is no such solution in the case of (32b). A semantic representation has a constructive solution if the underspecified elements (‘holes’) may be resolved simply by integrating the material that is already present in the semantic representation itself. For example, in the semantic representation for (32a), the semantics of the verb to read can be inserted into the hole immediately dominating it. By functional application, the verb’s semantic type is converted from <e, <e, <s, t>>> to <s, t>, which makes it of the appropriate type for insertion either into the hole of the semantics of the verb to begin or that of the quantifier. Depending on into which of these holes the verb’s semantics is inserted, the resulting representation can then be inserted into the remaining hole, thus determining the scopal relations between the quantifier and the verb to begin.43 The problem with the semantic representation of (32b) is that a constructive solution is not possible, simply because the verb to begin as it occurs in that representation is not of the appropriate type for insertion into the hole immediately dominating it (it would have to be of type <e, <e, <s, t>>> but actually is of type <<e, <s, t>, <e, <s, t>>). Thus, this is the place where an appropriate reinterpretation operator needs to be inserted in order to derive a well-typed solution for the underspecified semantic representation. One possibility is the operator that provides the reading-interpretation of (32b) and which would have to look like in (45). The final semantic representation for (32b) given in (46) is obtained by applying the operator in (45) to the semantics of the verb to begin and only integrating the result of this operation into the hole originally dominating the verb. (45) λz.2(λyλxλe. read∗ (x, y)(e)(z)) (46) λe00 ∀y.novel0 (y) → begin0 max0 , λxλe0 .read∗ (x, y)(e0 ) (e00 ) Thus, Egg’s approach handles what is traditionally called complement coercion by actually augmenting the verb’s semantics instead of that of the verb complement. Generally, his approach can also handle aspectual coercion as well as cases of metonymy. In all these cases, an underspecified semantic representation is build up, followed by the attempt of a constructive solution of the underspecifications. If no such solution is available, operators are made use of that integrate conceptual information which allows for a solution of the semantic representation after all. As Egg (2003, 2005) himself notes, calling the phenomena his approach handles reinterpretation phenomena is somewhat misleading, since his analysis actually does not involve a revision of some previously build up fully specified semantic form. 43 Following Egg (2002) and Egg (2003), I ignore the topmost hole and its dominance relations here. Empirical Investigations of Aspects of Semantics 2.4 67 Empirical Investigations of Aspects of Semantics In this section, I want to review the results of empirical studies that investigated different aspects of the semantic component assumed to form part of the language faculty. I will concentate on empirical investigations concerning the nature of lexically stored semantic information, on the one hand (2.4.1) and the question of whether a distinct neurological correlate of the process of semantic composition may be identified, on the other (2.4.2). 2.4.1 Polysemy vs. Underspecification in the Lexicon As stated in section 2.3.1, one of the challenges for a theory of meaning in the lexicon is the treatment of the phenomenon of systematic polysemy. Recall that systematic polysemy is characterised by the following facts. The different readings of a particular noun relate to one another but at the same time are clearly delimitable. There is no underlying meaning consciously detectable from which the different readings could be said to be derived, thus, intuitively they all seem basic. Crucially, the different readings are not characteristic for only one particular expression, rather, they are found across sets of expressions, thus, they are not idiosyncratic. As mentioned before, the question with respect to the lexical entries of systematic polysemous nouns is, how much semantic information is given. From the different approaches to meaning in the lexicon described above, theoretically, there seem to be at least three possibilities. First, as traditionally assumed, all of the relevant readings of a systematic polysemous noun might be listed as possible literal meanings of that noun. This would amount to a sense enumerating lexicon, where polysemy is not treated different from homonymy. Second, one of the readings could be assumed to be the ‘core’, or basic, meaning, from which all other meanings are derived. At least as far as intuitions are concerned, this view does not seem to capture the facts, however, intuitions might be misleading. Third, none of the different readings might be lexically coded, rather, the semantic information in the lexical entry might be maximally underspecified, allowing for different specifications as the interpretation enfolds, taking into account the specific context of an utterance of the respective noun. The different views make varying predictions concerning processing. Thus, if the sense enumeration approach is correct, then one should not find any differences in the processing of homonymous vs. polysemous expressions, as they are represented similarly. If, however, the core meaning approach is correct, then one would expect differences in processing, depending on whether a particular expression has a homonym or whether it is polysemous. Making predictions for the underspecification approach is a bit more difficult, as it does not seem quite clear what to expect. However, as we will we see below, studies that tested whether homonymy and polysemy are similar in processing actually found that there are specific differences and these in turn are explainable if one assumes underspecification of lexical semantic information. Klepousniotou (2002) took up the question of whether homonymy and polysemy 68 Against the Standard Notions of Literal Meaning and Non-literal Meaning are processed similarly, thus, supporting a sense enumeration approach. In this study, four types of ambiguity were tested, namely, homonyms and three different types of polysemous expressions (i.e., expressions with a metaphoric reading, words with metonymic count/mass readings and words with metonymic producerfor-product reading). The task was a cross-modal sentence-priming lexical decision task. Thus, subjects heard sentences that primed a particular meaning of a polysemous word or the meaning of a homonym. At the offset of the priming sentence, subjects were presented a target word on a computer screen. Reaction times for deciding whether the target word was a real or non-word were measured. The results showed that subjects reacted faster in the lexical decision task when the target word was polysemous as compared to homonymous. Klepousniotou (2002) takes this as evidence that different processes are involved in recognising words exhibiting different types of ambiguity. Thus, the results do not support a sense enumeration approach. Rather, Klepousniotou (2002) interprets the results as supporting a core meaning approach. Generally, she argues that the results are best explainable by assuming a basic difference in the representation of homonymous words as compared to a polysemous word. Thus, words that are homonymous really are distinct words in that they are distinctly represented. In contrast, in case of polysemy, there is only one entry, which, according to Klepousniotou (2002), gives the core or basic meaning of the polysemous word. As a consequence, there is a difference in processing. Thus, in case of homonymy, the interpretation process selects one of the given mental representations or entries, which goes along with a competition effect between the possible entries and results in slower overall processing of homonyms. In contrast, in the case of polysemy, no selection is necessary, as there is only one entry to begin with.44 However, as Beretta et al. (2005) note, a task such as lexical decision does not necessarily reflect early stages of processing. Thus, the effects attained in a lexical decision task could also be due to post-lexical processing. The authors therefore decided on a MEG experiment in order to find out whether the differences in reaction time noted by Rodd et al. (2002) for the processing of homonymy vs. polysemy would be paralleled by differences in the MEG data recorded. The component they focused on is the M350, of which they argue that it has been shown to be an index of initial lexical activation, but – and this is crucial – not of later processing stages (see also Pylkkänen et al. 2006). The hypotheses investigated are the following. If the sense enumeration approach is correct and homonyms and polysemous words are processed similarly after all, one would expect the M350 data to be similar for both conditions. In that case the M350 data would be in contrast to the reaction time data, suggesting that the latter reflects post-lexical processing. If, however an approach is correct that assumes a single entry for polysemous words (i.e., either the core meaning or an underspecification approach), the M350 data should be differential for the two conditions and thus, parallel that of the reaction time studies. Beretta et al. (2005) used the stimuli and design of Rodd et al. (2002)’s experiment, carrying out both the behavioural (reaction time) test and a MEG test. 44 Note that although this study did not take into account the possibility of an underspecification approach, Klepousniotou (2002)’s arguments for a core meaning approach are compatible with an underspecification approach as well. Empirical Investigations of Aspects of Semantics 69 Thus, subjects were visually presented with test words of which they had to decide whether they were words of the language or not. Test words were either homonyms (two meanings) or non-homonyms (one meaning). Within these two categories a further distinction was made between meanings that consisted of only a few and meanings that comprised many senses.45 Beretta et al. (2005) replicated the behavioural results of Rodd et al. (2002), thus, reaction times to homonymous were longer than to polysemous words. Moreover, the more senses a word had, the faster it was recognised (cf. figure 2.25). Crucially, the M350 data recorded paralleled the reaction time data. Thus, polysemous words led to an earlier peak of the M350 component than homonymous words. Again, the more senses a word had, the earlier the peak of the M350 component (cf. figure 2.26). Mean response time (ms): factorial design and main effects Homonymy Polysemy Few senses Many senses Mean Single meaning 626 611 619 (non-homonyms) More than one meaning 648 622 635 (homonyms) Mean 637 617 Figure 2.25: Reaction times from Beretta et al. (2005) Mean M350 latency (ms): factorial design and main effects Homonymy Polysemy Few senses Many senses Mean Single meaning 345 328 336 (non-homonyms) More than one meaning 359 349 354 (homonyms) Mean 352 338 Figure 2.26: M350 latencies Beretta et al. (2005) The results, thus, support the hypothesis that polysemous words are represented in a single entry in the mental lexicon, since both the reaction time as well as the MEG data show that they are processed differently from homonymous words. Similar results were attained by Pylkkänen et al. (2006), who also used the M350 component to investigate the processing of homonymous and polysemous words. These authors reach the same conclusion as Beretta et al. (2005), favouring a single-entry approach to polysemy. Note, however, that neither of the two studies makes any claims as to what type of single entry approach (core meaning or underspecification) is better reflected by the data. In contrast, Frisson (2009) does argue explicitly for an underspecification approach. 45 For example, there are two homonymous words bank, one with a meaning that might be glossed as ‘river bank’, the other with a meaning glossed as ‘financial institution’. For the latter, different senses are possible like ‘building’ or ‘institution’. 70 Against the Standard Notions of Literal Meaning and Non-literal Meaning Frisson (2009) recapitulates the results of experiments investigating the processing of homonymy and polysemy, concentrating on two factors, namely, the influence of frequency of occurrence of a meaning/sense and context of utterance on the processing of homonymous vs. polysemous expressions. Whereas for the former these factors have an impact on their processing, for the latter, they do not. More specifically, taken out of context, encountering a homonymous phonological/graphical form will result in strong activation of the meaning that is more frequent and less activation of less frequent meanings. Thus, Frisson (2009) speaks of homonym processing as a ranked parallel process, where the ranking of the individual meanings is provided by their individual frequency. However, this ordering can be influenced by contextual factors. That is, if the context biases the less frequent meaning of a homonymous phonological/grapical form, this meaning becomes better accessible. However, context cannot suppress the more frequent meaning, that is, the more frequent, or dominant meaning, is accessed in any case. Moreover, processing a less frequent meaning involves an extra effort. As Frisson (2009) puts it: . . . when the following context indicates that the subordinate meaning was the intended one, a great deal of extra processing is observed, indicating that the wrong meaning had been selected; when the preceding context disambiguates towards the subordinate meaning, increased fixation times are observed on the target word, indicating competition between the two meanings . . .’ (Frisson 2009, p. 114). What is important for Frisson (2009)’s argument is that these effects show up very early during processing and on the homonymous form itself. Thus, it seems that for homonymous expressions, a specific interpretation is selected immediately on its encounter. In contrast, for polysemous expressions, such frequency or contextual effects were not perceived. Thus, it seems that polysemous expressions are not subject to a ranked parallel process, where the various senses are activated in parallel and where the strength of activation depends on the individual senses’ frequency. Rather, as Frisson and Pickering (1999) argue, the results for the processing of polysemous expressions seem to be best captured by a model that assumes underspecification of semantic meaning. Thus, if it is an underspecified semantic form that is activated during early processing, one would not expect that the frequency of the individual potential senses plays a role at this stage. Similarly, the fact that none of the potential senses could be established as the dominant reading (although senses do differ as to their frequency of use) is explainable by an underspecification approach, but not by a core meaning approach. Moreover, as Frisson (2009) points out and as already argued above, a further characteristic that makes an underspecification approach attractive is that it allows for a similar treatment of so-called literal and non-literal senses of polysemous words. It explains the fact that there may be no differences in the time taken to process a literal vs. non-literal sense of a polysemous word. Thus, Frisson and Pickering (2007) tested whether there are differences in the processing of the (literal) person sense and the (non-literal) familiar metonymic producer-for-product sense of names for famous persons such as Dickens, using sentences as in (47).46 46 See also Frisson and Pickering (1999) where no significant differences were found in processing Empirical Investigations of Aspects of Semantics 71 (47) a. I heard that she often met Dickens when she had the time. b. I heard that she often read Dickens when she had the time. They found no significant early differences in reading time between the two senses, although overall the metonymic senses were much lower in frequency. To summarise this section, it seems that the results obtained in studies investigating the processing of homonymous vs. polysemous expressions are best explained by assuming an underspecification approach for expressions with at least two related senses, regardless of whether these senses are literal or non-literal, as long as the latter are familiar.47 2.4.2 Empirical Evidence for Semantic vs. Pragmatic Processing Several studies have been conducted to investigate the processing of sentences whose semantic composition is assumed to involve coercion processes. One of the questions investigated was whether coerced sentences would be harder to process than control sentences that do not involve any coercion processes, thus reflecting the insertion of additional operators into the semantic form of the coerced sentences. More generally, there is the question of whether coercion is any different from plausability violations, which are pragmatically based. That is, from a theoretical perspective coercion clearly is a semantically based operation and as such one might expect coerced sentences to have different effects on processing than sentences whose interpretation involves pragmatically based plausibility violations. Another question of interest is whether there are any differences in the processing of complement coercion as compared to aspectual coercion. If no such differences are found, this could be taken to indicate that the two phenomena really are two different cases of the same type of underlying process. If, on the other hand, there are differences in how these phenomena are processed, this would indicate that they really are two different kinds of phenomena. Furthermore, considering that in Egg (2003, 2005)’s theoretical approach, for instance, metonymy and coercion are treated similarly, the question arises whether the two types of phenomena show similar processing effects in empirical studies. The prediction that coerced sentences are harder to process follows from the assumption that coercion processes, in contrast to more basic semantic composition operations such as functional application or predicate modification, are not syntactically triggered, but are purely semantic in nature. Thus Pylkkänen and McElree (2006, p. 35) claim that ‘[i]f coercion requires an extra-syntactic process that is not part of the default repertoire of interpretive rules, coerced sentences should be more costly to process than non-coerced controls’. This hypothesis has been confirmed by place-for-institution (i) and place-for-event metonymies (ii) as compared to the ‘literal’ senses of the respective expressions. (i) That blasphemous woman had to answer to the convent. (ii) A lot of Americans protested during Vietnam. 47 However, it seems that there is not much difficulty involved in extending the number of senses of an expression, if the extension is systematic (cf. Frisson and Pickering 2007). 72 Against the Standard Notions of Literal Meaning and Non-literal Meaning longer readings times (cf. McElree et al. 2001) as well as differences in eye tracking (cf. Pickering et al. 2005, Traxler et al. 2002) for complement-coerced compared to non-coerced control sentences. However, for aspectual coercion the results are not so clear, with some studies reporting effects in secondary lexical decision and self-paced stop-making sense tasks (Piñango et al. 1999, 2006, Todorova et al. 2000) and some reporting no reading time or eye movement effects (cf. Pickering et al. 2006). Neurolinguistic results are more consistent, with both types of coercion modulating the same brain activity (cf. Brennan and Pylkkänen 2008), where this, in turn, differs from that associated with the ERP component N400, traditionally taken to reflect so-called semantic integration difficulties (cf. Pylkkänen, Oliveri and Smart 2009). Thus, Pylkkänen et al. (2004) used magnetoencephalography to investigate potential effects of complement coercion on processing. Most importantly, the idea was to test the hypothesis that in the case of complement coercion, semantic composition is more difficult and costly and should lead to some measurable effect different from pragmatically based plausibility effects usually engendering an N400. The underlying assumption is that the difficulty in so-called semantic integration taken to be captured by the N400 ERP component is in fact a reflex of the difficulty of integrating the meaning of a sentence with the hearer’s knowledge of the world. In other words, what the N400 actually captures is not a difficulty in building up the semantic form of a sentence, but rather the hearer’s difficulty of interpreting it in accordance with his knowledge of how the world is. Thus, the anomaly of a sentence such as (48b) below is not due to some semantic mismatch, but rather to the fact that its meaning does not match the knowledge we have concerning possible relations between journalists and articles.48 (48) a. The journalist began the article after his coffee break. b. The journalist astonished the article after his coffee break. c. The journalist wrote the article after his coffee break. The authors found an anterior midline field (AMF) at 350 – 500ms after the onset of the critical noun (article in 48a) for all sentences, but with the largest amplitudes in coerced sentences (as, e.g., 48a). Importantly, anomalous sentences (e.g., 48b) and control sentences (e.g., 48c) showed no significant difference in AMF amplitude. Moreover, while the left hemisphere M350 field pattern, taken to be the neural source for the N400 ERP component and occuring at 300 – 400ms, showed the largest amplitudes for anomalous sentences, it was not modulated by the coerced stimuli. Comparable results were obtained by Brennan and Pylkkänen (2008) for cases of aspectual coercion, indicating that the latter phenomenon is indeed similar to complement coercion. The authors were concerned that earlier conflicting results concerning the processing of aspectual coercion might be due to the fact that the plausability of the used verbs as expressing a single event or rather a complex event consisting of a number of similar single events was not controlled for. For instance, ‘most knocking events involve multiple knocks . . . [i]n contrast, a burping event is 48 It should be noted that this argument only holds true for semantic approaches that are entirely type-driven and where restrictions on the type of semantic role required of the arguments do not influence the process of semantic composition. Empirical Investigations of Aspects of Semantics 73 much more likely to involve only a single burp.’ (Brennan and Pylkkänen 2008, p. 133). Thus, for their experiments, Brennan and Pylkkänen (2008) pre-tested the material with respect to the likelihood of the verb describing a single or repeated events. Verbs that were rated as strongly punctual were then combined with durative (cf. 49a) or punctual (cf. 49b) adverbs to create critical and control stimuli. (49) a. Throughout the day the student sneezed in the back of the classroom. b. After twenty minutes the student sneezed in the back of the classroom. In a first experiment, subjects’ self-paced reading times were measured as well as the accuracy of sensicality judgements concerning the stimulus sentences. The results indicate that punctual verbs are more difficult to process in the environment of durative modifiers, as subjects took longer to read them in the critical condition than in the control condition. In a second experiment, using MEG, the authors tested whether the AMF effect found for complement coercion would also be found for aspectual coercion. If the same effect were obtained, this would indicate that the two phenomena are in fact similar and thus involve comparable processes. An AMF effect was indeed found, confirming the hypothesis that aspectual coercion involves processes of meaning shift similar to those found in complement coercion. Pylkkänen, Martin, McElree and Smart (2009) showed that the AMF effect obtained by Pylkkänen et al. (2004) (and Brennan and Pylkkänen 2008, for that matter) was not due to the experimental task subjects had to carry out, which involved decision making. Thus, even when subjects were simply reading for comprehension without having to judge the stimuli, AMF effects were obtained for the coerced stimuli. Moreover, Pylkkänen, Oliveri and Smart (2009) showed that the effect is not restricted to coercion phenomena, that is, resolvable semantic mismatches, but is also obtained in other cases of clearly semantically based but unresolvable mismatches between elements in a sentence. More specifically, in this study, Pylkkänen, Oliveri and Smart (2009) investigated the effects of violations of the constraints on verbal un-prefixation, which are much more restrictive than those for adjectival unprefixation. The stimuli used involved semantic violations (as, e.g., 50b), world knowledge violations (as, e.g., 50c) and well-formed controls (as, e.g., 50a). To keep the syntax constant across conditions, all stimuli were presented in the passive with progressive aspect, to force a verbal reading of the participle (rather than an adjectival reading). (50) a. The wine was being uncorked. b. The wine was being unchilled. c. The thirst was being uncorked. While an AMF effect was obtained for the critical semantic violation stimuli, no such effect was found for the well-formed and pragmatically deviant controls. This indicates that the AMF effect does not only reflect successful semantic composition, but more generally attempts at such a composition. Thus, these results make the AMF an even more likely candidate for a neural correlate of the linguistic operation of semantic composition. 74 Against the Standard Notions of Literal Meaning and Non-literal Meaning Note, however, that caution is called for when interpreting the AMF effect as directly reflecting difficulties encountered during the process of semantic composition. Thus, Brennan and Pylkkänen (2008) take the fact that critical stimuli such as (49a) engender an AMF effect as evidence against assuming that (at least strongly punctual) verbs are semantically underspecified with respect to aspectual properties. In contrast, Pylkkänen and McElree (2006) had still argued that aspectual coercion in fact differs from complement coercion in that the latter, but not the former, involves both a semantic type-shift and a shift in ontological sort, whereas aspectual coercion only involves a sortal shift which, however, is not semantically based. In other words, aspectual properties were not taken to be semantically coded. In addition, in their argumentation for a basic differentation between complement coercion and aspectual coercion, Pylkkänen and McElree (2006) refer to a parallel difference between metonymy and complement coercion. Thus – as was shown by McElree and Frisson (2006) – whereas interpreting a proper name such as Dickens metonymically (as in 51a) is not more taxing than a literal ‘person’ interpretation (as in 51b), a coerced interpretation (as in 51c) is harder than both a literal and a metonymic interpretation. (51) a. The gentleman met Dickens. b. The gentleman read Dickens. c. The gentleman began Dickens. On the basis of such results, Pylkkänen and McElree (2006) make the assumption that different types of meaning shifts may involve different degrees of difficulty. In particular, sortal mismatches are assumed not to block initial semantic composition of sentence meaning, whereas ontological mismatches do: ‘. . . ontological mismatch, such as [that] found in complement coercion, blocks basic composition and hence immediately mandates the on-line deployment of a costly coercion process (Pylkkänen and McElree 2006, p. 55). However, the results of Brennan and Pylkkänen (2008) militate against the assumptions made about the nature of aspectual coercion in Pylkkänen and McElree (2006). That is, either sortal shifts in fact do block initial composition and thus lead to the on-line integration of a coercion operator, or the AMF effect observed both for cases of complement coercion and aspectual coercion actually is not a reflex of a more difficult composition of the semantic representation of the respective sentences as such. If the first alternative is true, one would expect to see AMF effects also for metonymy, which was considered more similar to aspectual coercion by Pylkkänen and McElree (2006) than to complement coercion. Unfortunately, as far as I can tell, to date, potential AMF effects of metonymy have not been investigated. Concerning the second alternative, recall that both in Dölling’s as well as in Egg’s theoretical approaches to coercion, the result of semantic composition of a sentence’s meaning is in fact a highly underspecified meaning representation which, more often than not, is in need of specification by additional material provided by the conceptual system. I find it conceivable that the AMF effect reflects the relative difficulty of integrating such material or possibly of overriding the insertion of default values for particular variables. In Egg’s approach in particular, the AMF might be taken to signal the unavailability of a constructive solution for an underspecified Why the Standard Notions? 75 semantic form and the search this triggers for material that might be inserted into that form to get a solution after all. 2.5 Why the Standard Notions? Before going on to pursue the hypothesis that, actually literal meaning and nonliteral meaning are not so different in kind as traditionally assumed, there is another question connected with the traditional view that I would like to look at. That is, one might wonder why exactly the traditional view came about in the first place and why, for many people and a long time, it seemed such a reasonable view to hold. More specifically, one might wonder why if literal meaning is context-dependent it still seems possible to interpret sentences presented out of context. With regard to simple expressions one might ask why if literal meaning is context-dependent it nevertheless is possible to give the lexical meaning for that expression when presented out of context. Ruhl (1989) uses the notion of stereotype to explain why, for a given example, we arrive at what we think of as the literal meaning of that sentence. He argues that even in situations where no explicit context is given, we never interpret sentences without some assumptions about a possible context of utterance (cp. Searle 1978). The problem is that the consideration of these assumptions is not consciously available to us. Our conscious minds lead us to think that we grasp the meaning of a given sentence without any context whatsoever, but actually, that is not the case. At least certain background assumptions (Nunberg 1978, Searle 1978) or assumptions about stereotypical uses of the sentence at hand are necessary to be able to assign it an interpretation. Thus, in cases of so-called ‘zero-contexts’, an expression gets assigned a default interpretation, which is based on the assumptions that we make in interpreting that expression. Generally, the (literal) meaning that we take an utterance to express is the result of unconscious cognitive processes, which are applied to the semantic form of that utterance. Such processes make use of different kinds of information, e.g. stereotypical uses of an expression, frequency of uses of an expression in a certain reading, background assumptions that are relevant for the current utterance, etc. Since these processes take place unconsciously, we assume of literal meaning that it is context-independent. Thus, the reason why we interpret the sentence ‘The cat is on the mat’, as describing a situation as depicted in figure 2.3 is that the indicated state of affairs is a stereotypical one. This state of affairs seems to be a kind of stereotypical interpretation for that sentence. However, stereotypical interpretation is not to be equated with literal semantic meaning of a sentence. Apparently, what is actually referred to by the former term, is a pragmatically modulated meaning of a sentence, which has gained a stereotypical status and of which our conscious mind does not ‘see’ that it is not ‘isolated meaning’. Note that in contrast to Searle, Ruhl does allow for intrasentential influences on the pragmatically modulated word meanings. Thus, in a situation without a concrete context of utterance, it is very likely that take gets the interpretation of ‘steal’ in a sentence like in (52) due to the fact that its external argument is the NP the thief and its internal argument the NP the jewels. 76 Against the Standard Notions of Literal Meaning and Non-literal Meaning (52) The thief took the jewels. Again, this is the interpretation of the sentence in (52) which one stereotypically arrives at, under the assumption that there is no given context for it. That is, ‘. . . with no contrary evidence, we tend to assume a particular set of circumstances. The reasons for this assuming are largely unconscious.’ (Ruhl 1989, p. 86) However, of course the sentence could be uttered in a situation far from stereotypical. Thus consider (53). (53) Peter hired a gunman and a thief to do the job. After they had delivered the document, he met them in headquarters to give them their promised payment. The thief took the jewels. But the gunman was not satisfied with his payment. In the case of (53), take would not receive the interpretation ‘steal’, despite the specific NPs in its argument positions, because the sentence would not receive a stereotypical interpretation. In a sense, the stereotypical interpretation of a sentence is its default interpretation; it will only be chosen in those cases where there is no contrary evidence to its applicability. This view seems to go back to Putnam (1975) who argued that in explaining to someone the meaning of a particular expression, we give either an ostensive definition or a description, where the latter ‘typically consists of one ore more markers together with a stereotype – a standardized description of features of the kind that are typical, or “normal”, or at any rate stereotypical’ (Putnam 1975, p. 147). The features that are central to the stereotype are such that in normal situations they are sufficient for recognising whether some thing belongs to a certain category or not. Putnam takes a stereotype of some X to be a conventional idea that we have of X with respect to what it looks like, or how it acts, etc. The important difference between the two views is that Ruhl explicitly takes stereotypes to be of pragmatic nature, whereas this is not at all clear in the case of Putnam. Thus, he argues that we would not say of a person that he knows the meaning of the word tiger if he did not know that the stereotypical tiger is striped. Interestingly, although Putnam does not give an explicit definition of what exactly a stereotype is either, he does mention a number of assumptions regarding stereotypes which actually are not warranted. Thus, saying of some feature that it is stereotypical for a certain category does not warrant the inference that most members of that category actually have that feature, or that all normal members of the category exhibit it. For this reason it makes sense to say of stereotypes that they are conventional ideas. Ruhl (1989) also relates stereotype to the concept prototype, arguing that the latter is a specific kind of stereotype. Thus, the prototype is the best examplar of a given category. The prototype is selected from the set of denotata an expression picks out, where the element that is selected is, in some way, seen as being superior to the others. This is not so for stereotypes which are taken to be typical denotata. So, to make use of Cruse’s (1990) example: arguably, the prototypical mushroom is well-formed, whereas the stereotypical mushroom is not. In relating prototypes with stereotypes, one should, however, remember that there is an important distinction to be made between the two concepts in terms of their general nature; whereas stereotypes are generally seen as sociolinguistic concepts (cf. Putnam’s view on the nature of meanings), prototype is applied in psycholinguistic approaches to the Why the Standard Notions? 77 mental organisation of meaning. Thus, more needs to be said on how stereotypical meaning comes about and exactly how it can be defined. Putting such questions to the side for the moment, however, from Ruhl’s point of view, then, prototypes as well as stereotypes are pragmatic, not semantic. An expression’s linguistic meaning, in contrast, is very abstract and probably does not involve any fullfledged reading at all. This view goes well together with the general criticism against prototype theory as being a theory of word meaning. Thus, prototype theory assumes that concepts have an internal structure which consists of a set of properties that are typical for members of the category picked out by the concept. The objects one encounters in the world are categorised by recognising the similarities between them and the concept that underlies a particular category in terms of typical properties. However, assuming that complex expressions are composed from simple expressions, it is questionable whether the linguistic meanings of expressions are structured in the way prototype theory assumes. Consider, for example, the fact that the meaning of a complex expression is not necessarily the sum of the prototypical meanings of its constituent parts. Thus, Fodor (1998, p. 102) argues that the meaning of pet fish cannot be claimed to be the result of the combination of the prototype of pet and the prototype of fish. However, if word meanings were prototypically structured, we would expect the meaning of the complex expression to be a combination of the meanings of its parts and thus as having as its prototype a combination of the prototypes of its parts. However, this is not necessarily the case. Thus, prototype theory as a theory of word meaning makes the wrong predictions. Thus, it is neither stereotypes nor prototypes that form the linguistic meaning of expressions. The question then is, what status these types of meaning have: are they pragmatically derived or possibly stored as part of our encyclopaedic knowledge? I will take up this point below. Turning back to the relation of (non-)literal meaning and context-(in-)dependence: there is the question of why it is that we generally assume of literal meaning that it is context-independent, whereas we usually think of non-literal meaning as being context-dependent. As already mentioned our intuition that literal meaning is context-independent is misleading, because even literal meaning depends on some kind of context. However, the type of context minimally required for the interpretation of literal meaning is rather unusual in that it does not need to be any specific context, but rather something that Ruhl calls stereotypical and which we apply unconsciously in the interpretation of expressions seemingly produced ‘out of context’. In contrast, our ordinary understanding of non-literal meaning is such that this is a very special kind of meaning, that deviates from some underlying basic meaning and that has a secondary character. Best examples of this type of meaning are usually such that depend on a very specific context of utterance for their correct interpretation (e.g. cases of irony). However, as already indicated and as we will see in more detail below, just as it is a mistake to think of literal meaning as strictly context-independent, it is inaccurate not to allow for gradience of contextdependency in the case of non-literal meaning. That is, not all types of non-literal meaning depend on context for their interpretation, to the same extent. So, the reason for assuming that literal meaning generally is context-independent whereas non-literal meaning is generally context-dependent is an overgeneralisation, due to 78 Against the Standard Notions of Literal Meaning and Non-literal Meaning our inability to consciously recognise cases of non-literal meaning if they are not as clearly non-literal as e.g. cases of irony. On the other hand, there are examples of an expression having several possible readings within different sentences when those sentences are produced seemingly ‘out of context’. (54) a. b. c. d. Peter opened his eyes. The girl opened the book. Jane opened the file. The boy opened the window. (55) a. The shop assistant took the jewels. b. The thief took the jewels. c. He took the job. As already said, it is questionable whether interpretation ‘out of any context’ ever actually occurs. Moreover, what is underestimated in such cases is the contribution made by the immediate sentential context to selecting a particular reading for an expression. 2.6 Summary In this chapter, I argued against the standard notions of literal meaning and nonliteral meaning. More specifically, I argued against viewing literal meaning as essentially context-independent and non-literal meaning as always non-conventional. Subsequently, I discussed the consequences of these assumptions for the nature of lexical meaning, by reviewing different approaches to meaning in the lexicon and the problems they face. More specifically, I argued for assuming underspecification of lexical meaning and a general distinction to be made between semantic and pragmatic processes. Empirical evidence supporting such a view was reviewed. Generally, I hope to have shown that the standard notions are not sufficiently precise enough to allow the terms literal meaning and non-literal meaning to be used as basic concepts in definitions of semantic vs. pragmatic aspects of meaning. I will summarise the arguments below. Thus, arguments were given against the view according to which sentence meaning is at the same time fully propositional, context-independent and thus literal. Rather, either one takes the defining characteristic of sentence meaning to be that it is context-independent, then it cannot (always) be fully propositional, or one takes the defining characteristic to be that of full propositionality, then sentence meaning cannot (always) be context-independent. Whichever alternative one chooses, the question then arises, what it is that we can apply the term literal meaning to: a sentence’s context-independent, subpropositional meaning or the proposition expressed by a sentence when uttered? There are a few arguments in favour of the latter characterisation, according to which literal meaning is not equivalent to context-independent meaning, but rather associated with utterances. The crucial argument is that what we intuitively take to be the literal meaning of some sentence, is actually determined with respect to the context in which that sentence gets uttered. This holds true even for sentences seemingly uttered ‘out-of-context’, where it Summary 79 is our background knowledge and assumptions about potential situations in which an utterance of the sentence at hand would be most appropriate that determine what we take to be that sentence’s literal meaning. Thus, actually, one should not speak of ‘a sentence’s literal meaning’ at all, as it is rather utterances that may be used literally or non-literally. Thus, literal meaning, it was argued, in fact is not context-independent, contrary to traditional assumptions. A related issue is the question of what types of information we take the so-called context in which an utterance is made to involve. Thus, in this chapter, several types of context have been vaguely differentiated: the concrete context of utterance, the context build up during forgoing discourse between participants, background assumptions and aspects of world knowledge and finally the immediate sentential context of the elements in an utterance. However, sofar it has not been made clear what kinds of information or knowledge these types of contexts involve. With respect to the assumption that literal meaning actually is context-dependent, the question is what kinds of information have to be available for a hearer to be able to interpret a particular utterance as having literal (or, for that matter, non-literal) meaning.49 Having argued for a view according to which literal meaning is actually contextdependent, it does not come as much of a surprise that in empirical studies the assumption that literal meaning always and necessarily has to be processed first could not be confirmed. On the contrary, the results of some experiments even show that a potential non-literal meaning of an expression may be activated as early as, if not earlier than, that expression’s supposed literal meaning. Thus, it seems that differences in the difficulty of interpreting some expression are not correlated to whether that expression was intended in a literal or non-literal meaning. Note that under the assumption that literal meaning and non-literal meaning both are contextdependent, one would actually expect results where there is no necessary difference in the interpretation of literal meaning as compared to non-literal meaning. In addition, some of the experimental results suggest that another feature traditionally used to differentiate literal meaning from non-literal meaning in fact does not differentiate the two types of meaning: conventionality. Thus, the experimental results generally call into question the standard notion of non-literal meaning as non-conventional, speaker-intended meaning that results from a reinterpretation process. If one assumes that conventionality is reflected in relative ease of interpretation, then one has to allow for quite a number of conventional non-literal meanings. Also, since there are cases in which the so-called non-literal meaning of an expression is activated as early as, if not earlier than, that expression’s literal meaning, the assumption that non-literal meaning is always the result of some reinterpretation process cannot be appropriate. Moreover, since there are cases in which non-literal meanings cannot be suppressed even though it is clear that they are not speaker intended, the assumption that non-literal meaning generally is a feature of speaker meaning cannot be sustained. In addition to empirical evidence, we saw that there are also theoretical reasons for assuming that conventionality is a gradual property. Thus, such an assumption makes the description of semantic change much more straightforward, since such changes do not take place abruptly, but rather over a period of time. The revised characterisations of literal meaning and non-literal meaning have 49 This point will be taken up throughout the discussion both in chapter 3 and chapter 4. 80 Against the Standard Notions of Literal Meaning and Non-literal Meaning consequences for the assumptions made about the nature of lexical meaning. Thus, the meaning(s) stored in the lexical entry of some expression can no longer be viewed as literal, since literal meaning is taken to be context-dependent, whereas lexical meaning is not. Several reasons were given that actually make this a desirable outcome. First, if one wants to assume that all an expression’s literal readings, i.e., those which seem to be basic, are coded in that expression’s lexical entry, one has to also necessarily assume that the lexicon as such is characterised by a high degree of polysemy and (if one tends to extremes) as almost indefinite in size. If one wants to assume of the lexicon that it is highly economic and minimal, this is an unsatisfactory result. Second, listing all possible literal meanings an expression may have in its lexical entry might be problematic, on the one hand due to the sheer number of potential literal readings (cf. readings for the verb to open), and on the other hand, since one misses generalisations (such as the similarities between possible literal readings for expressions like newspaper, university, book etc.) by treating certain meaning aspects as idionsyncratic when they are not. Third, if only literal meanings are coded in the lexicon and if the disambiguation process first has to go through all potential literal sentence meanings, rejecting them as not fitting the context, one would not expect to get evidence of an early activation of non-literal meaning, as this would be the result of some process that follows the unsuccessful and protracted disambiguation process. From the various approaches to meaning in the lexicon reviewed, it seems that the revised characterisations are best reflected by an approach that assumes a high degree of underspecification for lexical meaning. Generally, one important aspect that makes such an approach superior to the maximalist and intermediate approaches is the fact that the minimally specified semantic form allows for both literal as well as non-literal ‘completion’. Thus, such an approach, generally, is compatible with the empirical results according to which under particular circumstances non-literal meaning may be interpreted as fast as – if not faster than – the literal meaning of the utterance in question. Another important advantage of an underspecification approach is that it can explain the differences in processing of homonymous vs. polysemous expressions. More specifically, an underspecification approach can explain the pecularities found for the processing of polysemous expressions. Moreover, it is also compatible with the assumption of some type of default readings for an expression, since these, as noted above, seem to be part of a level distinct from the purely semantic, that is, context-independent level. Thus, literalness neither is a property of lexical, nor of sentence meaning. Rather, these latter two types of meaning are characterised by being very abstract and heavily underspecified, generally allowing for a literal or non-literal interpretation/usage. As a conclusion to this chapter, the following points can be made. Both the concepts of literal meaning and non-literal meaning in their traditional characterisation are not precisely defined. Moreover, the properties captured by the standard notions of literal meaning and non-literal meaning are problematic at the least. The evidence gathered so far suggests that literal meaning and non-literal meaning should not be characterised in terms of dichotomies, as they do not seem to be so very different from one another as traditionally assumed. Still the question remains: what is literal meaning and what is non-literal meaning? Put differently, how are Summary 81 the processes characterised that lead to what we intuitively call literal meaning or non-literal meaning? Recall the fact mentioned in chapter 1 that the term literal meaning traditionally has not only been employed for the level of sentence meaning or lexical meaning, but also to describe a particular type of context-dependent meaning, namely the level of what is said. Thus, in the next chapter, we will take a closer look at the characterisation of this particular level of meaning and where literal (and other types of) meaning aspects might be found at it. 82 Against the Standard Notions of Literal Meaning and Non-literal Meaning Chapter 3 Utterance Meaning and the Literal/Non-literal Distinction In chapter 2, I argued for a characterisation of the terms sentence meaning and literal meaning along the following lines. Whereas sentence meaning is compositional, context-independent and (more often than not) sub-propositional, literal meaning is a particular type of meaning a sentence (or complex expressions generally) may have when used in a context. Thus, literal meaning is a term applicable at a level of meaning, at least partly resulting from context-sensitive processes. The question is which factors have to hold for the thus resulting meaning to be literal. Before this question can be answered, however, the more basic question needs to be addressed of how such a context-sensitive level of meaning generally is characterised, that is, which processes are characteristic for this level of meaning and how it is distinguished from other levels of meaning. Thus, in this chapter, I first want to review two particular approaches to the differentiation of levels of meaning in interpretation, namely those of Grice and Bierwisch (section 3.1). Apart from a level of context-independent meaning, both approaches actually differentiate two levels of meaning that can be characterised as being context-sensitive: what is said (Grice) or utterance meaning (Bierwisch) on the one hand, and what is meant (Grice) or communicative sense (Bierwisch) on the other. For the purposes of the present chapter, we will especially concentrate on the characterisations of the level of what is said or utterance meaning in the two approaches. However, since assumptions about the relation of this level of meaning to the level of what is meant/communicative sense will be important for the discussion in chapter 4, we will look at those as well. The bigger part of this chapter is dedicated to the discussion of various alternative approaches to the characterisation of the level of what is said/ utterance meaning and the particular processes that are assumed to contribute to it. As we will see, the discussion mainly centers around the issue of whether the processes contributing to what is said/utterance meaning have to be linguistically mandated and whether they should be taken to be independent of speaker intentions (section 3.2.1 – 3.2.3). In a sense, the issue here is not only the nature of the processes used in the interpretation of utterances, but also the nature of the context against which this interpretation takes place. A related issue is the question of whether it is possible and useful to differentiate another 84 Utterance Meaning and the Literal/Non-literal Distinction context-dependent level of meaning apart from what is said/utterance meaning. In other words, can ‘saying’ be clearly differentiated from ‘meaning’ and if so, by which criteria (section 3.2.4). This point, then, leads over to chapter 4. 3.1 Levels of Meaning In this section, I want to take a closer look at two approaches to the distinction of levels of meaning, namely those proposed by Grice and Bierwisch, respectively. Although Grice’s distinction between what is said and what is meant is probably best known, this is not to say that Grice generally only differentiated between two levels of meaning. In fact, in trying to specify what exactly his notion of ‘saying’ was, he considered a fourfold distinction of levels or types of meaning. Similarly, Bierwisch also differentiates more than just two levels of meaning. Moreover, he and Grice make a number of similar assumptions. However, as we will see, their approaches also differ in some crucial aspects. 3.1.1 Grice’s Four Types of Meaning Grice (1975) distinguishes between two general levels of meaning, namely what is said and what is meant. What is said is the truth-conditional meaning of complex expressions as generated by semantic composition plus such information as is the result of a number of specific processes, namely resolving reference, fixing indexical expressions and disambiguation. What is meant is the meaning the speaker intended to communicate in making a particular utterance. As noted already, Grice does not call the processes that lead to what is said pragmatic, rather, he reserves this latter term for such processes as lead to what is meant and are based on the co-operative principle and the maxims of conversation. In his view, pragmatic processes operate on what is said and deliver meaning aspects that are implied rather than ‘said’ and that together with what is said result in what is meant by a speaker of a particular utterance. Thus, what is said is the proposition explicitly expressed by an utterance. It forms the basis for drawing further inferences (e.g. conversational implicatures), that is, for determining the implicit meaning content of an utterance in a particular context. What is meant, on the other hand, comprises both explicit as well as implicit meaning aspects. Moreover, in contrast to the level of what is said, which is taken to be independent of speaker intentions, for the determination of what is meant speaker intentions are essential. Note that although speaker intentions do not play a role for determining what is said by an utterance, Grice still assumes that what is said by an utterance forms part of what is meant by a speaker. Thus, on a first blush, it seems as if Grice only identified two levels of meaning as playing an important role in the overall interpretation process: what is said and what is meant. However, as Grice himself points out, he uses saying in a ‘favoured’, somewhat ‘artificial’ sense (Grice 1989, p. 118). One of the aims of his investigations is to make it clear how exactly this sense (and the derived notion of what is said) is to be understood. Thus, Grice (1969, 1989) actually contemplates four types or levels of meaning: timeless meaning of utterance-types, applied timeless meaning of utterance-types, occasion meaning of utterance-types and utterer’s occasion meaning. Levels of Meaning 85 Semantics lexical meaning semantic composition sub-propositional sentence meaning reference resolution fixing indexicals disambiguation ? fully propositional Pragmatics is part of what is said + narrow context basis for further pragmatic inferences conversational implicature speech acts + broad context ... what is meant Figure 3.1: Grice’s distinction of what is said and what is meant more detailed To illustrate the differences between these types of meaning, Grice (1969, 1989) uses the following example sentence. (56) If I shall then be helping the grass to grow, I shall have no time for reading. The timeless meaning of utterance-types1 (x [utterance-type] means ‘. . . ’) can be stated both for complete as well as incomplete utterance-types, that is, both for sentences as well as phrases or words. Thus, in the case of (56), there are (at least) two possible specifications of the timeless meaning of a (complete) utterance-type, namely: ‘If I shall then be assisting the kind of thing of which lawns are composed to mature, I shall have no time for reading’ and ‘If I shall then be assisting the marijuana to mature, I shall have no time for reading’. Similarly, for the expression grass, the timeless (incomplete) utterance-type meaning can be specified: one of its meanings being something like ‘lawn-material’, the other ‘marijuana’. As the example in (56) shows, a complete utterance-type x may have more than one timeless meaning. However, one would assume that (at least in most circumstances) for a particular utterance of x, x only carries one of those possible meanings. Thus, the applied timeless meaning of utterance-types (x [utterance-type] meant here ‘. . . ’) specifies what x means on this particular occasion; in the case of (56) this being: ‘If I shall then be assisting the kind of thing of which lawns are composed to mature, I shall have no time for reading’ and in the case of grass ‘lawn-material’. The occasion meaning of an utterance-type (U meant by x [utterance-type] ‘. . . ’) is supposed to capture what a particular utterer (U) meant by the words of (56); for example: (i) ‘If I shall then be dead, I shall not know what is going on in the world’ and possibly (ii) ‘One advantage of being dead will be that I shall be protected from the horrors of the world’2 . If a particular U meant by (56) (i), he would certainly 1 Note that by utterance-type Grice not only referred to sentences, phrases or words; rather, he intended his explications to apply to non-linguistic utterance-types as well. However, in what follows, I will only mention specifically linguistic utterance-types. 2 Note, however, that in the original typescript of the William James Lectures, Grice writes of (ii) that one would maybe prefer that the utterer implied it, which means that it actually belongs to the utterer’s occasion meaning. 86 Utterance Meaning and the Literal/Non-literal Distinction also be taken to have meant by the words: ‘I shall then be helping the grass to grow’, ‘I shall then be dead’. However, what is important to note here is that since ‘helping the grass to grow’ is not a recognised idiom, the specifications given for the occasion meaning, or of what U meant by (56), are neither admissible specifications of the timeless meaning, nor the applied timeless meaning of (56). As Grice puts it: ‘The words “I shall be helping the grass to grow” neither mean nor mean here “I shall be dead.” ’ (Grice 1989, p. 90, emphasis in the original). Note that here, although Grice does not use the terms, there is a differentiation between what an utterance-type literally means, contrasting to what a speaker might mean by it. The utterer’s occasion meaning (U meant by uttering x that . . . ) captures the fact that there is a difference in that U meant by (56) (i), whereas in uttering (56) U meant that if he would then be dead he would not know what was going on in the world. Additionally, U, in uttering (56) meant (or conversationally implicated) that one advantage of being dead would be that he would be protected from the horrors of the world. As Wharton (2002) notes, the difference between the occasion meaning of an utterance-type and utterer’s occasion meaning is very subtle and it is not entirely clear what exactly Grice wanted to capture by this notion. Both types of meaning apparently involve speaker intentions (‘U meant by x’ vs. ‘U in uttering x meant that’). The only difference between them seems to be that the latter additionally comprises aspects of meaning that are contributed by conversational implicatures. Thus, it seems that for Grice the applied timeless meaning of utterance-types corresponds to what is said.3 It is the level at which disambiguation, the resolving of indexical expressions and reference assignment has taken place. What is meant, then, corresponds to the utterer’s occasion meaning, as it includes such meaning aspects as conversational implicature. Moreover, Grice also recognises a type of meaning that is totally independent of any context of utterance, namely the timeless meaning of utterance-types. From his characterisation of this latter type or level of meaning, it can be assumed that he took this to be the level at which literal meaning – in the traditional sense of conventional meaning – is to be found. In addition, Grice distinguishes a further type of meaning, namely occasion meaning of utterance-type. However, it is not quite obvious in which respect this is supposed to differ from utterer’s occasion meaning. As already hinted at, a problematic issue with respect to the relation of what is said to what is meant is that Grice actually assumes that the former constitutes a part of the latter. However, there seems to be an important distinction between saying something and meaning something in that a speaker in uttering a certain linguistic string can be saying one thing, but might mean more than or something different from what is included in the meanings of the linguistic expressions she uses in her utterance. Thus, there are cases, as in example (56) above, where it seems that a speaker means something quite different from what she literally says. This is especially obvious in instances of irony such as in (57), where, very crudely, what can be assumed of the speaker is that she meant something like ‘the opposite’ of what she said. (57) (ironically) It’s a great day for a picnic. 3 See, however, Wharton (2002), who argues that from the way Grice differentiated the four types of meaning, he might have had another notion of saying (and thus, what is said) in mind. Levels of Meaning Semantics 87 (simple) timeless meaning semantic composition (complex) timeless meaning reference resolution fixing indexicals disambiguation ? applied timeless meaning + narrow context basis for further pragmatic inferences conversational implicature speech acts + broad context ... Pragmatics utterer's occasion meaning Figure 3.2: Grice’s types of meaning and their relation to what is said/what is meant On Grice’s analysis, this leads to the problem that since the speaker of (57) apparently did not mean what her words mean, she has not actually said anything. Similar problems arise in case of metaphor. Thus, the speaker of (58) cannot literally be taken to mean what his utterance expresses. But neither is it ‘the opposite’ of what his utterance expresses that is what is intended here. In any case, what the speaker means, again, is not what her words mean, so once again she has not said anything. (58) You are the cream in my coffee. Examples such as in (57) and (58) force Grice to differentiate between saying and making as if to say. Thus, the speaker in uttering (57) does not ‘say’ the proposition expressed by her utterance. (Carston 2002c) points out that the move from ‘saying’ to ‘making as if to say’ is highly problematic. To give only one of the problems she mentions: ‘The Gricean account of the tropes depends on the flouting of the first maxim of quality: do not say what you believe to be false. However, in these cases nothing has been said so, trivially, the maxim is not violated and the account on which an implicature is required in order to preserve the supermaxim of quality (try to make your contribution one that is true) cannot get off the ground . . . ’ (Ibid., p. 115; emphasis in the original). Similarly, Bach (1994b) wonders why Grice analysed irony and metaphor in terms of conversational implicature at all, precisely because in their case he speaks of ‘making as if to say’. Moreover, Bach (1994b, p. 272) argues that ‘[i]ntuitively, one thinks of implicating as stating or meaning one thing . . . and meaning something else as well, not as meaning something else instead’, as is the case in both metaphor and irony. Note also that Grice’s treatment of metaphor and irony suggests that these two phenomena are similar in terms of the effort involved in their interpretation. In both cases, the assumption is that the literal meaning of the sentences used in making 88 Utterance Meaning and the Literal/Non-literal Distinction a metaphorical or ironic utterance, is rejected as not fitting in the context and the respective conversational implicatures are drawn instead. However, as suggested by the results of empirical studies, it seems that the interpretation of metaphor and irony actually differs in terms of cognitive effort (e.g. Happé 1993, Colston and Gibbs 2002, Eviatar and Just 2006). Considering the problems arising from an analysis of metaphor and irony in terms of conversational implicature, these two types of non-literal meaning should be – and as we will see, actually have been – analysed differently (see sections 4.1.1 and 4.1.2). Another aspect that might be viewed as being problematic has to do with the nature of conversational implicatures. Here, Grice distinguished two basic types. Thus, whereas particularised conversational implicatures (PCI) depend on the specific context in which an utterance is made, generalised conversational implicatures (GCI) seem to be drawn in any kind of context, unless the latter is such that it prevents it.4 Thus, one interesting aspect of GCIs is the fact that they are independent of the speaker’s intention in making an utterance (cp. Burton-Roberts 2006). In other words, if the context of some utterance is such that it allows a GCI to be drawn, the speaker of that utterance actually would have to make it explicit if he did not want the hearer to draw this inference. This is an interesting fact to note since Grice identified conversational implicatures in general as belonging to the level of utterer’s occasion meaning, which is characterised as the level of meaning at which speaker intentions come into play. It might be for this reason that Neo- and Post-Gricean theories of interpretation handled this phenomenon differently, both from Grice as well as from one another (see section 4.1.3). 3.1.2 Bierwisch’s Three Levels of Meaning Bierwisch (1979, 1983) differentiates between three levels of meaning: expression meaning, utterance meaning and communicative sense. The expression meaning SEM (E) is the linguistically determined meaning of an expression E.5 For a complex expression, its expression meaning is compositionally build up from the meanings of the simple expressions contained in it and their syntactic combination. That is, meaning on the level of expressions is the context-independent meaning of simple and complex expressions. Thus, similarly to Grice, Bierwisch assumes a level of meaning that is fully linguistically determined and therefore context free. The utterance meaning M (t) is the meaning an utterance token t of an expression E has when it is used in a context C. The utterance meaning M (t) can be equivalent to the utterance token’s literal meaning, but it does not necessarily have to be. Thus, contrary to Grice and similarly to Searle, Bierwisch assumes that what is called the literal meaning LM (t) of an utterance token t of a simple or complex expression E is not identical to the linguistically determined meaning SEM (E) of 4 Thus, the GCI (ii) of the utterance in (i) is only drawn, if the hearer does not already know that (ii) is not the fact. (i) Some of the students went to the party. (ii) Not all of the students went to the party. 5 Note that in different articles, Bierwisch uses different abbreviation terminology. For the sake of consistency, I shall use one set of terminology throughout. Levels of Meaning 89 that expression. Rather, an utterance token of an expression can only have a literal (or, for that matter, non-literal) meaning in a context. Thus, literal meaning LM (t) is a special case of utterance meaning M (t). As already mentioned, a consequence of this view is the assumption that the lexical semantic representations of simple expressions do not encode what we take to be their literal meaning, rather, their semantic representations are maximally underspecified. Thus, in contrast to Grice, Bierwisch assumes that the level of expression meaning is characterised by a high degree of underspecification. The third level of meaning Bierwisch differentiates is that of the communicative sense CS (t) of an utterance token t of an expression E.6 The communicative sense captures what a speaker meant by the utterance of an expression. As such, it includes implied meaning aspects such as conversational implicature. Thus, this third level of meaning can be equated with Grice’s level of utterer’s occasion meaning. Moreover, similarly to Grice, Bierwisch reserves the term pragmatic for processes that lead from the level of utterance meaning to that of communicative sense. Semantics (simple) expression meaning semantic composition (complex) expression meaning reference resolution fixing indexicals disambiguation Conceptual System ... utterance meaning Pragmatics basis for further pragmatic inferences conversational implicature speech acts ... communicative sense Figure 3.3: Bierwisch’s three levels of meaning Bierwisch (1979) illustrates the three levels of meaning by way of the example sentence in (59). (59) Das habe ich nur mit der linken Hand gemacht. ‘I made this only using my left hand.’ A speaker uttering a token of this sentence, might mean with it one of the following, depending on the context of utterance: 6 With respect to the discussion so far, the term Two-Level-Semantics which came to be applied to Bierwisch’s particular approach seems rather misleading since he actually differentiates three levels of meaning. (see Taylor 1994, 1995). However, what that term does reflect is the fact that Bierwisch differentiated two types of cognitive structure: semantic structure (SS) and conceptual structure (CS), more on which see section 2.3.2. 90 Utterance Meaning and the Literal/Non-literal Distinction (60) a. Ich bin stolz, dass ich eine so komplizierte Handarbeit sogar mit der linken Hand anfertigen kann. ‘I am proud that I can produce such a complicated piece of handicraft, even with my left hand.’ b. Ich entschuldige mich für die Fehler der Handarbeit, die daher rühren, dass ich sie mit der linken Hand anfertigen musste. ‘I apologise for the defects of the piece of handicraft, which result from the fact that I was obliged to produce it with my left hand.’ c. Von diesem Artikel halte ich nicht viel, denn ich habe auf seine Anfertigung nicht viel Mühe verwendet. ‘I don’t think much of this piece of handicraft, because I didn’t pay much attention to its production.’ What is given in (60) are three of the possible communicative senses CS (t) of the respective utterance tokens t of (59). Bierwisch claims that for the production of the CS (t) s in (60a) and (60b), (59) is used in its literal meaning, whereas for (60c) it is not. A rough paraphrase of the literal meaning of (59) in the context in which it is used to communicate (60a) and (60b) is given in (61a); the non-literal meaning with which (59) is used to communicate the CS (t) (60c) is given in (61b), in both of which X represents the result of the production. (61) a. Ich habe X nur mit der linken Hand gemacht. ‘I produced X only with my left hand.’ b. Ich habe X nur nebenbei gemacht. ‘I produced X without paying much attention to it.’ Obviously, the meanings given in (61a) and (61b) are two possible utterance meanings M (t) for utterance tokens t of the sentence in (59) in a context C. As already mentioned, the utterance meaning of an utterance exemplar t can be equivalent to its literal meaning with respect to C, as in (61a), but does not have to be, as in (61b). Thus, (61a) is the literal meaning LM (t) of an utterance exemplar t of the sentence in (59). t in turn is an utterance exemplar of an expression E, which has a linguistically determined meaning SEM (E). As mentioned above, Bierwisch assumes of expression meaning that it is characterised by underspecification. One advantage of such an approach is that it allows a non-lexical treatment of systematic polysemy as well as of the non-systematic primary meaning variation of verbs. Thus, Bierwisch specifically argues for two processes he calls conceptual shift and conceptual differentiation. Whereas the former leads to the different readings such nouns as newspaper in (13) can have, the latter accounts for the different readings of verbs like open in (20). (13) a. The newspaper was on the table. b. The newspaper was censored, yesterday. c. The newspaper was founded three years ago. Levels of Meaning (20) a. b. c. d. e. John John John John John opened opened opened opened opened 91 the the the the the door. exhibition. file. parcel. store. To repeat, the SEM of a lexical item like newspaper determines a family of concepts as possible readings for it. As the SEM is underspecified, there is no primary meaning variant stated in the lexical entry of newspaper. Depending on the context in which newspaper is uttered, an interpretation function chooses one member of the family of concepts as the reading in that context. In a sense then, the interpretation function shifts between different concepts, according to the relevant context. Thus, Bierwisch calls this type of interpretation process conceptual shift. In contrast, conceptual differentiation is a process in which an unspecified SEM variable is given a specific value. The variable is bound by an existential quantifier, which determines a specific variation possibility by inserting an appropriate entity from the interpretation context. Each specification of the variable within the context results in a possible differentiation of the meaning of the respective word. In contrast to expression meaning, the meaning of an utterance can be examined as to whether it is true or false. However, specific kinds of meaning transfer might be necessary to arrive at the utterance meaning.7 Thus, utterances such as in (15a) and (18) can only be assigned a truth value after a metonymic shift of the meaning of the italicised expressions has taken place. (15a) The newspaper called. (18) a. The ham sandwich in the corner wants some more coffee. b. Plato is on the top shelf next to Russell. c. John got a dent in his left fender. Thus, Bierwisch assumes that an expression gets assigned its literal or, if applicable, its non-literal meaning only at the level of utterance meaning. Under this view, systematic polysemy, the differentiation of meaning variants of verbs as well as metonymic meaning shift are all phenomena occurring at the level of utterance meaning. However, the last phenomenon is treated differently from the first two in that a metonymic meaning shift presupposes an already determined meaning that can be shifted. Note that Bierwisch recognises a further type of process, namely selection, which operates in cases where in determining the meaning of an utterance both processes of conceptual shift and conceptual differentiation apply but produce more than one possible reading for the utterance. (62) John left the university. Selection will consider the broader context of utterance in order to disambiguate whether (62) is interpreted as meaning ‘John moved his position from inside the 7 Cp. Nunberg (1995, 2002) 92 Utterance Meaning and the Literal/Non-literal Distinction university building to outside of it’ or ‘John severed his affiliation with the universityas-institution’. The nature of this process raises the question of how much and what kind of contextual information is available to the processes of conceptual shift and conceptual differentiation in the first place. Apparently, the kind of contextual information that these processes can make use of is restricted, as otherwise the process of selection would be superfluous. Incidentally, Bierwisch assumes that at the level of utterance meaning, speaker intentions do not yet play a role. They are necessary only when it comes to determining the communicative sense of an utterance. This reflects his very traditional view of pragmatics, as being concerned only with phenomena such as speech acts and conversational implicature, which he takes to hold at the level of communicative sense. Thus, the processes which lead from the level of expression meaning to that of utterance meaning Bierwisch calls conceptual, not pragmatic. Note also that although Bierwisch allows for such non-literal meaning aspects as metonymy to figure at the level of utterance meaning, it is not quite clear whether he would also assume metaphors to be already determined at this level of meaning. In contrast to Grice, Bierwisch does not assume that utterance meaning necessarily forms part of the communicative sense. However, although in principle utterance meaning and the communicative sense of an utterance may be independent of each other, in the normal course of communication, an utterance’s meaning does serve as the basis for a recovery of what the speaker meant. Thus, in example (59) above, Bierwisch assumes that the possible communicative senses in (60a) and (60b) are based on the utterance meaning in (61a), whereas the possible communicative sense (60c) is based on the utterance meaning as given in (61b). The question here, once again, is whether deriving the utterance meaning in (61b) includes the derivation of the literal utterance meaning in (61a) as a necessary intermediate step. From what Bierwisch says about literal and non-literal utterance meanings, this does not seem to have to be the case. Moreover, it should be noted that since there is no direct relation between the expressions used to convey an intended meaning and the intended meaning as such, it is of course possible that a speaker could have conveyed his intended meaning by the use of other expressions. That is, it has to be made clear what it means for the CS of an utterance to be based on that utterance’s M. 3.1.3 Summary In this section, two different views on the differentiation of levels of meaning were introduced. Grice differentiated between four levels of meaning, timeless meaning of utterance-type, applied timeless meaning of utterance type, occasion meaning of utterance-type and utterer’s occasion meaning.8 On Grice’s differentiation, literal meaning is determined by the language system, that is, it is already determined at the level of timeless meaning of utterance-type and therefore totally independent of any context. At the level of applied timeless meaning of utterance-type, contextual factors of the utterance of an expression are used, on the one hand, to disambiguate 8 However, as mentioned above, what Grice intended the notion of occasion meaning of utterance-type to capture as different from utterer’s occasion meaning, is far from clear. Therefore, in my summary, I will only refer to the notion of utterer’s occasion meaning. Levels of Meaning 93 between different possible literal meanings and on the other hand, to fix indexicals and assign reference. To get to the level of utterer’s occasion meaning, the speaker’s intentions in making a specific utterance are considered, leading to what he meant in uttering it. Furthermore, based on a number of conversational maxims, at this level, implicit aspects of the intended meaning of an utterance are inferred. Thus, the level of applied timeless meaning of utterance-type corresponds to Grice’s what is said, which he takes to be the basis for drawing further pragmatic inferences, leading to what is meant, which, in turn, can be assumed to correspond to Grice’s utterer’s occasion meaning. Similarly to Grice, Bierwisch also assumes more than two levels of meaning, namely the levels of expression meaning, utterance meaning and communicative sense. Whereas expression meaning is that level of meaning, which is solely determined by the language system, it is not the level at which the literal meaning of an expression is determined. That is, Bierwisch agrees with Grice in assuming a specifically context-free, linguistically determined level of meaning, however, he also agrees with Searle in assuming that the determination of the literal meaning of an expression is not free of extra-linguistically determined factors. Rather, literal meaning is determined at the level of utterance meaning, where contextual factors are considered. As we saw, this view allows a non-lexical treatment of systematic polysemy and the (non-systematic) meaning variation of verbs, which is not only preferable if one wants to assume an economical lexicon, but which also seems to be more in line with experimental results concerning lexical access of the respective expressions. Furthermore, Bierwisch differs from Grice (and from Searle, for that matter) in assuming that just like the literal meaning of an expression, its possible non-literal meanings (at least such as idiom and metonymy) can be determined at the level of utterance meaning as well, due to the fact that these are independent of the speaker’s intention behind the utterance of that expression (cf. the difference between the utterance meaning (61b) of (59) and the related communicative sense (60c)). Although more will have to be said concerning this issue, generally speaking, such a viewpoint seems reasonable as it can be used to explain experimental results which show that non-literal meaning is not harder to process than literal meaning. Thus, Bierwisch’s characterisation of the level of utterance meaning and its relation to the other meaning levels identified seems to best capture the data and empirical results reviewed so far. Both Grice and Bierwisch reserve the term pragmatic for the processes that contribute further inferences based on what is said/utterance meaning and leading to what is meant/the communicative sense. In fact, one could argue that under such a view pragmatics is not simply concerned with context-dependent information, but rather with such information as is determined based on a consideration of the speaker’s intention in making a particular utterance. Thus, recall that for both authors, speaker intentions only play a role after what is said/the utterance meaning has already been determined. However, as we will see below, there are arguments that such a view of the role of speaker intentions in the overall interpretation process is too restrictive. More specifically, it seems that speaker intentions already have to be considered in order to resolve reference and fix indexical expressions.9 Thus, I 9 Cp. Kaplan (1989a) on the role of directing intentions for fixing the referent of true demon- 94 Utterance Meaning and the Literal/Non-literal Distinction will continue to use the term pragmatic to refer to processes and meaning aspects that essentially are context-dependent, unless indicated otherwise. There is a great deal of discussion concerning the level of what is said as Grice characterised it, since this level results from an interaction of the semantics component and pragmatic10 system. Therefore, it is taken to reflect the semantics/ pragmatics interface, the nature of which is highly controversial. Thus, in the next section, we will be concerned with various problematic aspects that have been raised with respect to Grice’s level of what is said, most importantly the claim that this level is not rich enough in meaning to form the basis from which further pragmatic inferences are drawn. Note also that although Bierwisch’s characterisations of the levels of expression meaning and utterance meaning seem to better capture the data and empirical results reviewed above, generally his level of utterance meaning faces the same problem as Grice’s, when it comes to forming the basis for pragmatic processes leading to communicative sense. It is important to consider these problems as we have to be clear about the character of what is said/utterance meaning before we can consider the question of which of the processes involved in determining it actually lead to what we would like to call the literal meaning of an utterance. 3.2 The Problem of Characterising the Level of Utterance Meaning Taking Grice’s (characterisation of the) level of what is said as a starting point, there actually are a number of divergent views on exactly how the level of utterance meaning11 should be characterised and what kinds of processes take place when moving from expression meaning to utterance meaning. Thus, as we will see, there is a lot of discussion about whether only such processes should be assumed that provide the values to variables already present in the semantic form of an utterance (i.e. linguistically mandatory processes; roughly Grice’s view), or whether, in order to account for the facts, processes should be allowed that add pragmatic material freely. This issue is closely connected with the view one takes on how much and what type of contextual information is available when moving from expression meaning to utterance meaning. Both in Grice’s and Bierwisch’s approaches, it can be argued, context is taken to encompass such information as identified by Kaplan (1989b) to stratives. 10 in the sense used here 11 At the risk of seeming inconsistent, in this section, I use the terms expression meaning and utterance meaning with a less specific characterisation than intended by Bierwisch. Thus, expression meaning is the context-independent meaning of simple and complex expressions, with no assumptions made as to whether this level of meaning is characterised by a high degree of underspecification. Utterance meaning is that level of meaning which forms the basis for drawing further pragmatic inferences. As such, I assume that it is propositional, however, there are no particular assumptions made as to the kinds of processes involved in determining this level. My reason for doing so is that the different approaches I discuss here all assume (slightly) different characterisations for the level of meaning that forms the basis for further pragmatic inferences and they also have different views concerning the exact nature of the level of context-independent meaning. Therefore, I want to be able to refer to those two general levels of meaning without yet making commitments as to their specific aspects. The Problem of Characterising the Level of Utterance Meaning 95 be necessary for fixing the reference of indexicals. That is, the speaker/producer as well as the hearer/addressee of the utterance, the time and place at which the utterance takes place and the world of the utterance have to be (mutually) known. This particular set of information is often called narrow context. It differs from broad context in that the latter additionally includes information resulting from reasoning about the speaker’s intention in making the particular utterance he did. Remember that Grice as well as Bierwisch assumed speaker intentions to be relevant only at the level of what is meant/communicative sense. However, a case can be made that speaker intentions already have to be taken into account when resolving the reference of indexical expressions, which would mean that broad context already is at play in determining utterance meaning. In section (3.2.1), I will first look at RT’s notion of explicature, which is intended to replace Grice’s what is said as the level of meaning that forms the basis for further pragmatic inferences. As we will see, Bach’s notion of impliciture is very similar to RT’s explicature, however, the two approaches differ in some crucial aspects. In section (3.2.2), I will discuss Stanley’s characterisation of what is said as resulting exclusively from linguistically mandated processes. Section (3.2.3) is a discussion of two notedly semantic approaches – that of Borg and that of Cappelen and Lepore – to the determination of fully propositional sentence meaning as opposed to the two approaches discussed in section (3.2.1), which both assume that sentence meaning (more or less often) is sub-propositional. Finally, section (3.2.4) reviews a number of experimental studies aimed at answering the question whether or not the level of what is said as characterised by Grice has any significant role to play in the interpretation process. 3.2.1 Explicit/Implicit Meaning Explicatures Recall that although the processes of resolving reference, fixing indexicals and disambiguation clearly require contextual information, Grice apparently did not think of them as being pragmatic in nature. However, as has been pointed out especially in the relevance-theoretic literature, these processes are similar in nature to the processes that result in the derivation of such phenomena as conversational implicature (e.g., Sperber and Wilson 1995, Carston 1999, 2002a,c, 2004a). Thus, these two types of processes cannot be differentiated on the grounds that one type is pragmatic (i.e., context-dependent) in nature whereas the other is not. Another criterion has to be found. The criterion used by Relevance Theory is that of explicitness of meaning. That is, pragmatic processes are differentiated as to whether they result in explicitly or in implicitly expressed meaning. Thus, on the one hand, there are pragmatic processes that apply to the expression meaning of an utterance and result in a fully propositional form. The propositional form that is derived by these processes is the meaning which is explicitly expressed by an utterance (or rather by the speaker of that utterance, as we will see). On the other hand, there are pragmatic processes that apply to the fully propositional form of the utterance of some expression and derive further pragmatic inferences, which are taken to be implicitly conveyed by that particular utterance of the expression. That is, what is explicitly 96 Utterance Meaning and the Literal/Non-literal Distinction expressed by an utterance is the basis for inferring what is implicitly conveyed. Before going into more detail concerning the difference between explicit and implicit meaning in Relevance Theory, let us first look at some basic assumptions made in that approach. Relevance Theory (RT) basically differentiates two kinds of processes: the decoding process of the language system and the pragmatic processes of inference. This differentiation reflects the understanding of the semantics/pragmatics distinction in RT. Semantics deals with linguistically encoded meaning, totally free from context and context invariant. Pragmatics, essentially, is the recovery of speaker meaning and is entirely context-dependent. Thus, semantics is nothing more than the source of evidence required by the pragmatic system in order to achieve an interpretation of an utterance. The distinction is not one of representational format, that is, there are no pragmatic representations as such, only pragmatic processes for the derivation of meaning representations. The output of representations from the language processor, the semantic representations make no claims about language external reality. They simply are of the appropriate format that allows ‘. . . for integration with representations from other information sources’ (Carston 2004a, p. 820). As a result of the integration processes, fully propositional representations are derived. These full propositions, then, may represent possible circumstances and can thus be judged against an external reality concerning their truth or falsity. From a relevancetheoretic point of view, then, the only semantic part of an expression is its semantic form, which is the output of the context-independent linguistic encoding, but nothing of which could be said that it was the intended speaker meaning. In delivering the schematic, sub-propositional, decoded SF of an utterance, the semantics component is exhausted. This SF is then supplemented with contextual information supplied by the pragmatic system (cf. Carston 2004a). What is interesting here is that RT assumes that more processes have to take place when moving from SF to an utterance’s meaning than those traditionally assumed by Grice. Thus, there are cases in which a saturation process has to supply missing material before the utterance of some sentence can be fully propositional. Thus, saturation ensures that utterances of sentences as in (63) are understood by a hearer as expressing propositions. (63) a. b. c. d. Paracetamol is better. [than what?] It’s the same. [as what?] He’s too young. [for what?] She’s leaving. [from where?] Similarly to the process of fixing indexicals, saturation makes use of contextual information. For the examples in (63) this means that the missing material asked for in the brackets, is supplied. Moreover, the assumption is that, like the former, the latter process is linguistically mandated, namely by some variable present in the semantic form of the sentence, which is assigned a value during interpretation. Assuming that the most important function of the level of meaning Grice identified as what is said is to form the basis for further pragmatic inferences, RT notes that what is usually taken to be that level, in many cases, cannot be the basis from which further inferences are drawn (Sperber and Wilson 1995, Carston 1997, 1999, The Problem of Characterising the Level of Utterance Meaning 97 2002c). For example, Carston (2002c) shows that Grice’s level of what is said is not rich enough in meaning to actually form the basis for such further pragmatic inferences as conversational implicatures. Thus, consider the examples in (64) (chosen from Carston 2002c, 2004b). (64) a. b. c. d. It’ll take time for this wound to heal. She has a brain. I haven’t eaten. There’s nothing on telly tonight. Utterances of the sentences in (64) all have full propositions at the level of what is said, that is, after reference resolution, disambiguation and possibly other saturation processes have taken place. However, these propositions are banal tautologies (64a - 64b), or banal falsehoods (64c - 64d). That is, they are uninformative in that they do not add to or revise the hearer’s existing assumptions concerning the world. As such, they will not trigger any further pragmatic inferences and thus an utterance of sentences such as in (64) would have no cognitive effects at all. However, in reality this clearly is not the case. Rather, what the speaker uttering sentences such as in (64) would intuitively be taken to have expressed, depending on the particular context, might be the following. (65) a. b. c. d. It’ll take a considerable amount of time for this wound to heal. She has a high-functioning brain. I haven’t eaten yet/today. There’s nothing worth watching on telly tonight. As a consequence, RT assumes that there are processes involved in determining what was said by an utterance which crucially are not linguistically mandated. One of these is the process of free enrichment, which provides constituents that are not present in the semantic form of the uttered sentence. Thus, a proposition that is explicitly communicated by an utterance may contain what might be called unarticulated constituents.12 The proposition determined by the various linguistically mandated saturation processes plus the process of free enrichment, then, actually is rich enough to form the basis for further inferences. Thus, in a particular context, from (65a), the addressee might infer the implicature that he will be restricted in his usual mobility. From (65b), in an appropriate context, the addressee might infer that the person referred to is a good candidate for a particular position. Similarly, from (65c), the addressee might infer that the speaker is likely to want something to eat and from (65d) the addressee might infer that the speaker does not want to spend the evening watching TV. Carston (1999) describes another process that might take place when moving from an utterance’s SF to that utterance’s meaning. This process, however, does not insert additional material or provides the value for some variable, rather, it adjusts the concept picked out by some particular lexical item. Thus, Carston discusses the example in (66). 12 Cf. Perry (1998) for a first introduction to the phenomenon of unarticulated constituent. 98 Utterance Meaning and the Literal/Non-literal Distinction (66) Kato (of O. J. Simpson, at his trial): He was upset, but he wasn’t upset. [He was upset’ but he wasn’t upset”] Assuming that the expression upset is used to express one and the same concept, Kato’s utterance would have to be taken as blatantly contradictory. However, as Carston notes, it was not intended, nor was it understood as such. Rather, what Kato is taken to have intended is to express two different concepts by his use of the form upset, one of which he states that it did apply to Simpson at a particular time, and one of which he expresses that it did not hold of Simpson at that same time.13 Relevance Theory calls this process ad-hoc concept formation. Thus, there are cases in which some lexical item is not used to express the concept that it encodes, but some pragmatically adjusted concept, which may be narrower or looser than the original concept and, thus, merely overlap with the original concept’s denotation. Concepts that result from the process of ad-hoc concept formation are not captured by lexical approaches, since their defining property is that they are the result of an online, ad-hoc interpretation process. The idea is that such concepts are not permanently stored. That is, they are usually highly specific, picking out situations that are very roughly characterised as being non-normal. Thus, in example (66), the idea is that this ad-hoc concept is such that there is no economic or other reason for storing it in long term memory because it is rather unlikely that it will be needed again in that form. Having said that, the idea of ad-hoc concepts is also interesting from a diachronic point of view, as some cases of nonce-formation arguably are adhoc concepts, which may or may not become institutionalised. Be that as it may, ad-hoc concept formation is one of the processes that may apply when determining an utterance’s meaning and it is a process that results in what can be called another type of non-literal meaning, in the sense that the process results in a concept that deviates from and replaces the concept the expression is normally taken to encode. Having said this, it should be noted that the characterisation of ad-hoc concept formation is not unproblematic when considering RT’s general view of the nature of lexically encoded meaning. Remember that Carston (2002c) proposes that what is encoded by lexical items are not full-fledged concepts, but rather abstract meanings in that they merely are pointers to concepts. From this viewpoint, a lexical item is never used to express what it encodes and it does not make sense to speak of adjusting what is lexically encoded, as this is not a full concept.14 If what an expression encodes is only a concept schema, then this schema needs to be filled in before it can be adjusted. In that case ad-hoc concept formation would be similar (although more general and less systematic in its output) to the process of metonymic shift mentioned above. Note further that it is not entirely clear how ad-hoc concept formation actually works, since, generally in RT, it is assumed that concepts are atomic. If 13 Thus, a rough paraphrase of what Kato might have intended to convey is: ‘O.J. Simpson was sad and angry to a certain degree, but he was not sad and angry to a degree that led him to killing his wife.’ 14 Cp. Burton-Roberts (2005, p. 404): ‘. . . if “word meaning” is not a concept but merely a pointer to concepts, do we need – in fact, can we have – any notion of either “narrowing” or “loosening”? Surely, those notions only make sense if “word meaning”is taken to consist in “a concept” . . .’. The Problem of Characterising the Level of Utterance Meaning 99 that is so, then the question arises how, e.g., upset* can be an adjusted upset concept. That is, if all concepts are atomic, it is not possible to relate an adjusted concept to the original concept by claiming that they share a certain amount of features. Having said that, recall that RT actually associates three distinct entries with a concept and thus, allows for correspondence/deviation of features in those entries. However, as pointed out already, with the assumption of such entries, it is not clear in which sense concepts are still atomic. Leaving these problematic issues to the side for the moment, let us turn back to the propositional form resulting from the pragmatic saturation, enrichment and adhoc concept formation processes, which RT calls explicature in order to differentiate it from the traditional Gricean what is said. Explicatures differ from Grice’s what is said in a number of respects. Thus, explicatures are taken to be the proposition expressed by the speaker, rather than merely a full, but in a sense minimal proposition (cp. figure 3.4)15 . In other words, explicatures are determined not just on the basis of processes that make use of speaker-independent, ‘narrow’ contextual information, but are the result of a number of pragmatic processes, some of which do take into account assumptions about what the speaker wanted to express by making a certain utterance. Furthermore, some of these processes insert material into the SF of an utterance that is not linguistically mandated, whereas in the determination of traditional what is said, only processes are involved that provide values for variables already present in the respective SF (cf. Bierwisch’s approach above). Semantics lexical meaning semantic composition sentence meaning reference resolution fixing indexicals disambiguation Pragmatics what is said/minimal proposition free enrichment ... explicature/proposition expressed basis for further pragmatic inferences ... Figure 3.4: Minimal proposition vs. proposition expressed Thus, in RT, it is explicatures which are the basis for drawing further pragmatic inferences; the basis for implicit meaning content. Thus, in example (67), the conversational implicature is not drawn on the basis of Gricean what is said by B, but rather on the basis of the explicature of B’s utterance. (67) A: Would you like to join us for lunch? B: I had a very large breakfast. 15 The term minimal proposition, more on which below, is taken from Recanati (2004). 100 Utterance Meaning and the Literal/Non-literal Distinction EX: I had a very large breakfast today. CI: B doesn’t want lunch. In fact, in RT’s cognitively oriented approach to interpretation, the traditional level of what is said has no psychologically significant role to play. In other words, there seems to be no reason why the level of meaning that results from the application of the processes of disambiguation, reference assignment and fixing of indexicals should be recognised as cognitively significant, since this level has been shown not to be rich enough to form the basis for further inferences. This is why RT actually does not recognise what is said as an individual level in addition to, and distinct from explicature. Thus, it is explicatures which are taken to be explicitly expressed by the speaker of some utterance, whereas implicatures are implicitly communicated (cp. figure 3.5).16 Semantics lexical meaning underspecified semantic composition sub-propositional Pragmatics sentence meaning saturation processes free enrichment ad-hoc concept form. + broad context ... actual truth-conditions fully propositional explicature basis for further pragmatic inferences conversational implicature speech acts + broad context ... what is meant Figure 3.5: The Relevance-theoretic View However, since pragmatic processes are involved both in the determination of explicatures as well as implicatures, the question is, how these two types of meaning are actually differentiated. Carston (2002c, p. 124) offers the following definition for explicature. An assumption (proposition) communicated by an utterance is an ‘explicature’ of the utterance if and only if it is a development of (a) a linguistically coded logical form of the utterance, or of (b) a sentential sub-part of a logical form. Thus, in order to differentiate explicatures from implicatures, one has to check whether the communicated proposition under consideration is a development of the SF of the utterance in question. As Burton-Roberts (2005, 2006) notes, the problem here is the term ‘development’, which itself does not get a clear definition. He suggests the following definition: ‘. . . a communicated proposition P is a “development” of the encoded logical form L of the sentence uttered – and thus explicated 16 For a critique of RT’s use of the terms implicit and explicit see below. The Problem of Characterising the Level of Utterance Meaning 101 – if and only if P (asymmetrically) entails L’ (Burton-Roberts 2005, p. 397). Thus, in the examples below, the explicatures in (b) entail the logical form of the sentence uttered in (a). (68) a. He shrugged and left. b. He shrugged and then left. (69) a. He has three kids. b. He has exactly three kids. (70) a. I’ve had breakfast. b. I’ve had breakfast today. Intuitively, this criterion seems to work, however, Burton-Roberts (2005) points out that it actually is problematic. Thus, for an explicature of an uttered sentence to entail the logical form of that sentence, the latter would have to be propositional, which RT claims it is not. Another problem is that the entailment characterisation of explicature cannot be applied in the case of negated sentences. (71) a. I’ve not had breakfast. b. I’ve not had breakfast today. (72) a. You won’t die. b. You won’t die from that cut. In these examples the explicature in (b) does not entail the logical form of the sentence used to make the respective utterance, rather, the entailment relation is reversed (the logical form entails the putative explicature). There is yet a further aspect about explicatures that is controversial. As noted above, the level of explicature is taken to be that level at which an utterance becomes truth-evaluable. In this respect, it is similar to traditional what is said, which also was taken to be the level of meaning at which an utterance is truth-evaluable. More specifically, what is said was that level of meaning to the ‘expressing’ of which the speaker of an utterance was taken to be committed. However, explicature is the result of pragmatic processes which are similar in kind to those processes that derive conversational implicatures, where one of their characteristics is that they are cancellable. In consequence, Carston assumes that explicatures should be cancellable as well. However, as Burton-Roberts (2005, 2006) points out, an explicature of some utterance cannot be cancelled without contradiction.17 He suggests that explicatures may be clarified, but they may not be cancelled. That is because explicatures are taken to be the meaning the speaker intended to express by making the utterance she did. Burton-Roberts (2006) argues that such implemented intentions cannot be cancellable. For an illustration, take the following example from Carston (2002c, p. 138), which Burton-Roberts (2005, 2006) elaborates on. (73) a. She’s ready. b. Karen is ready to leave for the airport. 17 For another defence of the idea that explicatures are not cancellable, see Capone (2006, 2009) 102 Utterance Meaning and the Literal/Non-literal Distinction c. She’s ready but Karen isn’t ready to leave for the airport. Thus, the sentence in (73a) may be used by a speaker, in an appropriate context, to express what is given in (73b). Now, Carston (2002c) goes on to claim that this explicature is pragmatically cancellable without contradiction by an utterance of the speaker of (73c). As Burton-Roberts (2006) convincingly argues, this cannot be the case. To make sure that it is clear which possible explicature for (73a) we are interested in here, Burton-Roberts (2006) states it in a slightly more formal way as in (74). (74) Karen[K] is ready at time[U ] to leave for the airport[A] The explicature of interest here for the second conjunct in (73c), Burton-Roberts states as in (75). (75) Karen[K] is not ready at time[U ] to leave for the airport[A] But then, clearly, the second conjunct in (73c) cannot be used to cancel the explicature of the first conjunct without contradiction, if that is taken to be the speaker intended (74). The reason for this is very simple: what the speaker is assumed to have intended to express, he cannot take back without contradiction. The only thing a speaker can do is make it clear to the hearer what she actually intended to express, in case the hearer got her wrong. Thus, another example Burton-Roberts looks at is the following. (76) Ann: That fellow’s playing is lamentable. Bill: Too right. Cruelty to cellos, I call it. Ann: Not the cellist - the trombonist! Thus, Bill, with his reply to Ann’s first utterance, makes it clear that he assumed Ann was referring to the cellist when she said that fellow. Moreover, he indicates that he agrees with, what he thinks, she expressed about that person’s ability of playing his instrument. However, as Ann’s second utterance makes clear, Bill, unfortunately, simply ‘got her meaning wrong’. That is, the reference of that fellow which Bill’s pragmatic interpretation process supplied, simply was not the reference Ann had intended. Thus, Ann, with her second utterance clarifies what she actually intended to express by her first utterance. More specifically, she makes it clear to Bill that by her use of that fellow she did not intend to refer to the cellist, but rather to the trombonist. However, Ann does not ‘cancel’ the explicature that Bill thinks she expressed, because, in fact, she never did express that explicature. Thus, although explicatures are pragmatically derived, in contrast to implicatures, they are not cancellable. That is because a speaker cannot take back, without contradiction, what she did in fact intend to express. Implicitures Another approach that tries to bridge the gap between the semantic form of an utterance and what was meant by the speaker uttering the expression in question, is that advocated by Kent Bach (e.g. Bach 1994a, 1997, 2004, 2005, 2006a). In The Problem of Characterising the Level of Utterance Meaning 103 contrast to RT, Bach does recognise what is said as a distinct level of meaning, although he agrees with RT in that Grice’s notion of what is said does not play the role in utterance interpretation traditionally assumed. That is, Grice’s notion of what is said (and for that matter Bach’s as we will see) does not provide a rich enough basis from which to draw such pragmatic inferences as conversational implicatures. Nevertheless, Bach (1994a) recognises the level of what is said as the level of meaning explicitly expressed by a speaker, where this contrasts, on the one hand, with Grice’s level of implicated meaning what is meant and on the other hand with a level of meaning Bach (1994a) characterises as implicit and accordingly calls impliciture. As regards Grice’s level what is said, Bach (2001b) assumes that it is subject to what he calls the Syntactic Correlation constraint. This means that ‘. . . if any element of the content of an utterance, i.e., of what the speaker intends to convey, does not correspond to any element of the sentence being uttered, it is not part of what is said.’ (Bach 2001b, p. 15). Thus, what is said is the result of the semantic interpretation of the sentence uttered, that is, the meaning of the uttered sentence relative to the context of utterance, where the latter includes only a limited amount of information (comparable to the information allowed for by Kaplan) and does not include speaker intentions. As a consequence, what is said may not be fully propositional and it may not be the meaning the speaker is taken to have expressed. However, what is said forms the basis on which full propositions and, more specifically, the proposition intended by the speaker may be determined. Bach (1994a) assumes two processes which accomplish this task, namely completion on the one hand, and expansion on the other. Thus, completion comes into play if what is said is only of sub-propositional form, or what Bach calls a propositional radical. This process adds to what is said by an utterance content which is only implicitly communicated. Similarly, in cases where what is said is fully propositional or where completion already has successfully applied, the resulting proposition may still not be the one the speaker is taken to have intended to communicate. In such cases, the process of expansion will provide conceptual material which, again, was intended to be communicated, but which is only implicit. Since the material that is added to what is said by completion and expansion is not explicitly expressed, Bach calls the resulting propositions implicitures. As Bach (2006a) notes, RT’s characterisation of explicature as being what is explicitly expressed by an utterance is somewhat irritating. Thus, what differentiates an explicature of some utterance from that utterance’s semantic form is that the former may include material that is in fact not explicitly expressed at all by the SF of the utterance. In stating an explicature of some utterance, material that has been implicitly conveyed is being made explicit. In contrast, Bach uses the term impliciture to differentiate a level of meaning that is distinct from what is said, but also different from the level of meaning Grice calls what is meant. However, as he points out in Bach (2006a), his notion of impliciture and RT’s notion of explicature are rather similar in that an impliciture may result from an expansion of an utterance’s SF, that is, some kind of free enrichment process. Moreover, what makes explicatures and implicitures similar, and differentiates both from implicatures, is that they are conveyed directly. Thus, with implicatures, a speaker means one thing by saying what he says and conveys something in addition (and thus indirectly), 104 Utterance Meaning and the Literal/Non-literal Distinction Semantics lexical meaning semantic composition sentence meaning saturation processes sub-propositional actual truth-conditions fully propositional + narrow context what is said completion expansion impliciture + broad context basis for further pragmatic inferences conversational implicature + broad context speech acts ... Pragmatics what is meant Figure 3.6: Bach’s view whereas implicitures include what has been said, but also include material that remains implicit. Note that Bach (2006a) claims that there are more differences in the two frameworks that the notions of impliciture and explicature are part of than between the notions themselves. From this viewpoint, one major difference between the two notions is that RT’s explicature is taken to be the property of an utterance, whereas Bach takes implicitures to be more akin to implicatures ‘. . . in that both are things that speakers mean in saying what they say’ (Bach 2006a, p. 6). That is, in contrast to RT, Bach assumes that a differentiation between two levels of meaning (sentence meaning18 and speaker meaning) is sufficient and that there is no need for a third intermediate level (utterance meaning). Thus, ‘[u]tterances themselves . . . have no content apart from the semantic content of the uttered sentence and the content of the speaker’s intention in uttering it’ (Bach 2001b, p. 31). Interestingly, Bach (2000), like Carston with explicatures, also assumes that what is implicit in an utterance can be taken back, that is, implicitures are cancellable. Thus, consider the example below (from Bach 2000). (77) a. Not every bottle is empty. b. Not every bottle I bought today is empty. c. Not every bottle is empty – but every bottle I bought today is empty. Depending on whether one believes in quantifier domain restriction (cp. Bach 2000, Stanley 2000, Stanley and Szabo 2000, Recanati 2002, Carston 2004a, Pagin 2005), the material in italics in (77b) is, in Bach’s terms either the result of a completion or an expansion of what is said by the sentence (77a) as it is used in a particular context. Since it is possible for the speaker of (77a) to continue as indicated in (77c) without contradicting herself, what is left implicit in her utterance of (77a) is not part of what is said. However, the same criticism of this view as was given by Burton-Roberts (2006) for explicatures also holds here. That is, if the speaker of (77a) intends her 18 that is, the meaning of a sentence relative to a context of utterance as Bach conceives of it The Problem of Characterising the Level of Utterance Meaning 105 utterance to be taken as expressing (77b), she would be contradicting herself by saying something like (77c). This is because what is intended cannot be cancelled. Bach notes that implicitures differ from such phenomena as metaphor in that there is no use of particular words in a figurative way. Rather, implicitures capture the fact that one may communicate with an utterance more than what is explicitly expressed, namely information that is implicit in what one actually says. One particular type of such implicit meaning, namely the one resulting from the process of expansion, leads to what Bach calls sentence nonliterality. Thus, consider example (78), in the context of a mother talking to her son, who has just cut himself and is crying loudly. (78a) is what the mother says to him, (78b) is what, in this particular situation, she may be taken to have intended to express. (78) a. You’re not going to die. b. You’re not going to die from this cut. Bach says of this example: ‘. . . the mother is using each of her words literally but is omitting an additional phrase that could have made what she meant fully explicit. If her son had replied, “You mean I’m going to live forever, Mom?”, it would not be because she was being obscure but because he was being obtuse–he would be taking her utterance strictly and literally, not as she meant it. But if she were using each of her words literally, how could she not be speaking literally? This example illustrates a common but not widely recognized kind of nonliterality, whereby a sentence is used nonliterally without any of its constituents being so used.’(Bach 1994a, p. 135) Thus, Bach recognises a further type of non-literal meaning, namely one that differs from metaphor or metonymy in not being associated with particular constituents of an utterance, but rather with the utterance as such. Interestingly, concerning the son’s possible reply, Bach says that the son, in this case, simply was not able to figure out what his mother meant. However, in a slightly different context, assuming, for example, that the son is not all that young, he might actually reply with You mean I’m going to live forever, Mom?, while being perfectly aware of what his mother intended. One reason for doing so might be to achieve some comic effect. What is interesting here is that this presupposes that the son both understands what his mother intended to express by saying (78a) as well as, possibly, a more minimal proposition, which, in fact, she cannot plausibly be taken as intending to express. This is an interesting aspect to keep in mind, and we will come back to it below, when it comes to assumptions about whether minimal propositions actually play a role during the interpretation process. As mentioned above, Bach retains Grice’s distinction between what is said and what is implicated in that he assumes that there is a gap between the explicitly expressed (what is said) and the implicitly and additionally conveyed. However, if an impliciture is, in a sense, an expanded what is said, including implicit but nonetheless communicated material, the question arises, what differentiates between 106 Utterance Meaning and the Literal/Non-literal Distinction implicitures and implicatures. Similarly to explicatures implicitures are also the result of pragmatic processes, which are assumed to be similar to the processes leading to implicatures. Bach argues that implicitures in contrast to implicatures are built up from what is said, that is, the explicit meaning content of an utterance. An impliciture of some utterance includes material that might have been expressed explicitly ‘if the appropriate lexical material had been included in the utterance’ (Bach 1994a, p. 140). In contrast, implicatures arise from the saying of what is said. Thus, implicatures are conceptually independent of both what is said and the impliciture of an utterance. Note that Bach’s characterisation of implicitures as being ‘built up from what is said’ is as imprecise as RT’s notion of explicature being a ‘development’ of a logical form. Incidentally, neither RT nor Bach can characterise the level of implicitly expressed meaning (explicature/impliciture) as resulting from the ‘filling in’ of linguistically mandated ‘slots’ in the semantic form of some expression, since both allow for the insertion of information into the semantic form of an utterance without there being a linguistically predetermined ‘slot’, e.g., a variable for it. In this respect, both RT and Bach’s approach differ crucially from Stanley’s approach (e.g. Stanley 2000), who assumes that all the information that is part of what is said by an utterance needs to be mandated by something in the semantic form of that utterance, as produced by the process of semantic composition. 3.2.2 Unarticulated Constituents vs. Hidden Indexicals According to Stanley, in arriving at what is said by some utterance, extra-linguistic context only plays a role insofar as it provides the values for variables already present in the semantic form of that utterance, or in supplying the content of indexical expressions and resolving ambiguities. That is, only saturation processes take place in the determination of what is said. In other words, he assumes that all truthconditional effects that the extra-linguistic context in which an utterance is made has on what is said by that utterance, are linguistically mandated. Semantics lexical meaning semantic composition saturation processes actual truth-conditions fully-propositional sentence meaning rel. context/ what is said + broad context basis for further pragmatic inferences conversational implicature speech acts + broad context ... Pragmatics what is meant Figure 3.7: Stanley’s view With respect to the role of variables in the SF of utterances, Stanley’s approach is similar to Bierwisch’s view of the level of utterance meaning and the nature of The Problem of Characterising the Level of Utterance Meaning 107 the processes that contribute to it. That is, processes like conceptual shift and conceptual differentiation also rely on the presence of variables in the semantic form of the expressions concerned. Thus, these processes are linguistically mandated. Both in Stanley’s and in Bierwisch’s approaches, the variables are assumed to be part of the semantic form of the respective utterances, that is, they are contributed by one or another of the individual expressions used to make the utterance. In his (2000) article, Stanley argues against the existence of non-sentential assertions and generally against unarticulated constituents in the semantic form of an utterance, both of which phenomena have been used to argue for the application of pragmatic processes that are not linguistically mandated as, e.g. free enrichment. Thus, he notes that for examples such as (79) below, B’s utterance of Bill does not constitute a non-sentential assertion but rather is a case of syntactic ellipsis. By uttering ‘Bill’, B will be taken to have expressed the proposition Bill bought the bottle. Thus, the logical form of B’s utterance actually includes information which simply is not phonologically expressed. (79) A: Who bought the bottle? B: Bill. B can use his utterance of ‘Bill’ to express the proposition that Bill bought the bottle because A’s question, in a sense, provides the frame within which B’s utterance is interpreted. Thus, cases of syntactic ellipsis do not occur discourse-initially, because the correct interpretation of elliptical sentences depends on the context in which they are uttered. Looking once again at the examples given above, where RT assumes that the sentences do not express full propositions without some process of saturation, one could argue that these are cases of syntactic ellipsis, especially since an utterance of any of them would seem odd discourse-initially. (63) a. b. c. d. Paracetamol is better. [than what?] It’s the same. [as what?] He’s too young. [for what?] She’s leaving. [from where?] Generally, Stanley assumes that syntactically elliptical sentences need a linguistic antecendent. However, he argues that explicitly mentioning that antecedent is only one way of providing that antecendent. It should be noted, that Stanley has a very broad conception of what counts as a discourse and as such provides the necessary contextual information for the correct interpretation of a syntactically elliptic sentence. Thus, consider one of his examples, given in (80), for which Stanley provides the following context. ‘Suppose Bill walks into a room in which a woman in the corner is attracting an undue amount of attention. Turning quizzically to John, he arches his eyebrow and gestures towards the woman. John replies:’ (Stanley 2000, p. 404) (80) A world famous topologist. 108 Utterance Meaning and the Literal/Non-literal Distinction Stanley assumes that the context preceding John’s utterance, namely Bill’s quizzically turning to John, his arching of eyebrows and gesturing towards a specific woman, make salient a question that Bill might have uttered but did not, namely Who is that? Because the context makes this question salient, John can utter (80), which Bill will be able to recognise as the answer, as it where, to his implicitly asked question. Thus, in the example above, the necessary linguistic antecendet (Who is that?) is implicitly provided by the situation in which John’s utterance of (80) takes place. That is, it is plausible to assume that in this situation, Bill tries to make explicit his intention by using non-verbal clues such as gestures and facial expressions which are ostensively directed towards John. However, I think it questionable whether these non-verbal clues actually are of such a definite and determinate nature as to make salient a specific linguistic complex expression such as Who is that? Thus, consider another example, taken from Carston (2002c), where the context is the following. B is coming into the kitchen and sees A, apparently searching for something, as he is looking around the lower shelfs of a cupboard in the kitchen. It is breakfast time and B knows that A likes to have marmalade for breakfast, and he assumes that that is what A is looking for. Thus, B says the following. (81) On the top shelf Now, Carston (2002c) assumes that in order for A to interpret correctly that what B intended to express was in fact a full proposition (namely the proposition that the marmalade is on the top shelf of the cupboard), he has to pragmatically enrich B’s utterance to arrive at that proposition. According to Stanley, however, there are no such things as non-sentential assertions which are not syntactically elliptical or ‘shorthands’19 . However, (81) clearly qualifies as an assertion. Thus, Stanley would have to explain this example along the lines of the example given above. That is, the context of B’s utterance of (81) makes salient the question Where is the marmalade, a question A could have uttered but did not explicitly so. However, the context of B’s uttering (81) differs in some important aspects from the context of (80). Thus, there are no facial expressions or gestures on the part of A which are intentionally directed towards B and which could function as clues as to the determinate linguistic expression A could have but did not utter. In fact, in such a situation it is likely that A does not entertain any possible linguistic complex expressions applicable to the situation at hand at all. Thus, if cases of syntactic ellipsis need a determinate linguistic antecedent for the recovery of the missing material, it is not clear to me, how in examples as that given above, this implicit but determinate linguistic antecedent can be made salient by the context. This question is even more pressing in cases where the interlocutor who is attributed as entertaining this implicit question, by his behaviour, cannot even be taken to have intended to convey any question whatsoever. Stainton (2005) states a connected problem for Stanley’s assumption that the context in which a syntactically elliptic utterance is made may be taken to provide 19 For his notion of ‘shorthand’, see Stanley (2000). An example is an utterance of ‘nice dress’ by some passer-by on the street to some woman, where this can plausibly be taken to be a ‘shorthand’ of the statement ‘This is a nice dress.’. However, as noted by Stainton (2005), this is a very vague notion and according to Stainton (2005, p. 404), Stanley has given it up. The Problem of Characterising the Level of Utterance Meaning 109 the necessary linguistic antecedent. He assumes that true syntactic ellipsis is resolved wholly linguistically (a plausible assumption considering that a determinate linguistic antecedent is necessary). That is, cases of syntactic ellipsis are resolved within the language module. Apparently, recovering the missing material is a purely linguistic process, since it relies on the availability of a determinate linguistic antecedent. However, if that is the case, then pragmatic processes should not play a role here. Providing a linguistic antecedent by way of making it salient in a context, would require pragmatic processes to take place. Thus, resolving syntactic ellipsis could no longer be treated as a solely linguistic phenomenon. Moreover, as mentioned above, it is not even clear whether, and even less how, a determinate linguistic antecedent can be made salient. As a last point concerning Stanley’s hypothesis of the non-existence of genuine non-sentential expressions: Pagin (2005) states that he finds it implausible to assume that whenever a speaker manages to communicate a thought using some linguistic expression, this implies that he used a fully articulated sentence, if only as far as the expression’s logical form is concerned. Rather, he thinks that ‘. . . it is not too rare that a speaker manages to get across a thought with poor linguistic means, falling short of what would be needed to get the thought across to an audience less in tune with the speaker’ (Pagin 2005, p. 330). Let us now turn to Stanley (2000)’s more general argument against the presence of pragmatically provided unarticulated constituents in the proposition expressed by an utterance. Stanley (2000, p. 410) gives the following definition for unarticulated constituents. ‘x is an unarticulated constituent of an utterance u iff (1) x is an element supplied by context to the truth-conditions of u, and (2) x is not the semantic value of any constituent of the logical form of the sentence uttered.’ (emphasis as in the original) From what has been said above about non-sentential assertions, it should be clear that if one assumes that there are genuine non-sentential assertions which are not cases of syntactic ellipsis and which can be used to express a full proposition, then one has to also assume the presence of pragmatically provided unarticulated constituents in the semantic form of such assertions. That is, unarticulated constituents constitute the material that needs to be integrated with the semantic form of non-sentential assertions in order for that semantic form to become fully propositional. However, even if one accepts Stanley’s argument against the existence of non-sentential assertions that are not cases of syntactic ellipsis, unarticulated constituents may still be at play where the semantic form of some utterance already is a full proposition, but not the proposition the speaker can be plausibly taken to have intended to express, as in cases of free enrichment. However, Stanley (2000) and Stanley and Szabo (2000) argue that what seems to be an unarticulated constituent pragmatically provided and somehow inserted into the semantic form of an utterance is in fact simply the pragmatically provided value of some variable that was present in the semantic form all along. The argument they pursue is that if some constituent is wholly pragmatically provided, it should not be able to be bound by some explicit quantifier occurring in the utterance in question. This is because 110 Utterance Meaning and the Literal/Non-literal Distinction the operators within a sentence can only interact with variables that are also part of the same sentence and which lie within the operators’ scope. The idea is that pragmatically provided constituents cannot lie within the scope of such operators. Let us look at some of the relevant examples. (82) It’s raining. Provided the speaker uttering (82) does so in a neutral context, she would normally be understood as expressing the proposition that it is raining at the time of utterance at the place where she is making that utterance. Thus, assuming that the semantic form of the utterance does include a temporal, but not locational variable, the truth conditions for an utterance of (82), according to RT, look as follows, where the location of the raining is pragmatically determined. (83) An utterance of ‘it is raining (t)’ is true in a context c iff it is raining at t and at l, where l is the contextually salient location in c. Now consider the possible interpretations of an embedded version of the simple sentence. (84) Every time John lights a cigarette, it rains. (85) a. For every time t at which John lights a cigarette, it rains at t at the location l in which John lights a cigarette at t. b. For every time t at which John lights a cigarette, it rains at t at some location which is salient in the context of utterance. Whereas RT can account for the reading in (85b), it cannot account for the reading in (85a), which is the preferred reading in this case. This is because in the preferred reading the location variable is within the scope of the universal quantifier. According to Stanley, however, a variable that is introduced by some pragmatic process cannot come within the scope of an explicit quantifier. Thus, the argument is that since in (84) the location variable is within the scope of the quantifier and thus varies with varying values for the temporal variable, the location variable is already part of the compositionally built up semantic form of the utterance at hand. Thus, there is no unarticulated constituent present here. As Carston (2002c) states, Stanley’s argument rests on the assumption that the pragmatic mechanisms cannot provide a reading that corresponds to a bound variable reading. Stanley’s arguments against such a possibility are based on what he assumes to be the nature of denotations and his view of what constitutes context. ‘. . . denotations of bound variables are odd, theoretically complex entities. It is difficult, if not impossible, to see how, on any account of salience, such an entity could be salient in a context. Certainly, neither it, nor instances of it, could be perceptually present in the context.’ (Stanley 2000, p. 414) Thus, it seems that for Stanley, only perceptible objects and properties constitute the context of an utterance. In contrast, RT assumes that, essentially, context The Problem of Characterising the Level of Utterance Meaning 111 consists of a set of mentally represented assumptions, where only some of those are representations of perceptible environmental features. Thus, for the case at hand, Carston (2002c, p. 200) suggests that the addressee might access a ‘. . . general knowledge assumption about the way in which times and places pair up when a certain type of event (such as “raining”)’ occurs. This will allow the preferred ‘bound-variable’ interpretation, unless the context is such that it prevents it. To be sure, the hidden indexicals analysis explains the preferred reading of (84) more elegantly, in assuming that there actually is a variable present in the semantic form of the utterance, which, in case of the preferred reading, gets bound by the universal quantifier. However, an argument can be made against the semantic necessity of such a location variable. For instance, Recanati (2004) argues that there is no such necessity for assuming a semantically provided location variable in the case of rain. Thus, he claims that one can imagine situations in which no particular place is required for a successful interpretation of an utterance such as (82).20 Since this is the case, there is no reason to assume that the semantics has to provide a location variable, requiring a value in every context. However, if there is no location variable in the semantic form of (82), how is it that we get a bound reading in (84)? As mentioned above, Stanley claims that such a reading is only possible, if one assumes that there actually is some variable in SF which gets bound by the quantifier. More specifically, he assumes that this variable is provided by the semantics of the verb to rain. Recanati (2004) offers an alternative analysis, where the variable is provided by the quantificational phrase. He claims that such quantificational phrases may be treated in parallel to singular phrases functioning as modifiers, where the latter introduce what are called variadic functions. ‘A variadic function is a function from relations to relations, where the output relation differs from the input relation only by its decreased or increased adicity’ (cf. Recanati 2002, p. 319). Thus, for a sentence such as (86), ‘. . . the phrase “everywhere I go” contributes both the adicityincreasing variadic function and the operator which binds the extra argument-role.’ (cf. Recanati 2002, p. 330). (86) Everywhere I go, it rains. Thus, (86) is properly represented by (87), where the quantifier binds the free variable in the subformula ‘in l (it rains)’, which it has supplied to it in the first place. (87) [For every place l such that I go to l ](in l (it rains)) The advantage of this approach is that a variable only has to be assumed in cases where the simple sentence It’s raining is embedded, whereas the simple sentence as such does not have such a variable.21 20 This is Recanati’s example context: ‘. . . imagine a situation in which rain has become extremely rare and important, and rain detectors have been diposed all over the territory (whatever the territory – possibly the whole Earth). In the imagined scenario, each detector triggers an alarm bell in the Monitoring Room when it detects rain. There is a single bell; the location of the triggering detector is indicated by a light on a board in the Monitoring room. After weeks of total draught, the bell eventually rings in the Monitoring Room. Hearing it, the weatherman on duty in the adjacent room shouts: “It’s raining!!”’ 21 However, see Pagin (2005) for a critique to the effect that even in the case of an occurrence of the simple sentence it is raining the prefixed form ‘in l (it rains)’ has to be assumed as its logical form, in which case the additional variable would no longer be optional. 112 Utterance Meaning and the Literal/Non-literal Distinction Note, however, that if what the verb to rain is taken to express is a particular type of event, conceptually speaking there is no question as to the status of the location variable. Thus, events in general are conceptually conceived of as necessarily involving a time (span) and a place at which they take place, although neither time nor place necessarily needs to be made explicit or be specified. In other words, a raining event of necessity involves a location at which the event takes place. Now, it seems that when expressing the taking place of a raining event as in (82), what the speaker standardly is taken to also express is that the location of the raining event is in close approximation to the speaker. That is, the location variable that events have in general, in this case is specified to something like in close proximity to speaker. However, the specification of the location variable as in close proximity to speaker is not part of the semantics of to rain but rather is a standardised, though defeasible meaning aspect. In fact, it reminds one of the notion of deictic centre that plays a role in the interpretation of deictic expressions. Thus, turning back to the example in (84) and its preferred interpretation as ‘whenever John lights a cigarette, it rains where John is when he lights a cigarette’: here the location variable associated with the raining event is not bound to where the speaker is, but rather to where the subject of the embedding sentence (John) is whenever he engages in a cigarette-lightning event. In other words, the location variable of the raining event co-varies with the location variable of the cigarette-lightning event. And since John is said to be involved in the cigarette-lightning event, the location of the raining-event seems to co-vary with where John is. However, again this seems to be a preferred interpretation, because if more thematic roles are introduced, the interpretation of the location of the raining event shifts. (88) a. Every time Susann calls Mary, it rains. b. Susann: Every time I call Mary, it rains. c. Mary: Every time Susann calls me, it rains. Thus, in an utterance of (88a), I would content, the location of the raining event cannot easily be interpreted as the location at which either Susann or Mary or even both are, when the calling-event takes place. However, the location variable of the raining event is still understood as varying with the location of the calling-event. The problem here is that it is much more difficult to indentify the location of that calling-event, as it involves two participants and the two participants may be in very different locations. Interestingly, as regards utterances of (88b) and (88c) by Susann and Mary, respectively, it seems that the location variable of the raining event is bound to where the respective speaker is when a calling event by Susann of Mary takes place. What these examples show is that even if one assumes that a location variable is semantically supplied due to the fact that the verb to rain is understood as expressing a particular type of event, supplying a value to that variable is not a trivial matter and depends very much on the overall structure of the utterance the verb is part of as well as on the specific circumstances holding in the particular utterance situation.22 22 See, however, section 5.2.1 for a view where the location variable is not taken to be part of the semantics of the verb but may still enter into the interpretation of an utterance involving that verb, albeit not ’freely’. The Problem of Characterising the Level of Utterance Meaning 113 Moreover, the location at which a particular event is assumed to take place does not necessarily have to be specified. Thus, note that in (84) no specific location is inferred for the event expressed in the embedding sentence and still the location variable of the event expressed in the embedded sentence is assumed to co-vary with that unspecified location variable of the embedding sentence.23 Turning back to Stanley’s hidden indexicals approach: a general criticism is the fact that it is not clear how many such variables should be assumed to be present in the semantic form of a particular utterance. Thus, Wilson and Sperber (2000, pp. 238) discuss the example below, where the context is that Lisa looks in at her neighbours, who are about to start supper. (89) Alan: Do you want to join us for dinner. Lisa: No, thanks. I’ve eaten. According to a standard semantics analysis, Lisa’s utterance of I’ve eaten has the truth-conditions stated below. (90) At some point in a time span whose endpoint is the time of utterance, Lisa has eaten something. However, in the context at hand, this clearly is not what Lisa will be taken to have intended to express. Rather, what she probably intends to express in this situation and what Alan will understand her to have expressed is something like the following. (91) Lisa has eaten supper this evening. That is, Lisa both communicates the object of the eating-process as well as a temporal specification of when the eating-process took place. Wilson and Sperber (2000), and following them Carston (2002c), assume that under the hidden indexical approach the semantic form of Lisa’s utterance includes two variables – one for the object eaten and one for a specification of the time span. (92) I have eaten x at t. Wilson and Sperber (2000) go on to show that in other circumstances, the proposition expressed by a speaker uttering I’ve eaten, might also give a specification of the place or manner of eating and possibly others. (93) I’ve often been to their parties, but I’ve never eaten anything [there]. (94) I must wash my hands. I’ve eaten [using my hands]. To capture the propositions expressed by these utterances, even more hidden indexicals would have to be postulated, yielding a semantic form for I’ve eaten as in (95), although, as Carston (2002c, p. 204) notes, in many cases the variables present would not get a definite specification at all (as in the example of Lisa’s utterance above)(cf. 96). (95) I’ve eaten [object x] [in manner y] [at location l] [within time span t]. 23 The discussion of these examples will be taken up once more in section 5.2.1. 114 Utterance Meaning and the Literal/Non-literal Distinction (96) I’ve eaten supper in some manner at some location this evening. However, in the context given above, (96) is not what we take Lisa to have expressed. That is, if she had chosen to make explicit the allegedly implicit material, the proposition expressed would have differed considerably from the one expressed by her simply uttering I’ve eaten. There are several points that have to be made concerning this criticism and the specific examples used to make it. First of all, conceptually, the act of eating clearly does involve something that is being eaten. Thus, ‘eating’ is a relation between someone who eats and something that gets eaten. Therefore, it is not unreasonable to assume that there is a slot for the eaten object in the semantic form of the verb, even though that slot does not need to get specified and it does not need to be overtly expressed. When Lisa says ‘I’ve eaten’, she is understood to have expressed that she has eaten something or other before the time of utterance. Moreover, in this particular situation, Lisa can be taken to use ‘I’ve eaten’ in order to give an explanation for her declining the invitation. It is questionable whether Alan in his interpretation of Lisa’s utterance has to go as far as to interpret her as having specifically expressed (91). While it is true that Lisa will be taken to have expressed that she has eaten recently, there is no reason to assume that she intended to express the specific object of the eating process she engaged in. Having said this, even if Alan does interpret her as expressing that she has eaten dinner, all that is necessary to arrive at that interpretation is to supply a value to the semantically supplied eating-object variable. Similarly, all that is needed for an interpretation of Lisa’s utterance as expressing that she has eaten recently (from which Alan can infer that it might make it undesirable for her to join Alan for dinner), is a pragmatically determined restriction of the time span semantically given by her use of the past tense.24 Similarly, the criticism that the semantic form of I’ve eaten, based on examples such as (93) and (94) would have to be assumed to include as many variables as given in (95), actually is not warranted. Thus, note that both examples (93) and (94) involve not just one utterance, but rather small texts. Thus, the [there] used in (93) to indicate a particular location of the eating event expressed by the second sentence actually is a shorthand for ‘at the contextually relevant party’ in this particular setting. The idea is that in order to interpret the second utterance as involving a particular location, no semantic location value needs to be posited, rather, the particular interpretation arises due to the particular discourse context in which the utterance is made. Part of that context is the proposition added to it by the utterance of the first sentence in (93). Similar points can be made regarding (94). Thus, I would assume that the examples stated above do not force Stanley to posit semantic variables for manner or location of eating.25 Rather what they show is that the discourse context in which an utterance takes place influences how that 24 This is similar to the today interpretation of ‘I’ve had breakfast’, where when this standard interpretation is cancelled by saying ‘I’ve had breakfast, but not today’, the speaker will still be understood to have expressed that he has had breakfast in the past, but will be understood as not including the day of utterance in that time span. Thus, in this example, the time span is pragmatically adjusted as well, but it was originally introduced by the past tense and not totally pragmatically supplied. 25 Although, as noted for to rain, one might actually argue that the location variable does form The Problem of Characterising the Level of Utterance Meaning 115 utterance is understood. Moreover, the fact that an utterance by Lisa of (96) would have been interpreted differently from the utterance she actually made, should not be seen as a problem of the hidden indexicals approach, but rather as being grounded in pragmatic considerations. That is, overtly expressing the proposition that seems to be covertly expressed by uttering ‘I’ve eaten’ raises questions concerning Lisa’s reasons for doing so.26 Having said this, although there are cases of variables in the SF of an utterance which are ‘only’ provided a value by pragmatic saturation processes, it seems that one has to allow for pragmatic processes that provide constituents which are not articulated by the utterance’s SF in any way. However, these constituents do not arise out of the blue, rather, one has to turn to the context in which the utterance that seems to include them is made. This holds also for non-sentential utterances, which can be understood to express propositions because they are interpreted within the context in which they are made, even though that context may not necessarily involve previous verbal interaction. 3.2.3 Minimal Semantic Content and Full Propositionality In chapter 2, I argued that the traditional view of semantics as both being concerned with context-independent meaning as well as delivering full propositions for sentences cannot be upheld. I argued for a view of semantics as essentially dealing with context-independent meaning, but not necessarily determining fully propositional sentence meanings. However, an alternative view is possible, in which the determination of fully propositional sentence meanings is taken to be an essential characteristic of semantics. Under such a view, semantics cannot be assumed to be only concerned with context-independent meaning and, accordingly, the differentiation between semantics and pragmatics has to be based on some other criterion. In fact, there are approaches which assume that what the semantic component returns as output is not some sub-propositional semantic form, but rather, as traditionally assumed, a full proposition, that is, something which is truth-evaluable. Two such approaches are those suggested by Borg (2004b) and by Cappelen and Lepore (2005), respectively.27 28 Thus, Emma Borg (Borg 2004b,a, 2007) argues for what she calls a minimal semantic component which, however, retains one of the most characteristic properties of the traditional approach: Borg claims that what the semantic component returns as output actually is fully propositional/truth-evaluable, although it may not be identical to the proposition expressed by a particular utterance of the senpart of the verb’s semantic form – and Stanley probably would do so in any case – if it is understood as expressing a particular type of event. 26 Thus, compare this to Levinson’s M-heuristic (more on which in section 4.1.3): what is said in an abnormal way, is not normal (marked messages indicate marked situations). 27 Strictly speaking, Stanley’s approach also falls into the category of approaches discussed in this section. Thus, the characterisations of semantics and pragmatics given here also hold for his approach. However, there are some crucial aspects in which his general approach differs from Borg’s and Cappelen and Lepore’s, which, I think, justify treating the respective approaches separately. 28 Bach’s approach does not fit in this section, because, although he also assumes that semantics is not concerned with context-independent meaning exclusively, he does not require of sentence meaning even relative to a context of utterance that it be fully propositional. 116 Utterance Meaning and the Literal/Non-literal Distinction tence/linguistic string in question. Also, like Stanley, she assumes that only syntactically mandated processes take place when determining the truth-conditional content, or sentence meaning, of an utterance. Thus, there are no unarticulated constituents and there is no such process as free enrichment at this stage in the interpretation process. In fact, she denies that speaker intentions play any role whatsoever at this stage. Cappelen and Lepore (2005)’s approach differs from Borg’s, and from Stanley’s, in that they deny the context-sensitivity of any other than the expressions in, as they call it, the ‘basic set’, that is, such expressions as traditionally were considered context-sensitive (cf. Kaplan 1989b). Thus, while they also only allow linguistically mandated processes to take place when building up the semantic content of a sentence, they argue that there are considerably fewer such processes than assumed by dual pragmatists29 or contextualists, as they call them. Generally, Borg argues that there are (only) two tasks that an appropriate formal semantic theory has to fulfill. On the one hand, it has to capture how the meaning of a complex expression is determined on the basis of the meanings of its component expressions and the way they are combined, on the other hand it should make clear the relations that hold between individual complex expressions, for example, which inferential relations possibly hold between them. Borg claims that other possible tasks one might expect a semantic theory to fulfil actually lie beyond the reach of an appropriate formal semantic theory, such as explaining our communicative abilities, or being ‘. . . sensitive to the kinds of epistemic relations we bear to objects . . .’ (Borg 2004b, p. 2). From what Cappelen and Lepore say about the role of semantics, it can be assumed that they generally share this view. Interestingly, both approaches agree with dual pragmatists that in order to deliver the fully propositional semantic content of a sentence/utterance, contextual information has to be considered. Thus, neither Borg nor Cappelen and Lepore draw the dividing line between the semantics component and pragmatics along the traditional dichotomy ‘context-independent vs. context-dependent’, rather, what differentiates between semantics and pragmatics is the nature of the contextual information integrated in the semantic content of a sentence (especially for Borg – as for Stanley) and the conditions under which such integration takes place (for both Borg and Cappelen and Lepore). In other words, what the semantic component deals with are sentence-types relativised to a context of utterance. However, as mentioned already, for Borg, the contextual information used in determining the semantic content of such a sentence-type is restricted to purely objectively available data, such as information concerning the speaker and addressee of the uttered sentence as well as the time and place of utterance. Information that requires theory of mind reasoning, such as for example the possible intention behind the speaker’s uttering the particular sentence he did, are not part of this kind of contextual information, as this is based on subjective data, resulting from abductive mind-reading processes. This is another aspect in which Borg’s approach differs from Cappelen and Lepore’s. While Borg rules out any involvement of such non-demonstrative information, Cappelen and Lepore seem not to mind that such information is necessary when, e.g. determining the actual referent of some indexical expression. However, what is 29 Borg’s term for the proponents of approaches such as put forward by Sperber and Wilson, Carston or Recanati. The Problem of Characterising the Level of Utterance Meaning 117 problematic about this latter viewpoint is that it allows non-objective information to figure in determining the semantics of an expression, when the assumption is that semantics is part of our general language system and as such only involves processes that are formal in nature, syntax-driven and computational. To repeat, the nature of the semantic component is usually taken to be such that it reflects (a particular part of) our knowledge of language. Reasoning about plausible intentions behind a speaker’s use of a particular expression does not form part of this kind of knowledge. For Borg in particular, there is another reason for not assuming that speaker intentions play a role in the determination of semantic meaning, since she assumes of semantics that it exhibits the characteristics of a Fodorian mental module (cf. Fodor 1983). One such characteristic is informational encapsulation. That is, the semantic component only has access to a restricted amount and type of information. Thus, in computing the meaning of a sentence, the interpreter’s assumptions about why the speaker made a particular utterance, are not taken into consideration, as they constitute aspects of knowledge which is not specifically linguistic. As already mentioned, Borg claims that what the semantic component delivers are truth conditions for sentences (relativised to a context of utterance). However, the truth conditions figuring in the semantic component are what she calls liberal truth conditions. In this respect her approach differs from Stanley’s, who seems to assume that semantics delivers the actual truth conditions of a sentence relative to a context. Liberal truth conditions capture the conditions that have to hold for the sentence under consideration to be true. Thus, they differ from actual truth conditions (or the proposition expressed by a speaker’s utterance of a particular sentence), where the latter capture what a speaker intended his utterance to mean and are therefore a solely pragmatic matter. Borg (2004b) says of liberal truth conditions that they may be satisfied in various possible, more specific contexts. Moreover, ‘[a] liberal truth-condition posits ‘extra’ syntactic material . . . only when it is intuitively compelling to do so, or when there is good empirical evidence to support the move.’ (Borg 2004b, p. 230). She gives the following examples of liberal truth conditions. (97) a. If u is an utterance of ‘Jane can’t continue’ in a context c then u is true iff Jane can’t continue something in c. b. If u is an utterance of ‘Steel isn’t strong enough’ in a context c then u is true iff steel isn’t strong enough for something in c. c. If u is an utterance of ‘The apple is red’ in a context c then u is true iff the apple is red in c. Looking in more detail at the example sentence Jill can’t continue: in a context where it is clear that Jill did not pass enough university course units to qualify for admission to second year study, it is plausible that what the speaker intended her utterance to mean is something along the lines of (98c). However, the semantic content of the uttered sentence is simply (98b). (98) a. Jill can’t continue. b. Jill can’t continue something. c. Jill can’t continue university education. 118 Utterance Meaning and the Literal/Non-literal Distinction Thus, actual truth conditions capture the conditions under which the utterance meaning30 under consideration would be true, where utterance meaning is not identical to sentence meaning relativised to a context of utterance, as the determination of the former may rely on non-objective contextual information, more specifically, a consideration of possible speaker intentions. Thus, while speaker intentions for Borg do not figure in the semantic meaning of an utterance, they do already play a role when moving to its utterance meaning. In this sense, everything that goes beyond the semantic content of a sentence may be termed non-literal meaning, whereas the semantic component delivers a sentence’s literal meaning (its liberal truth conditions) – this is how Borg uses the terminology. Whereas utterance meaning may include unarticulated constituents, resulting from processes of free enrichment, Borg argues that sentence meaning does not allow for unarticulated constituents as the semantic component only deals with syntactical objects. Thus, variables that are given in the syntactic structure of a sentence do figure in determining the semantic content of a sentence, but purely pragmatically supplied constituents, naturally, do not. Note that whereas Borg allows for covert syntactic constituents in the case of continue and enough, Cappelen and Lepore argue that these are not context-sensitive expressions in the sense of those in the ‘basic set’. That is, Cappelen and Lepore are even more strict about which expressions may be taken to count as being contextsensitive as Borg is, who cautions that covert syntactic constituents should only be assumed if there is empirical evidence of a very strong intuition that it is appropriate to do so. What this means for Cappelen and Lepore’s approach is that, in a sense, the semantic content of a sentence is taken to be even more minimal, at least in one sense, than assumed by Borg. Thus, Cappelen and Lepore claim that the truth conditions for a sentence such as Jill can’t continue or Steel isn’t strong enough simply are as follows31 . (99) a. If u is an utterance of ‘Jane can’t continue’ in a context c then u is true iff Jane can’t continue in c. b. If u is an utterance of ‘Steel isn’t strong enough’ in a context c then u is true iff steel isn’t strong enough in c. However, as Recanati (n.d.) points out, sentences such as the above invite the question ‘for what?’ which only gets answered once one takes the context of utterance into account. Nevertheless, Cappelen and Lepore insist that the truth conditions for such sentences are as given above and that they are propositional. However, as Bach (2006b) comments ‘Cappelen and Lepore evidently assume that producing a T-sentence automatically provides a truth condition. [. . .] After all, if a sentence is semantically incomplete, the corresponding T-sentence will be incomplete too.’ (Bach 2006b, p. 4). Furthermore, both Recanati as well as Bach criticise that Cappelen and Lepore – other than providing T-sentences such as above – do not explain what the propositions such sentences as Jane can’t continue or Steel isn’t strong enough express, are. Cappelen and Lepore (2005) in turn argue that that’s beyond 30 Borg (2004b) does not use the term utterance meaning, but rather sticks to the traditional Gricean term what is said. 31 Note that throughout their book Cappelen and Lepore (2005) ignore tense, thus, their version of the truth-conditions for these sentences actually do not refer to a context of utterance. The Problem of Characterising the Level of Utterance Meaning 119 the semanticist to tell and something that has to be clarified by metaphysics. However, if that is so then one simply has to take the authors’ word for it that the given truth conditions are actually such that they allow one to evaluate the truth of the respective sentence (for further discussion, see Bach n.d.b,n, Cappelen and Lepore n.d.b,n). In addition, Borg (2007) suggests that for Cappelen and Lepore’s account, preserving the law of non-contradiction is problematic.32 For instance, consider the sentence ‘Jill is ready’: according to [Cappelen and Lepore (2005)] this expresses the minimal proposition that Jill is ready and presumably the sentence expresses something true in any situation in which Jill is ready. On the other hand, the sentence ‘Jill is not ready’ expresses the minimal proposition that Jill is not ready and this presumably expresses something true in any situation in which Jill is not ready. Obviously however there will be innumerable situations in which both sentences are true together (for instance, where Jill is ready to go to the party but not ready to take the exam). Yet, since, for [Cappelen and Lepore (2005)], the second sentence is just the negation of the first, this looks problematic. Another point Borg (2007) raises is that Cappelen and Lepore (2005) seem to be exclusively interested in ‘bottom-up’ context-sensitivity, that is, the type of contextsensitivity exhibited by the members of the ‘basic set’. As she argues, from their characterisation of radical vs. moderate contextualists, Cappelen and Lepore do not seem to recognise a difference between syntactically triggered context-sensitivity and what might be called ‘top-down’ context-sensitivity. The latter is manifested in the claim by, e.g., RT that even with sentences which do express full propositions (of the type You are not going to die) some kind of pragmatic enrichment is necessary in order to determine the proposition that was intended by the speaker. However, Cappelen and Lepore (2005) classify RT as a form of radical contextualism, which they characterise as follows: Radical contextualists: claim that every expression of a natural language is context sensitive (Cappelen and Lepore 2005, 5-6) Borg (2007) claims that this characterisation does not apply to RT or other approaches categorised as radical contextualist, at least not if context-sensitivity is understood in the ‘bottom-up’ sense. As Borg (2007) argues, this can be seen by the fact that RT allows for processes such as free enrichment, which are specifically characterised as introducing material into the semantic form of an utterance which is not syntactically triggered. In other words, the material free enrichment adds does not constitute the value of some context-dependent variable that forms part of some linguistic expression’s semantic form. However, considering RT’s view of lexical meaning as simply constituting a pointer to the meaning an expression will be assigned on a particular occasion of use, Cappelen and Lepore (2005)’s characterisation of RT as radical contextualist does not seem totally inappropriate. 32 For a similar concern cp. Pagin and Pelletier (2007). 120 Utterance Meaning and the Literal/Non-literal Distinction Turning to the notion of what is said, Borg’s view of semantics as delivering sentence meaning relativised to a context of utterance actually corresponds to the traditional Gricean characterisation of the level of what is said. However, this is not how Borg views this latter notion, nor do Cappelen and Lepore. As should have become obvious, both Borg as well as Cappelen and Lepore clearly take what is said to be pragmatic33 although their characterisations differ in detail. Borg takes what is said to be pragmatic principally because it involves such non-objectively obtained component parts as the intended referent of indexical expressions like that, whereas this particular type of information is allowed to figure on the level of sentence meaning relativised to a context of utterance by Cappelen and Lepore. However, both approaches have in common that they assume that what is said, in contrast to sentence meaning may include unarticulated constituents. Thus, what is said is part of speaker meaning, where the latter is independent of sentence meaning. That is, what a speaker means by making a particular utterance may be totally independent of what the sentence he utters means. In contrast to Borg, Cappelen and Lepore (2005) claim that the semantic content of a sentence, actually does form a proper subpart of the content of the speech act carried out by the speaker of that sentence. However, as Borg (2007) points out, this assumption is problematic. . . . we should note that speakers could normally only count as asserting minimal propositions given a quite specialised notion of assertion, one which does not rely on speakers being consciously aware of what they are asserting. For it is clear that the speaker who looks in the fridge and says ‘There is nothing to eat’ would be surprised to learn that they had asserted the quite general proposition that there is nothing to eat (in the universe). (Borg 2007, p. 15) And, as Borg (2007) remarks further, there are of course cases in which the speaker clearly does not want to be taken to assert the minimal semantic content of the sentence she utters, e.g., cases of irony. It should be noted, however, that Cappelen and Lepore (2005) do not claim that speakers in carrying out a particular speech act actually assert the minimal semantic content. It is important for us (a) not to identify the proposition semantically expressed with the propositions asserted (or said), and (b) for that reason not to reserve the label ‘what is said’ for the proposition semantically expressed. (Cappelen and Lepore 2005, p. 150) This would mean, however, that speech act content does not only capture what the speaker intentionally asserted and intentionally conveyed, but may also include meaning aspects that he distinctly does not want to assert. It is questionable whether 33 Remember that for Borg as well as Cappelen and Lepore pragmatic does not simply mean context-dependent, as they assume of semantics that it deals with context-dependent meaning aspects at least up to a certain degree. Moreover, and as mentioned above, they use the term mainly to pick out Grice’s functional characterisation of what is said, namely that it is the level of meaning that is the basis for further pragmatic inferences (e.g. my utterance meaning). The Problem of Characterising the Level of Utterance Meaning 121 this is desirable. On the one hand, in cases as given above by Borg, this would mean that the speech act content may be contradictory. On the other hand, in those circumstances in which the addressee has no particular difficulty in understanding what the speaker said and meant, there is no reason to assume that he, nevertheless, takes the speech act content to consist of contradictory information. As Borg states, Cappelen and Lepore are led to this particular assumption because they claim that, contrary to assumptions made by contextualists, the minimal semantic content or minimal proposition of a sentence does have a unique role to play in the comprehension process. It can serve as a ‘fall back’ in those cases where for some reason it is not clear what the speaker intended (cf. Ariel’s (2002) characterisation of linguistic meaning, discussed in section 5.1.2). The authors argue that assuming that minimal semantic content is part of any possible speech act content the sentence in question might be used to express has the advantage that in cases where only little is known about the situation in which an utterance is made or about the speaker of that utterance, a hearer may still be able to at least capture the minimal semantic content. Borg, however, argues against this view. She mentions a number of experiments which show that it is likely that the minimal proposition of a sentence may not even get processed during comprehension, namely in such cases, where pragmatics operates locally. It should be noted that Borg does not want to deny that semantic content may play the role of a ‘fall back’ in particular cases. What she wants to deny is that it is therefore necessary to assume that semantic content need be part of speech act content. Semantics lexical meaning semantic composition saturation processes liberal truth-conditions fully-propositional sentence meaning rel. context free enrichment ... actual truth-conditions + narrow context what is said + broad context basis for further pragmatic inferences conversational implicature speech acts + broad context ... Pragmatics what is meant Figure 3.8: Borg’s view Thus, although Borg’s and Cappelen and Lepore’s approaches differ in a number of details, there still are some basic assumptions they have in common. Both approaches assume that the semantic component delivers the truth-conditions of the sentence used in an utterance. That is, the results of semantic composition are not sub-propositional forms, but full propositions which are truth-evaluable. However, these truth conditions are in some sense liberal in that they do not necessarily reflect what the speaker of an utterance intended to express by using a particular sentence. Thus, the truth-conditions determined for sentences relative to contexts 122 Utterance Meaning and the Literal/Non-literal Distinction Semantics lexical meaning semantic composition saturation processes liberal truth-conditions fully-propositional sentence meaning rel. context free enrichment ... actual truth-conditions + broad context what is said + broad context basis for further pragmatic inferences conversational implicature speech acts + broad context ... Pragmatics what is meant Figure 3.9: Cappelen and Lepore’s view are not identical to the truth-conditions of the utterance meaning of that sentence. The advantage of such an approach to the nature of the semantic component is that it can be assumed to be rather restricted in its operation, while still fulfilling the role it traditionally was supposed to have, namely to account for the logical relations holding between particular (simple as well as complex) expressions. However, a disadvantage is that it is unclear, which role such minimal propositions play in the overall interpretation process. As mentioned above, it seems that they do not necessarily have to be computed during the interpretation of a particular utterance. Furthermore, they are actually irrelevant (unless they correspond to the actual truth-conditions), in a sense, because what the hearer is normally interested in is the actual not the liberal truth-conditions of some utterance. Moreover, it was assumed that they may serve as a kind of fall back in the case of misunderstanding or insufficient information concerning the context of an utterance, etc. However, there seems to be no reason why a hearer trying to reconstruct what the speaker might have intended to express should fall back to this particular level of meaning. After all, in a situation in which the hearer is not sure about the contextual conditions under which an utterance including indexicals was uttered, the semantic form that he falls back on is going to be sub-propositional in form. Note also that this is not just so in cases of misunderstanding but also holds in all such situations in which the speaker utters a sub-sentential syntactic form. As mentioned already, in the case of sub-sentential utterances, the semantic component simply cannot provide a truthevaluable semantic form just on the basis of the expressions used in the utterance and their possibly attending context-sensitivity. Thus, liberal truth conditions may not actually have any role to play in the interpretation process after all. Borg would concur in this conclusion. However, she argues that although liberal truth conditions may not have a role to play in the actual interpretation process, they are still an important and necessary level of meaning, as they establish the inferential relations holding between expressions in a language. As mentioned already in section 2.3.3, the question is whether this type of inference is actually useful at this level. On the one hand, hearers draw inferences that matter to them in communication based on The Problem of Characterising the Level of Utterance Meaning 123 the actual truth conditions of an utterance. On the other hand, capturing the logical relations between particular readings of particular expressions within the semantic component seems only possible if the lexical meanings of expressions are taken to be full readings. However, we saw that there are arguments against generally assuming full-fledged readings as part of the lexical entries of linguistic expressions. Thus, if semantic meaning is taken to be characterised by a high degree of underspecification, then capturing the logical relations between particular full readings should not be viewed as a task of linguistic semantics anyway, but possibly rather a property of the language of thought (cp. Burton-Roberts 2005). 3.2.4 Minimal Proposition vs. Proposition Expressed Looking again at the difference between minimal proposition (corresponding to Grice’s what is said and Borg’s liberal truth conditions) and proposition expressed (corresponding to RT’s explicature or Bach’s impliciture and Borg’s actual truth conditions): although an utterance of the sentence in (64a) necessarily expresses a proposition, it is questionable whether it is the one the speaker of (64a) intended. Imagine a situation in which a mother says to her small son, who has just had a crash with his bike: (64a) It’ll take time for this wound to heal. It is more likely that what the speaker of (64a) wanted to express by her utterance is the proposition that ‘it’ll take a considerable amount of time for the contextually relevant wound to heal’. This is especially important considering that what is said is supposed to be the basis for further inferences that lead to what the speaker actually meant by her utterance, that is, the basis for conversational implicatures. However, from the apparent tautology expressed by an utterance of (64a), no further implicatures seem plausible. In contrast, depending on the particular contextual circumstances, from the speaker intended proposition expressed ‘it’ll take a considerable amount of time for the wound to heal’ further inferences may be drawn. For example, the child might draw the inference that he will not be able to ride his bike together with his friends the next day. In consequence, it has been claimed (specifically by RT proponents) that the minimal proposition or Gricean what is said by an utterance, actually does not have any role to play in the interpretation process. In fact, this level of meaning is claimed to not be psychologically real. Thus, if what is said has no psychological reality, the assumption is that it will not be consciously available to interpreters. There are a number of experiments that have tested that claim (e.g. Gibbs and Moise 1997, Nicolle and Clark 1999, Bezuidenhout and Cutting 2002). Thus, Gibbs and Moise (1997) report a number of experiments they carried out in order to verify whether what is taken to be an utterance’s minimal proposition (or what is said), is what naive speakers actually do take to be what the speaker of that utterance has said. Thus, they assume Recanati’s availability principle34 and 34 ‘In deciding whether a pragmatically determined aspect of utterance meaning is part of what is said, we should always try to preserve our pre-theoretic intuitions on the matter’ (Recanati 1989, p. 310). 124 Utterance Meaning and the Literal/Non-literal Distinction accordingly were interested in testing naive speakers’ intuitions on the matter of what is said and whether they are able to consciously recognise a level of meaning below the level of communicative sense or what is meant.35 Thus, their first experiment asked subjects to choose one of two given paraphrases of a particular utterance they were presented. One such utterance is that given in (100a), where subjects were asked to choose either (100b) (the minimal proposition) or (100c) (the proposition expressed) as what they thought best captured what the speaker said. (100) a. Jane has three children. b. Jane has at least three children but may have more than three. c. Jane has exactly three children. The hypothesis was that if interpreters use pragmatic information in order to determine what the speaker said with his utterance, they should choose the enriched paraphrase. This hypothesis was verified by the results of that experiment, as subjects chose the enriched paraphrase significantly more often as what they thought the speaker had said with his utterance. In a second experiment, Gibbs and Moise (1997) checked whether subjects possibly do access the minimal proposition during their interpretation, but simply are not consciously aware of that fact. Thus, before presenting subjects with the actual data, they were informed about the difference that can be made between what a speaker says and what he implicates by making an utterance. Although in this experiment, subjects chose the minimal proposition as what they thought best reflected what the speaker said slightly more often, overall they still preferred the enriched proposition over the minimal proposition. A third experiment tested whether naive speakers are actually able to differentiate between what a speaker says and what he implicates by his utterance. Thus, in this experiment, the critical sentences were preceded by short contexts which suggested certain implicatures as arising from the speaker’s utterance of the critical sentence. (101) a. Bill wanted to date his co-worker Jane. But Bill really didn’t know much about her. Being a bit shy, he first talked to another person, Fred. Fred knew Jane; fairly well. Bill wondered if Jane was single. Fred replied, Jane has three children. b. Jane has exactly three children, but no more than three. c. Jane is married. Once again, participants were asked to chose the paraphrase that they thought captured what the speaker had said. This time, one of the paraphrases included the suggested implicature, the other the enriched form of what is said. As before, subjects overwhelmingly chose the enriched paraphrase over the paraphrase that included the implicature, which shows that they are able to differentiate between a level of meaning that does not include implicatures and one that does. Furthermore, 35 Note that Gibbs and Moise (1997), following RT, assume that GCIs are part of what is said rather than ‘proper’ implicatures which are only part of what is meant by an utterance. The Problem of Characterising the Level of Utterance Meaning 125 it shows that subjects in the first two experiments did not simply choose the most enriched paraphrase available. The general conclusion that can be drawn from the results of this experiment is that, apparently, the level of meaning representend by the minimal proposition expressed by an utterance is one which speakers do not consciously make use of when they interpret an utterance. Although the experiments do not prove that speakers do not access such a level of meaning at all, from a psychological point of view, the level of minimal proposition seems to have no role to play in the overall interpretation process. It is not consciously accessed by speakers, nor is it a helpful level of meaning when it comes to hearer’s inferring implicated aspects of meaning. Rather, what hearers consciously take a speaker to have said is a more enriched proposition, the proposition expressed. Moreover, subjects choosing as what is said by an utterance the enriched proposition that included what traditionally are called GCIs shows that these pragmatically determined aspects of meaning are part of the level of what is said rather than what is meant. This is especially apparent when looking at Gibbs and Moise (1997)’s third experiment, which tested subjects’ ability to differentiate between those two levels of meaning. However, Nicolle and Clark (1999) point out quite a number of critical aspects concerning Gibbs and Moise (1997)’s experiments. For example, they are concerned that Gibbs and Moise (1997)’s data includes quite a number of utterance types which may not actually be amenable to a minimal proposition analysis. Moreover, they criticised Gibbs and Moise (1997)’s method of instructing subjects to choose what they thought best captured what a speaker said over the Gricean characterisation of what is said they were informed of in the second experiment. Most importantly, Nicolle and Clark (1999) claim that it is not clear whether subjects, especially in the third experiment, actually did distinguish between the two levels of proposition expressed and what is meant or whether their choices may have been influenced by some other criterion. Thus, they put forward the hypothesis that speakers choose that paraphrase which is most similar with respect to the way in which the original utterance achieved relevance. Using an original example from Gibbs and Moise (1997), given in (101) above, they explain what this underlying criterion means for subjects’ selection of the best paraphrase of what the speaker said (in a nontechnical sense). Thus, they point out that on the basis of the explicature of the critical utterance together with contextual information given in the preceding text, the addressee computes the intended implicatures of that utterance, which may include any of the following. ‘Jane is married; Jane has a partner (but is not necessarily married); Jane is experienced with men (whereas, Bill is too shy even to approach her, hence they may not be compatible); Bill should only date Jane if he likes children; Jane has family responsibilities and so may not want to enter a relationship’ (Nicolle and Clark 1999, p. 350). Since (101c) only expresses one of the possible communicated implicatures, it does not qualify as an ‘. . . optimally relevant paraphrase’ of the original utterance (Nicolle and Clark 1999, p. 350). Thus, speakers choose the paraphrase which most likely allows the same amount of cognitive effects as the original utterance. If, in a particular 126 Utterance Meaning and the Literal/Non-literal Distinction instance, this paraphrase happens to be an implicature rather than an explicature of the utterance, subjects will choose the implicature paraphrase. Thus, consider example (102). (102) a. An informal five-a-side football match had been arranged for Saturday morning at 11:30h. That morning was warm and sunny, and over twenty people showed up. ‘Can we play two five-a-side matches at the same time?’ asked Steve. John answered, ‘Billy’s got two footballs’. b. Billy has got exactly two footballs and no more than two. c. There are enough footballs to play two matches. The relevant cognitive effect of John’s answer in this particular situation is, of course, that there are enough footballs to play two matches. Although there may be other inferences that may be derived from John’s utterance (e.g. Billy is an organised person, Billy likes football), in the given context it seems unlikely that they would be relevant. Thus, the second paraphrase makes the relevant cognitive effect immediately accessible. Therefore, the assumption is that subjects presented with an example such as (102) will choose the second paraphrase as best reflecting what the speaker had said. This hypothesis was tested in a first experiment, which had three conditions. In the first condition, subjects were instructed to choose the paraphrase they thought best reflected what the utterance said. In condition two, subjects were explicitly asked to choose the paraphrase that best captured what the words the speaker used mean, taking into account that naive speakers use the phrase ‘what is said’ differently from its theoretical use. In the last condition, subjects were asked to choose the paraphrase that best reflected what the speaker wanted to communicate. The results of all three conditions show that speakers mostly choose relevant implicatures over enriched explicatures, even when they are instructed to determine what the speaker had said. Moreover, the results were similar across the three different conditions, suggesting that subjects did not differentiate between what is said by an utterance, what is meant by the speaker of an utterance or what the words used in an utterance mean. Thus, Nicolle and Clark (1999)’s results refute the assumption that naive speakers intuitively differentiate between the two theoretical levels of meaning what is said (or, more accurately here, explicature) and what is meant.36 Rather, the criterion by which they determine the best paraphrase of a speaker’s utterance is the criterion of relevance. That is, they choose the paraphrase that is most similar to the original utterance in the amount of cognitive effects that it allows the hearer to gain from it. Having said this, Nicolle and Clark (1999)’s and Gibbs and Moise (1997)’s results both show that speakers do not usually opt for the original Gricean level of what is said, that is, they do not decide on the minimal proposition as what is said by some utterance. However, as pointed out by Bezuidenhout and Cutting (2002), this may simply be due to the fact that they concentrate on what is most relevant in a context and often that is not what, strictly speaking, has been said. Thus, although in the experiments ‘. . . the minimal paraphrase was rarely chosen, this result by 36 Cp. Bezuidenhout and Cutting (2002),who replicated Nicolle and Clark (1999)’s results. The Problem of Characterising the Level of Utterance Meaning 127 itself does not show that hearers do not recover the minimal proposition at some level of processing’ (Bezuidenhout and Cutting 2002, p. 443). In a last experiment, Nicolle and Clark (1999) tested whether naive speakers can be made aware of the difference between theoretical what is said and what is implicated, although their choice of paraphrases in the experiments before seems not to have been based on an intuitive differentiation of these levels. Thus, as in Gibbs and Moise (1997)’s experiment, subjects were given a tutorial on the respective differentiation. They were then presented the material from the first experiment. However, in contrast to Gibbs and Moise (1997)’s experiment, subjects here were explicitly asked to make their choices based on the technical definitions of what is said and what is implicated. The results show that speakers can be made aware of the technical difference, although normally, they do not understand what is said in the technical sense. Thus, as shown already by Gibbs and Moise (1997)’s experiment, naive speakers seem to have quite a different understanding of the expression ‘what the speaker said’ from what is captured by the technical definition. What is more, none of the experiments in Nicolle and Clark (1999)’s study verify Recanati’s availability principle, that is, speakers do not seem to have reliable intuitions concerning the technical differentiation between what is said by an utterance in contrast to what is meant. Therefore, it does not seem very helpful to rely on intuitions in characterising the levels of what is said and what is meant.37 Thus, the fact that subjects in Gibbs and Moise (1997)’s experiment chose the enriched proposition including so-called GCIs as what is said by the speaker, may be due to the particular design of the experimental material used. In other words, the results of that experiment may in fact tell us nothing about whether GCIs belong to the level of what is said/explicature or whether they are genuine implicatures and as such belong to the level of what is meant. Be that as it may, Nicolle and Clark (1999)’s results do corroborate RT’s assumption that the processes that derive explicatures are not so very different from those deriving implicatures. If this assumption is correct, one would actually predict that it should be difficult to find a particular level of meaning which includes explicatures but does not include implicatures. However, Bezuidenhout and Cutting (2002) claim that neither the experiments reported in Gibbs and Moise (1997) nor those in Nicolle and Clark (1999) actually tell us what exactly happens during the interpretation process, since both are only concerned with the products of that process. More specifically, they do not allow us to draw conclusions about potential stages in the overall interpretation process, and therefore, do not provide an answer to the question of whether or not a minimal level of meaning such as Grice’s notion of what is said plays a role during interpretation. In order to shed some light on what exactly is going on during interpretation, Bezuidenhout and Cutting (2002) designed the following experiment. They constructed little stories in which the last sentence was the target sentence. The stories were constructed so as to either bias a minimal or an enriched interpretation of the target sentence. Time taken to read the target sentence was measured, as well as the 37 Note that this also makes the appropriateness of Recanati’s distinction between different types of non-literal meaning doubtful (more on which below), since the distinction again is based on the availability principle. For a general critique of assuming such a principle, see e.g., Bach (2001b). 128 Utterance Meaning and the Literal/Non-literal Distinction time taken to make a match/mismatch decision on a sentence that was shown after the target sentence. This sentence never matched the target sentence, but rather corresponded to the Gricean, and thus minimal, notion of what is said by the target sentence. The stories were presented line by line on a computer screen and subjects had to press a button for the next line to appear. After the target sentence, they saw a sentence of which they had to decide whether it exactly matched the respective target sentence they had seen just before. Bezuidenhout and Cutting (2002) tested the predictions made by three different models of processing for target sentence reading times in the two contexts. According to the Literal-First Serial (LFS) model, target sentences in minimal contexts should be read faster than in enriched contexts, since their minimal proposition is computed in both contexts and in the minimal context it is the only proposition computed. This is in contrast to the Local Pragmatic Processing (LPP) model, which assumes that computation of minimal propositions is not necessary, but only takes place if warranted by the context. Thus, reading times for target sentences should be equivalent for the two contexts. The third processing model tested is the Ranked Parallel (RP) model, which assumes that all interpretations are determined in parallel, but that the interpretations are ranked in terms of their accessibility. This model assumes that enriched interpretations generally are higher ranked and thus more accessible than minimal interpretations. The results show that subjects took longer reading the target sentences in those contexts which biased a minimal interpretation, than in the contexts which biased an enriched interpretation. As regards the time taken for the match/mismatch decision, whether or not the preceding context was biasing a minimal or enriched interpretation of the target sentence did not make any significant difference. Generally, Bezuidenhout and Cutting (2002) interpret their results as supporting the RP model of the interpretation process, in which both the minimal as well as the enriched interpretation are determined in parallel, but where the enriched interpretation is more accessible than the minimal one. Since the enriched interpretation is more accessible, it interfers with the minimal one in the contexts that are biased towards the latter. However, as the authors themselves point out, there are some problems with such an approach. Thus, that the enriched interpretation generally is more accessible than the minimal one is a questionable assumption. Moreover, it is not clear whether there is only one single highly accessible enriched interpretation in a particular situation, which seems to be implied by such an approach. Finally, the authors also are concerned that their match/mismatch task actually is inconclusive concerning the question of whether during interpretation a minimal proposition is accessed (cp. Garrett and Harnish 2008, p. 71). As already mentioned, the time taken to make a decision did not significantly differ for the two types of contexts. As Bezuidenhout and Cutting (2002, p. 454) state, ‘[i]t is necessary to have some baseline time for this sort of match/mismatch judgment task, to show that there was in fact the slowing in judgment times that would be (indirect) evidence of interference from the minimal interpretation’. Thus, although the authors chose an experimental design of which they argue that it should allow conclusions to be drawn concerning the question of whether a minimal proposition is accessed during the interpretation of some utterance, the result was, at least partly, a null effect. Summary 3.3 129 Summary For a summary of the approaches to the characterisation of utterance meaning considered in this chapter, let us start by recapitulating the relevant aspects of Grice’s level of what is said. Grice took what is said to be a full, and therefore truthevaluable proposition, determined on the basis of narrow contextual information. It is the proposition the speaker asserted and thus can be taken to be committed to. Moreover, it forms the basis for further pragmatic inferences. One problematic aspect of Grice’s characterisation of the processes that lead to what is said is that they may only make use of narrow context. Thus, it has been argued that this might result in what is said actually not being fully propositional. Moreover, even if what these processes determine is a full proposition, it has been claimed that Grice’s what is said actually is not informationally rich enough to form the basis for further pragmatic inferences. That is, if what is said is a proposition at all, very often it is only a minimal one and not what the speaker will be taken to have intended to express. However, it seems that further relevant pragmatic inferences can only be drawn on the basis of the proposition actually expressed by the speaker of a particular utterance. Concerning the determination of the proposition expressed, the question arises whether only linguistically mandated processes should be assumed to operate, or whether one should allow for unarticulated constituents to be integrated by pragmatic processes. RT’s as well as Bach’s solution to the problems Grice’s level of what is said faces is to assume that more processes than traditionally recognised are involved in the determination of the level of utterance meaning, which level RT calls explicature and Bach impliciture. Thus, in order for explicatures/implicitures to get determined, processes are assumed that take into account the broad context of utterance. The explicature/impliciture of an utterance, then, is the proposition the speaker intended to express by his utterance. Its determination involves not only linguistically mandated processes, but also unarticulated constituents. Moreover, it is taken to be the proposition the speaker is committed to by his utterance. In contrast to Bach, RT does not recognise Grice’s what is said as a psychologically real level of meaning. Rather, pragmatic processes apply directly to an utterance’s sub-propositional logical form and determine the proposition expressed by the speaker. Although Bach does recognise Grice’s what is said in that it forms the basis for implicitures, he agrees that the former may be of sub-propositional form, since for its determination only narrow context is considered. What is problematic about the respective characterisations of explicature and impliciture is that they are taken to be cancellable. If explicature/impliciture is the proposition the speaker is taken to have intended to express, the question is how this can be cancelled without contradiction? In contrast to both RT and Bach, Stanley claims that only linguistically mandated processes are at work in the determination of utterance meaning. This means that there are no such things as unarticulated constituents at this level of meaning, only hidden indexicals. Generally, Stanley’s approach can be taken as an attempt to rescue the traditional Gricean notion of what is said, in arguing that particular meaning aspects that can be found at the level of utterance meaning, but are not part of the semantic form of an utterance, are actually the result of the process that 130 Utterance Meaning and the Literal/Non-literal Distinction resolves indexicals. The major critique levelled at the hidden indexical approach is that such hidden indexicals would have to be taken to pervade semantic forms, inspite of the fact that in many cases they will not even be provided a value by the relevant pragmatic process. One advantage of this approach, however, is that it allows a clear definition of the level of utterance meaning or what is said as the level of meaning resulting from the application of saturation processes to the semantic form of an utterance. In contrast, both RT and Bach’s approach suffer from an imprecise definition of explicature and impliciture, respectively, as the ‘development’ of or being based on a semantic form. Both Borg’s and Cappelen and Lepore’s approaches are similar to Stanley’s in that they only assume linguistically mandated processes to be involved in the determination of the truth-conditions of a sentence relative to a context of utterance. However, one major point they differ on is that the former take what is said to be a level of meaning distinct from that of sentence meaning relative to a context of utterance. More specifically, both these approaches assume that the semantics component determines the proposition expressed by a particular sentence in a particular context, where this proposition does not necessarily have to correspond to the proposition intended to be expressed by the speaker of that sentence. Thus, whereas Borg’s approach is similar to Stanley’s in that she assumes that there are more context-sensitive expressions than those in the ‘basic set’, the former differs from the latter approach in that Borg does not allow for information based on reasoning about the speaker’s intentions in making a particular utterance to enter into the determination of the proposition expressed by that sentence. In contrast, Cappelen and Lepore do not rule out that speaker intentions might play a role in determining the proposition expressed by a sentence, however, they argue that the set of context-sensitive expressions as such is restricted and consists only of those in the ‘basic set’. In both approaches, what the semantic component determines are liberal as opposed to actual truth conditions for utterances. The question here is what role these liberal truth conditions play in the mental life of speaker-hearers and, thus, why they should be determined in the first place. That is, it seems rather plausible that it is the actual truth conditions that a speaker-hearer would (unconsciously) base any further inferences on actually relevant for him. One of the questions concerning Grice’s level of what is said that came up in several of the approaches discussed is whether it is psychologically real, that is, whether hearers actually compute this level of meaning when engaging in utterance interpretation. Recanati’s assumption based on his availability principle is that if hearers access this level of meaning, they should be consciously aware of it and they should be able to reliably differentiate this level of meaning from the level of what is meant, since the latter is assumed to be based on what is said. However, as the results of experiments show, the levels of meaning computed during utterance interpretation cannot be verified by relying on hearers’ intuitions. One of the reasons is that naive speaker-hearers do not understand the term ‘saying’ as envisaged by theorists. Another reason is that the fact that hearers may not consciously be aware of a particular level of meaning as such is not evidence for that level not being accessed at all. Having looked at different approaches to the nature of utterance meaning – under- Summary 131 stood as that level of meaning which forms the basis for further pragmatic inferences – the question of how this level of meaning relates to the revised characterisations of literal and non-literal meaning can be approached. To repeat, from what has been said in chapter 2 concerning the time-course of the interpretation of literal and non-literal meaning, it should be clear that we want literal meaning and non-literal meaning to be present at the same level(s) of meaning. Thus, it is not possible to identify what is said as an utterance’s literal meaning on which any non-literal meaning is based. In fact, the discussion in the present chapter concerning the exact nature of the level of meaning what is said and the kinds of meaning aspects identified as belonging to it make such a characterisation seem even more inappropriate. For instance, intuitively at least, one would probably not want to characterise meaning aspects due to ad-hoc concept formation or free enrichment as literal. One possibility of capturing similarities and differences between literal meaning and non-literal meaning is to assume that of the processes that are involved in determining what is said, a subset can be identified which results in the literal meaning of a particular utterance. In this case, one has to make sure that the processes involved need not be assumed to take place in a particular order, since in that case one might again end up predicting that a particular type of meaning is temporally prior to another. Another possibility is to assume that literal meaning and non-literal meaning can not only be found at the same level of meaning, but that they actually result from the same set of processes. The differences between the two types of meaning would then have to be explained in terms of differences in the information used by those processes.38 Before we can turn to an investigation of these ideas in more detail, however, there are a few more issues that I would like to look at first. One point concerns the treatment of metaphor as a phenomenon arising at the level of what is meant and thus being treated as rather different from such phenomena as metonymy or idioms and more similar to irony and indirect speech acts. Another point concerns the possible treatment of generalised conversational implicatures as already arising at the level of utterance meaning due to the fact that they differ in some crucial aspects from particularised conversational implicatures. Recall that traditionally, one criterion that seems to differentiate phenomena such as metaphor, irony, CI, speech acts from those found at the level of utterance meaning – or rather Grice’s what is said – was that a consideration of speaker intentions is necessary to determine their interpretation. However, as we saw in the present chapter, there are theories that assume processes to contribute to the level of utterance meaning, where these processes already take into account the potential speaker intentions in making the respective utterance. Thus, the question arises whether the theoretical differentiation between the two levels of what is said/utterance meaning and what is meant/communicative sense actually is warranted. These issues are the topic of the next chapter. 38 This idea will be taken up in chapter 5. 132 Utterance Meaning and the Literal/Non-literal Distinction Chapter 4 Utterance Meaning and Communicative Sense – Two Levels or One? The aim of chapter 3 was to characterise the level of what is said/utterance meaning as a preliminary to specifying the conditions that have to hold when some utterance is said to have literal meaning. The question of how what is said differs from what is meant was touched but not elaborated on. However, this is an important question to investigate, since, as with the level of what is said, Grice’s assumptions concerning the nature of and phenomena found at the level of what is meant have not gone undisputed. Thus, recall for instance that at least for metaphor – traditionally assumed to belong to the level of what is meant – we already saw that apparently a prior determination of a literal meaning on which the metaphoric meaning can be said to be based is not always necessary. According to Grice, to get from the level of what is said to the level of what is meant of an utterance, further, implied aspects of meaning get inferred on the basis of the former level. That is, what is meant combines both directly expressed and indirectly conveyed aspects of meaning content (cp. Bach 2006a). More specifically, what the level of what is meant is supposed to capture is the whole of the meaning the speaker of some utterance wanted to convey. In other words, it is speakers who mean something by their utterances.1 Since speaker intentions play an important role in getting to this level, the meaning at the level of what is meant is also often called speaker’s meaning or speaker’s intended meaning. The phenomena traditionally assumed to belong to the level of what is meant importantly include conversational implicatures and speech acts.2 Recall further that Grice originally took metaphor and irony to be instances of conversational implicature and thus to belong to the level of what is meant. However, as mentioned already, there are arguments both against a treatment of metaphor and irony in terms of conversational implicature as well as against generally treating metaphor 1 Remember Grice’s differentiation of the four levels of meaning and the formulation ‘U meant by uttering x that . . . ’. 2 Thus, although, e.g., conventional implicatures also play a role at this level, the discussion will center on the notion of conversational implicature and the phenomena originally subsumed under it. 134 Utterance Meaning and Communicative Sense – Two Levels or One? and irony as involving similar comprehension processes. In fact, there are reasons to assume that metaphor, in contrast to irony, actually belongs to the level of (explicated) utterance meaning. Similarly, of the two general types of conversational implicature identified by Grice – particularised conversational implicature (PCI) and generalised conversational implicature (GCI) – the latter has received a considerable amount of attention leading to the question of whether it actually should be assumed as contributing to the level of what is said rather than to the level of what is meant. As regards PCIs, it has been suggested by RT that their inference is not necessarily based on a fully propositional utterance meaning/explicature, which suggestion calls into question the usefulness/possibility/necessity of differentiating between the two pragmatically determined levels of meaning what is said and what is meant at all. In fact, it has been claimed that there actually is no principled way of differentiating what is said from what is meant (Cappelen and Lepore 2005). Thus, in this chapter, I will review the phenomena originally assumed to belong to the level of what is meant, concentrating on metaphor (section 4.1.1), irony (section 4.1.2) and (mainly generalised) conversational implicature (section 4.1.3). I will only mention a few issues concerning speech acts and here specifically indirect speech acts, since in their treatment the notions of literal and non-literal meaning have been made use of and since they have also been treated as a type of conversational implicature (section 4.1.4). Finally, I will turn to the discussion of the question whether it is actually useful/possible/necessary to differentiate between two pragmatically determined levels of meaning (section 4.2). The overall aim of this chapter, then, is to find out which of the phenomena traditionally identified as belonging to the level of what is meant actually do need a fully propositional utterance meaning as a basis and also how the former level of meaning is characterised. Thus, the discussion here will also, again, contribute to the understanding of the level of utterance meaning, by possibly adding some phenomena to it that are not traditionally assumed to arise at it. Moreover, the question will be pursued what kind of context the individual phenomena discussed require for their interpretation, that is, what types of information are necessary for the interpretation process to determine the various phenomena presented. This is an important question for the discussion in section 4.2 concerning the differentiation of two levels of context-dependent meaning, on the one hand, as well as for determining the relations between the individual phenomena traditionally identified at the two levels, on the other. 4.1 4.1.1 Problematic Phenomena Metaphor Traditional Characterisation and its Problems As already mentioned in section 3.1.1, Grice analysed metaphor and irony similarly in that he assumed that both are cases of a violation of the first maxim of quality: ‘Do not say what you believe to be false’. Thus, recall the treatment of example (58) repeated below. Problematic Phenomena 135 (58) You are the cream in my coffee. The idea is that in case of an utterance of (58) by a speaker, it is clear that the speaker cannot be trying to mislead the hearer, due to the blatant falsity of the proposition expressed by the utterance the speaker uses. Thus, the literal proposition cannot be what the speaker intended to express. Assuming that the speaker is adhering to the co-operative principle, the hearer will try to make sense of his utterance, by drawing an appropriate implicature. For instance, depending on the particular context of utterance of (58), the hearer might infer that the speaker meant to convey that the addressee is very important to him. There are several problems with Grice’s approach to metaphor understanding. First, as already mentioned, his view that what is said is entailed by what is meant forces him to argue that in case of metaphor, the speaker is not actually saying anything, as the proposition expressed by the sentence used is not assumed to be part of what the speaker meant in making his utterance. If that is the case, then it actually seems that metaphor cannot be treated in terms of conversational maxim violation (since the maxims refer to what is said). Second, in Grice’s framework, the hearer is traditionally taken as rejecting the literal meaning of the expressions used in a metaphoric utterance. However, it is arguable that the literal meaning plays an important role in determining the exact non-literal or metaphoric meaning the speaker did in fact intend to convey. Third, and as mentioned several times before, Grice’s treatment of such non-literal meaning aspects as metaphor suggests that in terms of the temporal enfolding of the interpretation process, non-literal meaning aspects will always be interpreted after the potential literal meaning of some complex expression has been processed and rejected as not fitting the particular context of the utterance. We already saw that there is empirical evidence against such a view. Fourth, there are both theoretical considerations as well as empirical evidence suggesting that understanding of metaphor involves processes different from those at play in understanding irony.3 Thus, in this section, I will discuss several alternative approaches to the understanding of metaphor as well as the results of empirical studies aimed at investigating the interpretation of metaphor. Whereas Grice treated metaphor as a phenomenon arising at the level of what is meant, RT takes metaphorical meaning to belong to the explicit meaning expressed by an utterance (Carston 2002b, Sperber and Wilson 2008) and thus, as part of utterance meaning. In fact, RT claims that the interpretation of metaphor does not essentially differ from that of (other types of not strictly) literal language. The underlying process involved in metaphor interpretation is adhoc concept formation, which we have already come across. In a way similar to RT, Glucksberg and McGlone (1999) assume that in (nominal) metaphor interpretation, a ‘new’ category is created by the interpreter on the basis of particular properties of the metaphor vehicle that are taken to be relevant for the particular metaphor topic. This new category, then, is what is attributed of the metaphor topic. Thus, both in RT as well as in Glucksberg and McGlone (1999)’s approach, the literal meaning of the metaphor vehicle plays a crucial role, as it forms the basis 3 This point will be taken up in section (4.1.2) 136 Utterance Meaning and Communicative Sense – Two Levels or One? from which the metaphoric meaning of the metaphor vehicle is inferred. Thus, in contrast to the traditional view of metaphor, the literal meaning of the metaphor is not simply rejected or discarded. This is verified by the results of empirical studies investigating the interpretation of metaphors. Thus, Rubio Fernández (2007) showed that irrelevant literal meaning aspects of a metaphor vehicle are activated initially and remain active for some time before being suppressed. There are also formal approaches to the interpretation of metaphor which aim at offering an explanation of how a metaphorical meaning is determined without compromising the principle of compositionality and generally, the autonomy of the semantic component. Stern (2006, 2009), for instance, treats metaphor interpretation as paralleling that of indexicals and demonstratives. Thus, he assumes a special operator in the semantics of natural languages, which indicates that its argument is to be understood metaphorically. Dölling (2001) proposes an operator that obligatorily applies to all first-order predicates and provides them with structure that can be, but does not have to be exploited in the further interpretation process. Metaphor and The Similarity of Various Types of Meaning Recall that one of the processes assumed by RT to play a role when moving from expression meaning to utterance meaning is ad-hoc concept formation. Thus, this process adjusts the concept lexically encoded by a particular expression, so that the expression is understood in that particular situation as expressing the adjusted concept.4 Sperber and Wilson (2008) argue that it is this general process that underlies the interpretation of various types of not strictly literal language use, metaphor amongst them. Generally, in RT literal meaning as well as different types of non-literal meaning are assumed to be similar in that they result from the same overall interpretation process, which is geared towards achieving maximal relevance. In addition, there are types of meaning which are not easily characterised as belonging to one or the other of the traditional categories, so called loose uses as in (103). (103) a. Holland is flat. b. France is hexagonal. c. I must run to the bank before it closes. Taking utterances of the sentences in (103) as constituting statements, it is not clear whether they should be judged true or false. That is, strictly speaking, Holland, of course, is not flat. Nor is France, strictly speaking, hexagonal. And the speaker of (103c) might utter it in a context where he actually starts walking rather fast in direction of the bank in question, but does not, strictly speaking, run. However, the addressee of such an utterance would not intuitively and consciously assume that the speaker had said something false. Rather, such uses of the terms flat, hexagonal and run are ‘loose uses’ in that the idea or concept they are used to convey does not fully correspond to the strict concept they are taken to encode. That is, the 4 See, however, Stern (2006), who argues against RT’s treatment of metaphor in terms of ad hoc concept-formation and generally the view according to which metaphor is only one point in a continuum of strictly non-literal meanings. Problematic Phenomena 137 concepts they are understood as expressing are the result of the process of ad-hoc concept formation. The existence of such a phenomenon as loose use together with the problematic nature of Grice’s analysis of such non-literal meaning as metaphor and irony as being cases in which the speaker actually only ‘makes as if to say’ something and, thus, in effect would have to be taken to not having said anything at all, Wilson and Sperber (2000) use to argue against the suitability of a maxim of truthfulness. That is, since there apparently are many cases in which speakers make utterances which, strictly speaking, are not literally true and where this fact does not seem to irritate the addressee much, they conclude that interpretation of a speaker’s utterance is not guided by some maxim of truthfulness, but rather by considerations of which interpretation would make the speaker’s contribution most relevant to the addressee at the time and in the situation in which it occurs. Moreover, since, according to RT, what is encoded by the speaker’s utterance only forms the basis for the actual interpretation of what the speaker tried to convey, addressees do not have any particular expectations that the encoded material expressed by the speaker be true. Thus, since under RT’s view, the only principle guiding an addressee in interpreting an utterance is the principle of relevance, one would actually expect different kinds of meaning (i.e. literal, non-literal, ‘loose’, etc.) not to differ from one another in terms of the processes involved in deriving them. ‘The very same process of interpretation gives rise to literal, loose, hyperbolical or metaphorical interpretations, and there is a continuum of cases which cross-cut these categories’ (Wilson and Sperber 2000, p. 245). Consider a situation in which Peter and Mary are planning their next holiday, which is supposed to be a cycling trip. Peter has just warned Mary that he does not feel in very good shape, whereupon Mary replies: (104) We could go to Holland. Holland is flat. This, of course, would be an example of ‘loose use’ of the expression flat. What Wilson and Sperber (2000) claim is that the interpretation process Peter engages in to derive Mary’s meaning is not much different from the process that would have been triggered by a slightly different utterance on Mary’s part, namely one involving metaphor. (105) We could go to Holland. Holland is a picnic. What makes the two interpretation processes similar is that they start from the same assumption, namely that Peter assumes Mary’s utterance to be relevant. Moreover, in both cases, what seems to be involved in interpreting the respective utterances is an adjustment of the concepts expressed by flat on the one hand and picnic on the other. Generally, the resulting ad-hoc concepts might be narrower or broader (or a combination of both) than the encoded concept the process started out from.5 In Sperber and Wilson (2002) the authors compare a metaphorical to a literal interpretation of the sentence in (106), where the important point, once again, is 5 To be precise, actually what is narrower or broader due to ad-hoc concept formation is the denotation of the resulting adjusted concept. 138 Utterance Meaning and Communicative Sense – Two Levels or One? that the interpretation procedure is similar for the two interpretations and where, crucially, in case of the metaphorical interpretation of (106), the literal interpretation of the sentence used does not figure as an intermediate, much less, necessary step. (106) John is a soldier! Thus, imagine the following exchange between Peter and Mary. (107) Peter: Can we trust John to do as we tell him and defend the interests of the Linguistics Department in the University Council? Mary: John is a soldier! Due to the particular context in which Mary makes her utterance, particular aspects of Peter’s mental concept soldier that is activated by Mary’s utterance will be more accessible than others. This difference in accessibility of particular aspects of the soldier concept correspond to a difference in the accessibility of possible implications the interpretation process will deliver based on Mary’s utterance, cf. (108). (108) a. John is devoted to his duty. b. John willingly follows orders. c. John does not question authority. d. John identifies with the goals of his team. e. John is a patriot. f. John earns a soldier’s pay. g. John is a member of the military. According to the relevance-theoretic comprehension procedure, Peter, unconsciously, considers these implications in the order of their accessibility in this particular context. Since the idea is to stop the interpretation process when an interpretation is reached that satisfies the interpreter’s expectations of relevance, Peter will consider the implications down to (108d). Note that Peter, in this particular situation will not consider the implication in (108g), that is, what may be considered the literal interpretation of Mary’s utterance. However, in a different situation he may do. (109) Peter: What does John do for a living? Mary: John is a soldier! Now, Mary’s utterance in this particular context, may lead to a different degree of accessibility for the various aspects that are part of Peter’s mental concept of soldier. Most importantly, it may lead to a difference in accessibility of the implications Peter may draw on the basis of Mary’s utterance, such that in this particular context, the implication (108g) might be the most and (108d) the least accessible implication. Thus, in this particular situation, Peter might stop his interpretation of Mary’s utterance after having considered implications (108g) and (108f). As Sperber and Wilson (2002) note: . . . by applying exactly the same comprehension procedure (i.e. following a path of least effort and stopping when his expectations of relevance are satisfied), Peter arrives in the one case at a metaphorical interpretation, and in the other at a literal one. (Sperber and Wilson 2002, p. 20) Problematic Phenomena 139 Thus, Relevance Theory rejects the traditional view according to which literal meaning and non-literal meaning are the results of different interpretation processes. More specifically, and similarly to Giora (1997, 1999, 2002) and Gibbs (1994), RT claims that the interpretation of the literal meaning of an utterance need not form an intermediate step in the determination of its intended non-literal meaning. However, assuming that both literal meaning as well as non-literal meaning essentially are computed by the same process does not explain why in clear cases of metaphor or metonymy, these types of meaning are perceived as deviating from some underlying literal meaning.6 Moreover, the same problems concerning the nature of encoded concepts and the process of ad-hoc concept formation mentioned already also apply here. Thus, if one assumes that what is encoded by an expression is only a pointer to potential concepts, then the process of ad-hoc concept formation cannot be taken to operate on encoded meaning. Furthermore, if one assumes that concepts are atomic, then it is not clear how the process of ad-hoc concept formation derives a contextually fitting concept from the concept it applies to. Similarly, if concepts are atomic, it is not clear in which sense aspects of it can differ in accessibility. Metaphor and Attributive Categories Glucksberg and McGlone (1999) investigate the question of how people understand such metaphors as in (110). (110) a. My job is a jail. b. Sam is a pig. c. Our love has become a filing cabinet. Consider, for example, (110a). Here, the metaphor topic is my job, whereas the metaphor vehicle is jail. What seems to be asserted by an utterance of the sentence in (110a) is the idea that what is denoted by my job is part of what is denoted by jail, that is, jail is taken to be the superordinate category to which my job belongs. Literally speaking, we know that this is not the case. Nevertheless, Glucksberg and McGlone (1999) claim, what superficially looks like a category-inclusion statement (i.e. 110a) is in fact a category-inclusion statement, only that the properties of the category denoted by the metaphor vehicle are changed such that the metaphor topic may be included in the category denoted by the vehicle. ‘When someone says that “my job is a jail”, job and jail are cast into a common category, viz., situations that are confining, difficult to get out of, unpleasant, etc.’ (Glucksberg and McGlone 1999, p. 1542). In fact, what is ‘normally’ denoted by the expression jail, actually can be seen as the stereotypical subset of the newly created category. Thus, in (110a) jail gives the name to the superordinate category to which both the literal jail as well as the metaphor topic, my job, belong. Glucksberg and McGlone (1999, p. 1543) call such a category that ‘. . . is used to characterize a metaphor topic . . .’ an attributive category. As regards the question of how such an attributive category is created, Glucksberg and McGlone (1999) argue that the properties of metaphor topic and vehicle that 6 Cp. Stern (2006, p. 248, footnote 3), who argues that RT ‘plays down the intuitive figurativity of metaphor by placing it on a continuum with loose use’. 140 Utterance Meaning and Communicative Sense – Two Levels or One? may be relevant in this ‘creation’ have to be inferred by the interpreter. Thus, in contrast to Lakoff (1987), the authors do not assume that the creation of attributive categories or the interpretation of metaphors, generally, relies on the representation of conceptual mappings between a source and a target domain. Rather, the hearer of such sentences as in (110) determines an interpretation of the respective sentence, based on the properties of the metaphor vehicle taken to be relevant for the appropriate characterisation of the particular metaphor topic. Thus, even in the case of an already conventionalised metaphoric vehicle, the properties that will be important in a particular instance of use will depend on the metaphor topic and contextual specifics. The authors’ view of metaphor interpretation is supported by the results of an experiment they carried out, which was intended to investigate whether for their interpretation of cases of metaphor interpreters rely on conceptual mappings assumed to be already established. Thus, subjects were presented three love-metaphors, which they were asked to paraphrase. According to Lakoff’s view, the abstract concept love (the target domain) is understood in terms of more concrete concepts, for instance, as a journey or container (source domains). Specific properties of the source domain are mapped onto specific properties of the target domain. These mappings, it is argued, enable people to interpret such metaphors as in (111). (111) a. Our love is a bumpy rollercoaster ride. b. Our love is a voyage to the bottom of the sea. c. Our love is a filing cabinet. However, the paraphrases subjects offered for the metaphors in (111) do not support such a view. Thus, in the paraphrases for the metaphor in (111a), no journeyrelated references were made. Glucksberg and McGlone (1999) consider the possibility that the meaning of the metaphor vehicle bumpy rollercoaster ride might actually not entail any of the necessary journey-love mappings. This, however, should not be the case for (111b), where explicit reference is made to journey by the expression voyage. Thus, in this case, if Lakoff’s view is correct and subjects interpret the metaphor in terms of the conceptual mappings between the concepts love and journey, one can expect their paraphrases to be consistent. In fact, although some of the paraphrases made reference to journey-related aspects, not all of them did and not in the same way. Similar results were obtained for (111c), which was designed as a non-conventional instance of the conceptual metaphor relationships are containers. Thus, overall the results indicate that interpreters do not automatically and invariably use hypothesised knowledge about domain-to-domain mappings in interpretation. Rather, what seems to guide their interpretation of novel metaphors is the question which properties of the metaphor vehicle can plausibly be attributed to the particular metaphor topic.7 7 For similar results, supporting the attributive category approach to metaphor interpretation, see McGlone (1996). For an attributive category-supporting study investigating the neural correlates of metaphor interpretation, see Stringaris et al. (2007). Problematic Phenomena 141 Empirical Results Concerning Metaphor Interpretation We already looked at a number of empirical studies which suggest that the traditional view of metaphor as being secondary to literal meaning and necessarily non-conventional is mistaken. Thus, recall the results attained by Glucksberg et al. (1982) and Keysar (1989), which indicated that if some sentence can have a metaphorical interpretation, it will be processed, regardless of whether the sentence’s potential literal meaning is false or not and even in cases where subjects were specifically asked to simply judge sentences as to their truth or falsity and, thus, a metaphorical interpretation was assumed not to be necessary. Thus, it seems that the processing of metaphor is not optional as suggested by the traditional approach to the understanding of non-literal meaning, but rather obligatory. This is also assumed by Giora’s graded salience hypothesis. More specifically, recall that she assumes differences in processing depending on the familiarity status of the meaning in question. Thus, for familiar metaphors she assumes that both the expression’s literal as well as its metaphorical meaning are activated initially, regardless of the nature of the preceding context (i.e., whether it is literally or metaphorically biasing) (Giora and Fein 1999). However, as also mentioned already, her results have to be treated with caution, since it is not clear on which basis she differentiated between metaphors on the one hand, and idioms on the other.8 Having said that, Blasko and Connine (1993) carried out a range of experiments in which they tested the influence both of familiarity as well as aptness on metaphor interpretation. The method they used was a cross-modal lexical priming task. Thus, subjects were auditorily presented utterances involving nominal metaphors (e.g., X is Y) and had to carry out a lexical decision task on words presented visually at different time points during the auditory presentation, namely at the offset of the metaphor vehicle (Experiment 1) or 300ms afterwards (Experiment 2). Blasko and Connine (1993) found that for highly familiar metaphors, their figurative interpretation was available immediately at offset of the metaphor vehicle as evidenced by similar reaction times to target words either related to a literal or metaphorical interpretation of the metaphor vehicle, both of which were reacted to more quickly than to unrelated control words. However, for low-familiar metaphors, the metaphorically related target word was more slowely responded to than both the literally related and the unrelated control word. Similar results were obtained with the 300ms interstimulus intervall. In a third experiment, the authors tested whether aptness played a role in the interpretation of low-familiar metaphors.9 The results suggest that it does, as the response times to the metaphorically related targets in the case of highly-apt metaphors were significantly faster than for moderate-apt metaphors. A further point Blasko and Connine (1993)’s results suggest is that the literal meaning of a metaphorically interpreted expression is not simply and not immedi8 See however Schmidt and Seger (2009, p. 385) who investigated the effect of figurativeness, familiarity and difficulty on metaphor processing, using fMRI. They too suggest that ‘. . . [t]he quality of being figurative may not be the crucial factor in determining the neural activation during metaphor processing.’ 9 A post hoc analysis of the second experiment had suggested that aptness might play a role in the case of low-familiar metaphors, but apparently does not have any effect on the processing of highly familiar metaphors. 142 Utterance Meaning and Communicative Sense – Two Levels or One? ately suppressed during interpretation. Generally, the results of empirical studies investigating the role of the literal meaning of a metaphorically used expression give support to the idea that during interpretation, aspects of the literal meaning of the metaphorically used expressions are enhanced, whereas inappropriate aspects are suppressed. Thus, Rubio Fernández (2007) makes explicit reference to both the attributive category and RT’s approach to metaphor. Although the two approaches differ in the specific theoretical assumptions they make concerning what is going on in metaphor interpretation, they actually make similar predictions concerning the activation levels of relevant vs. irrelevant meaning aspects of the metaphor vehicle used in actual metaphoric utterances. Thus, according to the attributive category approach, the interpreter has to construct an attributive category, which ideally the metaphor vehicle stereotypically exemplifies. In order to engage in this creation, the interpreter has to infer the relevant properties of the metaphor vehicle, with respect to the metaphor topic and the specific context. In processing terms, this means that properties of the metaphor vehicle that are relevant are enhanced, whereas properties that are inappropriate or irrelevant are suppressed.10 Similarly, in ad-hoc concept formation, the encoded concept is adjusted by being narrowed and/or broadened. Recalling RT’s characterisation of concepts as consisting of a logical, lexical and encyclopaedic entry, ad-hoc concept formation can be understood to involve the following changes to the ‘organisation’ of elements in the respective entries of a concept. In cases of narrowing, a meaning aspect in the encyclopaedic entry of the concept is ‘promoted’ to the status of content-constitutive, i.e., it is transferred to the logical entry. In contrast, in cases of broadening, a logical aspect is discarded. Rubio Fernández (2007) suggests that in cases of narrowing, the meaning aspect that is promoted becomes and stays activated during interpretation, whereas in broadening, the meaning aspect that is eventually discarded is initially activated but then suppressed. Using the cross-modal lexical priming paradigm, Rubio Fernández (2007) tested this hypothesis. Subjects heard little texts, which always ended in a nominal metaphor of the form X is a Y, Y being the metaphor vehicle and, thus, here the prime. A target word was visually presented at 0, 400 or 1000ms after the word-recognition point of the respective prime and had to be judged as to whether it is a word of the language. Target words either refered to the superordinate category or to distinctive properties of the metaphor-vehicle. In the control condition, the targets where unrelated to the primes. The context in which the metaphor was presented was designed so as to render the superordinate category metaphorirrelevant and the distinctive properties metaphor-relevant. Thus, if the hypothesis that metaphor-irrelevant properties of the metaphor-vehicle are suppressed, while metaphor-relevant features are enhanced is correct, one would expect differences in the reaction times of subjects when presented with the superordinate category target as compared to the distinctive property target during the interpretation process. The results support the tested hypothesis. That is, at 0 and 400ms after wordrecognition point of the prime, both the superordinate category as well as the dis10 Two further experimental studies in which this hypothesis was tested and confirmed are Gernsbacher and Robertson (1999), Gernsbacher et al. (2001). Problematic Phenomena 143 tinctive properties of the metaphor vehicle were equally active. However, at 1000ms inter-stimulus intervall, subjects took significantly longer to judge superordinate category targets than to judge distinctive properties targets, suggesting that the metaphor-irrelevant superordinate category was suppressed during interpretation. This result is corroborated by results on the activation of superordinates in literal neutral contexts, where they are still active at 1000ms ISI (cf. Rubio Fernández et al. 2003). Note that the same paradigm has been used to investigate the activation of the meanings of homonyms, where, depending on the context given one meaning would be intended and the other irrelevant. The results there parallel the results for metaphor interpretation with one important exception. Thus, similarly to metaphor interpretation, in homonymy interpretation both potential meanings of the homonyms are activated initially, but the irrelevant one is suppressed later on. However, this suppression is already manifest at a 200ms ISI. In other words, although both in homonymy and metaphor interpretation suppression plays a role, it does so at different time points. More specifically, Rubio Fernández (2007) argues, whereas in homonymy interpretation, suppression occurs early and thus, in a sense, takes place automatically, in metaphor interpretation, suppression takes place later and seems rather due to attentional processes. She explains the difference as follows. In homonymy interpretation, the potential readings are immediately available to the interpretation process, which simply has to select one as the appropriate reading. In novel metaphor interpretation, the potential readings of the metaphor vehicle are not immediately available, since they are actually only created during interpretation. Thus, irrelevant aspects of the literal meaning of the metaphor vehicle stay activated longer, because the actual metaphorical meaning has to be created before it can enter into the overall utterance meaning. Rubio Fernández (2007, p. 364) suggests that it is ‘possible that metaphor awareness is related to the fact that more attentional resources may be involved in metaphor interpretation as opposed to disambiguation . . .’. It would be interesting to see how the results of a similar study with, in Giora’s terms, familiar or highly salient metaphors as material would fall out. That is, are familiar metaphors similar to homonyms in that both their literal as well as their non-literal meanings are stored and immediately available to the interpretation process, or are they more like novel metaphors in that both their literal and nonliteral meanings are initially activated and retained beyond 400ms ISI, or are they somewhere in between? Formal approaches to metaphor interpretation Stern (2000, 2006, 2009) proposes a treatment of metaphor along the lines of indexicals and demonstratives. Thus, he assumes an operator, ‘Mthat’, in analogy to Kaplan (1989b)’s ‘Dthat-operator’, which lexicalises the potential metaphorical interpretations for a particular expression type. The operator applies at the level of logical form and constrains the interpretation of its argument. Thus, similarly to RT and Glucksberg and McGlone (1999), Stern (2000, 2006, 2009) does not assume that a metaphorical interpretation of some utterance depends on the prior determination of that utterance’s literal proposition, rather, what is metaphorically 144 Utterance Meaning and Communicative Sense – Two Levels or One? interpreted is the metaphor vehicle (which may be simple or complex) and it is the literal meaning of the metaphor vehicle, its character, that plays an important role in the metaphoric interpretation. Stern (2006, p. 262) characterises the ‘Mthat’ operator as follows. . . . when prefixed to a (literal) expression φ, [the operator ‘Mthat’] yields a context-sensitive expression ‘Mthat[φ]’ whose tokens in each context c express a set of properties presupposed in c to be m(etaphorically)associated with the expression φ, such that the proposition < . . . {Mthat [φ]}(c) . . . > is either true or false at a circumstance. Thus, the character of ‘Mthat[φ]’ is similar to the character of indexicals in that it yields different contents when applied to different contexts of utterance. Note that, in order for the context to function as an argument to the character of the metaphorical expression ‘Mthat[. . .]’, it has to be richer than the Kaplanian context for demonstratives and indexicals in that it additionally has to include sets of presuppositions or presupposed properties. In other words, the properties or propositions that are taken to be metaphorically associated with a particular expression, have to be determined before the content of that expression in a particular context can be established. Unfortunately, Stern does not address the question of how this is achieved.11 Obviously, determining the set of m-associated propositions for a particular expression is not a task that can draw on objectively accessible information, as is the case in determining who are the speaker and addressee of an utterance and where and when that utterance takes place. Stern (2009) draws support for his view from the fact that metaphor can be shown to pass the tests on context-sensitivity employed by Cappelen and Lepore (2005) to identify truly context-sensitive expressions.12 For example, one test is the InterContextual Disquotational Indirect Report Test (ICDIR). Here two contexts C and C* are assumed, where the latter differs from the former in some relevant aspect(s). A sentence S is uttered in context C and this utterance of S is disquotationally indirectly reported in context C*. If any such report is true, then that is taken to be evidence for the context-insensitivity of the sentence S. For an illustration, consider the following example, where the assumption is that (112a) and (112b) are uttered in contexts that differ at least in the speaker of the respective contexts. (112) a. Mary: I am too tired to go to the cinema. b. Peter: Mary said I am too tired to go to the cinema. The idea is that Peter by his utterance of (112b) cannot report what Mary has said by using the same sentence Mary did, because in his context I will pick out an individual different from the individual picked out in Mary’s context of utterance. Now consider example (113), which involves a metaphorical interpretation of the predicate is a fish. 11 And he does not seem particularly interested in this question either: ‘From the perspective of the semantics, all that is significant is that they are presupposed, abstracting away from their psychological sources’ (Stern 2009, p. 17). 12 It should be noted, however, that the tests put forward by Cappelen and Lepore (2005) have been claimed to actually not differentiate between the expressions the authors claim are contextsensitive and those of which they want to show that they are not. Cf., e.g., Bezuidenhout (2006). Problematic Phenomena 145 (113) a. Tip: Nixon is a fish. b. John: Tip said Nixon is a fish. The context of Tip’s utterance is the time of the hearings after Watergate in Congress. A journalist asks how the hearings are going and Tip responds with (113a) ‘. . . in a context C in which it is commonly presupposed that fish are slippery, slimy, and hard to catch’ (Stern 2009, p. 6). However, if Tip’s utterance is reported in a relevantly different context, namely, in a context in which different presuppositions hold concerning the metaphorical interpretation of is a fish, then the metaphorical interpretation will in fact be different from that intended by Tip. Thus, if John, who utters (113b), is an experienced fisherman and associates with the metaphorical is a fish ‘. . . the salient presupposition that fish are easy to catch with a little patience and bait’, then the truth-conditions of the embedded sentence (114a) will differ from those of Tip’s original utterance (114b). (114) a. < Nixon, being easy to catch with a little patience and bait > b. < Nixon, being slippery, slimy, and hard to catch > Thus, it seems that metaphorically interpreted expressions behave similarly to such indexical expressions as I in that they do not allow an ICDIR. In both cases, the reason is that the semantic content in the two relevantly different contexts is not identical.13 Stern can explain this behaviour since he assumes that the ‘Mthat’ operator turns the expression that it applies to into a context-sensitive expression, whose content varies with the contexts in which it is used. Thus, both Tip and John use the same expression type, namely that given in (115). (115) Nixon Mthat[is a fish]. However, they make their utterances in different contexts and, thus, the contextual information relevant for the interpretation of the metaphorically used expression differs and leads to the different propositions given above being expressed. To repeat, Stern (2009) assumes that there is a set of properties presupposed in the particular context of utterance to be metaphorically-associated with the argument of ‘Mthat’. Note though that, at least for the example discussed above, the properties that figure in the propositions expressed by the respective utterances of Tip and John seem to me to be properties that are literally associated with the concept fish. This is especially obvious in the proposition taken to be expressed by Tip, which seems to me to be just a metaphorical paraphrase of what is expressed by his utterance, but does not capture the actual proposition expressed. That is, (114b) is no more true in the particular context of utterance than the literal interpretation of Tip’s utterance is. Generally, I think that it is a simplification of what is actually required in metaphor interpretation to assume that the metaphorically-associated properties of a metaphor vehicle are given and part of the context before the interpretation of the metaphorically intended expression even begins. Thus, recall Glucksberg and McGlone (1999)’s assumption that the interpreter’s identification 13 Of course, this argument only works if one accepts that the propositions given in (114a) and (114b) actually are the semantic contents of the respective utterances. See Stern (2009, pp. 9) for a defence of this view. 146 Utterance Meaning and Communicative Sense – Two Levels or One? of the properties of the metaphor vehicle that can plausibly be attributed to the metaphor topic actually is part of the interpretation of a metaphor and cannot be taken to be presupposed. Another critical point concerns the predictions Stern’s approach seems to be making conerning the relative difficulty of metaphor interpretation. That is, since Stern (2000, 2006, 2009) takes metaphors to work similarly to indexicals and demonstratives, one might expect that they are similarly difficult to interpret. However, results from neuropathological investigations go against that expectation. Thus, de Villiers et al. (2007, 2010), for instance, show that autistic people have no great difficulty interpreting indexical expressions, but it is well known that they do have problems interpreting cases of metaphor (e.g. Happé 1993, Gold et al. 2010, Rundblad and Annaz 2010, Wearing 2010). Thus, treating metaphor as similar to indexicals, where their content is, in a sense retrieved from the context of utterance, cannot capture what it is that leads to differences in the difficulty of interpreting metaphor in contrast to indexicals in autistic people, unless one wants to argue that although the general mechanism is the same, what is difficult for autistic people is to provide the presuppositions necessary for successful interpretation.14 However, as I have already said, I think the idea of potential metaphoric content being provided by the context is a problematic assumption itself. Another approach that grounds metaphoric interpretation of utterances in their semantics is that proposed by Dölling (2000). Thus, Dölling (2000) assumes an operator, which, generally, allows both for metonymic and metaphorical interpretations of the argument it applies to. The operator applies as part of the semantic composition of an utterance’s logical form and enhances the general variation potential of the meaning of the expression it applies to by introducing a range of parameters which have to get fixed during interpretation. (116) met: λP λx.S 3 n (x) & ∃y [R3 n (y)(x) & S 2 n (y) & ∃z [R2 n (z)(y) & S 1 n (z) & Qn v [R1 n (v)(x)][P(v)]]]15 16 Note that the application of the met-operator is obligatory, that is, it applies to the semantic form of all predicates that are of the appropriate type (namely <e,t>), regardless of whether that predicate will or will not be used with a shifted meaning in a particular utterance. In fact, Dölling (2000)’s approach ensures semantic compositionality in two ways. On the one hand, the met-operator is not inserted into the semantic form only once a conflict arises, but rather by default. On the other hand, the actual content of the met-operator (and, for that matter, of all SF-parameters, generally) is only supplied after semantic composition is finished. In case the enhanced variation potential is not realised in a particular utterance interpretation, 14 This line of argument is taken by Wearing (2010) in her defence of RT’s approach to metaphor interpretation. Thus, she writes ‘. . . the difficulties with metaphor exhibited by some speakers with autism can be explained as the result of having to interpret while lacking crucial information, rather than as a defect in the processing abilities required to understand a metaphor correctly’ (Wearing 2010, p. 207). 15 S n and Rn as well as Qn are SF-Parameter which need to get fixed during interpretation, S n by a predicate of sorts of entities, Rn by a predicate of relations between elements of two sorts of entity and Qn by some quantifier. 16 The complexity of the operator is due to the fact that there can be chains of metonymy which need to be captured. Problematic Phenomena 147 default values will be supplied to fix the SF-Parameters in the structure provided by the met-operator and, as a result, the contribution of the operator will be reduced to zero.17 Thus, in contrast to Stern (2006, 2009)’s approach, the interpreter does not have to disambiguate, as it were, whether the speaker used a non-metaphoric or a metaphoric expression type.18 Moreover, Dölling (2000)’s met-operator is clearly restricted in its application to one-place predicates, where this does not seem to be the case for Stern’s ‘Mthat’-operator. Let us look at some examples involving the lexical item Lamm (lamb). Dölling (2000) gives the following semantic form (SEM) for Lamm. (117) λx. x K n lamm Recall that K n is an SF-Parameter which may be fixed by (‘instance of’), ≤ (‘subkind of’) or = (‘is a’) and thus abstracts from the different readings of a systematic polysemous expression. Thus, for the occurrences of Lamm in (118a – 118c), the parameter will be fixed as in (119a – 119c), respectively. (118) a. Das Lamm ist ein Tier. ‘The lamb is an animal.’ b. Das Lamm springt über den Zaun. ‘The lamb jumps over the fence.’ c. Das Jungtier eines Hausschafes ist ein Lamm. ‘The offspring of the domestic sheep is a lamb.’ (119) a. λx. x = lamm b. λx. x lamm c. λx. x ≤ lamm That is, in (118a), Lamm is understood as denoting the kind, whereas in (118b) it denotes an instance of that kind and in (118c), a subkind of the kind. Lamm, of course, can also be used metonymically as, for example, in (120). (120) a. Hans möchte heute Lamm essen. ‘Hans wants to eat lamb today.’ b. Maria trägt seit gestern Lamm. ‘Since yesterday, Maria wears lamb.’ c. Anna weigert sich, Lamm zu essen oder zu tragen. ‘Anna refuses to eat or wear lamb.’ 17 Thus, the default value for Qn is ∃, that for S n is ENT (for ‘entity’) and that for Rn is =. That is, in Stern’s approach the utterance of (i) might be a token either of the expression type in (ii) or the metaphoric expression type in (iii). 18 (i) Juliet is the sun. (ii) Juliet is the sun. (iii) Juliet Mthat[is the sun]. 148 Utterance Meaning and Communicative Sense – Two Levels or One? It is in such cases that the material contributed by the met-operator comes to bear. Thus, applied to the semantic form of Lamm, the parameters available will be fixed for the respective occurrences in (120) as indicated in (121).19 (121) a. λx. FLEISCH(x) & ∃y [MAT(y)(x) & (y) (lamm)] b. λx. FELL(x) & ∃y [MAT(y)(x) & (y) (lamm)] c. λx. FELL∨FLEISCH(x) & ∃y [MAT(y)(x) & (y) (lamm)] That is, in (120a), Lamm is understood as denoting a particular type of meat, in (120b), a particular type of fur and in (120c) something that is either or both a particular type of meat/fur. Note again that although in Dölling (2000)’s approach metaphorical as well as metonymic interpretations are grounded in semantics, the actual content is supplied only after semantic composition is finished. That is, the parameters provided by the met-operator need to be fixed just as the parameters that may be involved in the SEM of the individual lexical items involved in an utterance (i.e., K n ). The process of fixing parameters, then, takes into account contextual and background conceptual information and results in a truth-evaluable full proposition. Thus, the predicates that are inserted as values for the parameters are based on a range of conceptual axioms that capture the relations holding between them. For instance, for the predicates involved in (121), at least the following axioms have to be assumed. (122) a. ∀x [DING(x) → ∃y [STOFF(y) & MAT(x)(y)]] b. ∀x [LAMM(x) → DING(x)] c. ∀x [TIER*(x) → ∃y [FLEISCH(y) & MAT(x)(y) & TYP ESSBAR(x)]] d. ∀x [TIER**(x) → ∃y [FELL(y) & MAT(x)(y) & TYP PELZBAR]] e. ∀x [LAMM(x) → TIER*(x) & TIER**(x)]20 Finally, the same met-operator is also involved in possible metaphorical uses of Lamm, such as in (123), where Lamm seems to be used with the meaning indicated in (124). (123) Annas Mann ist ein Lamm. ‘Anna’s husband is a lamb.’ (124) λx. PERSON(x) & ∃y [INST(y)(x) & ART(y) & ∃z [ÄHNLM (z)(y) & (z) = (lamm)]]21 Thus, in (123) Lamm denotes a person that instantiates a particular kind, where that kind is similar to the kind of lambs with respect to the set of features M. Note that although there is only one operator involved in metonymic and metaphorical interpretations of utterances, the difference between the two is still captured by the type and nature of the information that is involved in fixing the parameters 19 The material fixed by default values and thus, in a sense, cancelled out, is not stated. TIER* and TIER** denote particularly restricted sets of animals, namely such that their flesh is typically edible (TIER*) and such that their fur can typically be used for coats (TIER**). 21 ART is a predicate of the domain of kinds and ÄHNL is a predicate for the relation of similarity relative to the set of features M. 20 Problematic Phenomena 149 provided by the operator in the two interpretation situations. Thus, in metonymy the ontological relations between the predicates that serve as values are captured by axioms that can be taken to be part of the background knowledge. That is, the relevant conceptual relations for a metonymic interpretation are, in a sense, given. In contrast, part of metaphor interpretation seems to be the establishment of new conceptual entities (ART) and relations (ÄHNLM ), usually based on an assumption of similarity of features between the kind denoted by the metaphor vehicle and the metaphor target. One question that arises at this point is how the members of the set M are determined. Dölling (2000) acknowledges the importance of this question, but does not answer it. ‘Die kognitiven bzw. pragmatischen Bedingungen, die zum einen die Festlegung der jeweiligen Merkmale und zum anderen die Auswahl des metaphorisch verwendeten Ausdrucks bestimmen, gilt es zu untersuchen’ (Dölling 2000, p. 50).22 Another question is how, generally, the conceptual values are determined that are inserted for the respective SF-Parameters. Dölling (2000) mentions abductive interpretation as an appropriate mechanism (cp. Hobbs et al. 1993). Generally, assuming one basic operator that ensures unproblematic integration of conceptual material during utterance interpretation appropriately captures the amount and variety of meaning variation exhibited by many lexical items. In fact, the same operator may also be involved in cases of what RT calls ‘loose use’ of an expression. The differences in perceived non-literality could be correlated to the differences in the type of information that is inserted as the value of the parameters provided by the operator or in the effort of providing/creating that information in the first place. Thus, as in Stern (2000)’s approach, in Dölling (2000)’s approach, the difficulties autistic people exhibit in understanding of metaphor (or metonymy for that matter, cp. Rundblad and Annaz 2010) cannot be captured by a general difference in the semantic operations involved in the two types of figurative language use as both are based on one and the same operator. Rather, it has to be assumed that autistic people are not able to provide (or in the case of metaphor to create) the necessary conceptual information. A similar point has to be made concerning typically developing children, where there also seems to be a difference in metonymy as compared to metaphor understanding, in that the former is generally easier than the latter and competency for both types of figurative language use increases with chronological age (cf. Rundblad and Annaz in press). Assuming one operator in the semantics that is involved both in cases of metonymy and metaphor, the differences in understanding these two types of figurative language use have to be explained by differences in providing/creating the conceptual information that is integrated by that operator. As we saw above, generally, Dölling (2000)’s approach allows for such a difference to be made. 22 ‘The cognitive or pragmatic conditions, which determine the selection of the respective features, on the one hand, and the choice of the metaphorically used expression, on the other, have to be investigated.’ 150 Utterance Meaning and Communicative Sense – Two Levels or One? Summary None of the alternative approaches discussed here assume that metaphor interpretation is based on the prior recovery of utterance meaning, rather, it is simple (RT, Glucksberg and McGlone 1999, Dölling 2000) or more complex expressions (Stern 2000, 2006, 2009) within some utterance that are interpreted metaphorically.23 That is, metaphorical interpretation seems to be based on the lexical meaning of the metaphor vehicle. In fact, if one assumes – and I argued above that one should – that lexical meaning is characterised by a high degree of underspecification, then metaphor is not actually based on lexical meaning, since this is not even a full concept, rather it is based on, in Dölling (2001)’s terms, one of the set of possible values for the SF-parameter in the SEM of the expression under consideration. The important point to note is that in all the approaches discussed here, metaphor interpretation is taken to be part of the processes that determine utterance meaning rather than being based on the latter. In other words, contrary to Grice’s assumptions, metaphor is not taken to be a phenomenon of the level of what is meant. In fact, treating metaphor as arising at the level of utterance meaning or what is said actually is compatible with Grice’s assumption that what is said is entailed by what is meant, since the metaphoric meaning will be treated as forming part of what is said by the respective utterance. Thus, there is no need to assume that in case of metaphor, the speaker only ‘makes as if to say’ some literal proposition which, in ‘non-metaphorical circumstances’, actually would be taken to constitute what is said. Rather, in the case of metaphor, there is no ‘literal proposition expressed’ to be determined in the first place. Having said this, we saw that the meaning the metaphoric interpretation is based on does play a role in the actual interpretation, at least in the case of novel metaphors. Thus, it seems that metaphor interpretation involves the creation of a new concept on the basis of relevant features of the ‘literal’ concept the interpretation starts out from. In case of the interpretation of novel metaphors, therefore, one would expect that people first access the metaphor vehicle’s literal meaning and that the metaphorical reading becomes available afterwards. This is the pattern observed in the experiments cited in this section. Generally, it seems that metaphor interpretation proceeds independently of the intentions the speaker had in making the utterance that includes the metaphor.24 This assumption is implicit in Glucksberg and McGlone (1999)’s as well as Stern’s and Dölling’s approaches to metaphor interpretation. Thus, in Glucksberg and McGlone (1999)’s approach, an attributive category is created on the basis of relevant features of the metaphor vehicle, where the relevant features are determined by comparing the concepts associated with both the metaphor topic and the metaphor vehicle. This process seems to be free of considerations of speaker intentions. Similarly, in Stern’s approach, it is simply assumed that the potential metaphoricallyassociated meaning aspects are presupposed by the context, thus, it seems they are not determined based on a consideration of what the speaker intended with his ut23 Compare this to Borg (2001), who seems to assume that metaphor is based on the proposition literally expressed by an utterance – i.e., its minimal semantic content – and involves a set of associated propositions. 24 Cf. Borg (2001) Problematic Phenomena 151 terance. In Dölling (2001)’s approach, the metaphoric interpretation of a particular expression is also constituted by the creation of a new concept (ART) and a relation (ÄHNLM ) holding between that new concept and the concept associated with the metaphor vehicle. Again, the features with respect to which the two concepts are taken to be similar seem to be determined independently of the speaker’s intentions. In contrast, in RT, metaphor interpretation cannot strictly be characterised as being independent of speaker intentions, since it results from an overall interpretation process which crucially is assumed to make reference to speaker intentions. Recall that in chapter 2, I cited experiments investigating the interpretation of familiar metaphors, which indicated that their figurative meaning is available as early as their literal meaning in the interpretation process. Thus, in such cases it seems that during interpretation the metaphoric meaning no longer has to be created but is already stored as a potential reading for the expression in question. In other words, in case of familiar metaphors, their metaphoric reading might have become a member of the set of values of the SF-parameters in the SEM of the metaphor vehicle. 4.1.2 Irony Traditional Characterisation and its Problems Traditionally, verbal irony is characterised as a figure of speech which involves the substitution of a literal by a figurative meaning. More specifically, in irony the figurative meaning is characterised as being the opposite of the original literal meaning. Thus, consider the example in (125). (125) Mary: Paul really is a fine friend. [uttered after Peter has told Mary that Paul has repeatedly run down Mary behind her back.] In the context in which this utterance is made, it is clear that its utterance meaning – determined on the basis of narrow context – would be judged false. However, once the speaker’s intentions are taken into account and the hearer has realised that the speaker intended her utterance to be taken ironic, he will consider what was meant or the communicative sense as being true. In his treatment of irony, Grice essentially adopts the classical view. His innovation is to consider irony along with other types of non-literal meaning such as metaphor as implicatures. Thus, his analysis of example (125) involves the flouting of the maxim of quality and a calculation of the irony-implicature by the hearer along the following lines. (126) a. The speaker has expressed the proposition that Paul really is a fine friend. b. However, the context in which she expressed that proposition makes it obvious that this proposition is blatantly false. c. Thus, it is clear that the speaker cannot have intended to mislead the hearer. d. If the speaker is to be taken to adhere to the co-operative principle, in saying that Paul really is a fine friend, she must have meant the opposite of what she said. 152 Utterance Meaning and Communicative Sense – Two Levels or One? As has been pointed out by different authors (e.g., Wilson and Sperber 1992, Happé 1993, Dews and Winner 1999, Colston 2000, Schwoebel et al. 2000, Colston 2002, Colston and Gibbs 2002, Giora et al. 2007), there are several problems with the traditional view of irony, based on classical rhetoric and Grice’s reanalysis of it in terms of conversational implicature. For instance, in irony, speakers do not always and simply mean the opposite of what they say. Moreover, there are both theoretical considerations as well as empirical evidence that shows that the processing of metaphor and irony differs. Also, there is empirical evidence that shows that people do not simply replace and thus disregard the inappropriate literal meaning of an ironic utterance. Irony as echoic interpretive use Wilson and Sperber (1992) give a range of examples which show that irony cannot be characterised simply as involving a meaning that is the opposite of what has actually been said. The examples include ironical understatements (127a) and non-ironical false-hoods (128a). (127) a. You can tell he’s upset. [uttered by A to B in a situation in which a customer C complains in a shop, blind with rage and making a public exhibition of himself] b. You can’t tell he’s upset. c. You can tell he’s not upset. Wilson and Sperber (1992) claim that contrary to what is suggested by the traditional view of irony, (127a) neither communicates (127b) nor (127c). Rather, what is communicated by (127a) – as is typical of understatements in general – is somehow more than what is actually said. Thus, this is an example where the traditional characterisation of irony is not applicable but which, nevertheless, intuitively seems to be ironic. Let us now turn to an example, which conforms to the traditional definition of irony, but fails to be ironic. A and B are out for a walk. They pass a car with a broken window and A says (128a) to B. (128) a. Look, that car has all its windows intact. b. That car has one of its windows broken. Assuming that B, in saying (128a) actually intended to communicate the opposite (128b), B’s utterance meets the criterion of being ironic, but intuitively it fails to be so. Thus, the traditional characterisation of irony cannot account for all cases of irony and it subsumes cases which actually are not ironic. Moreover, the traditional view cannot explain why ironic utterances should be used at all. That is, if in irony the speaker simply says the opposite of what he means, the question arises what rational explanation there is for the fact that the speaker did not simply say what he meant in the first place. Wilson and Sperber (1992) offer an alternative characterisation of irony in terms of RT based on the idea that in irony, a speaker makes an utterance which echoes Problematic Phenomena 153 a thought she implicitly attributes to someone else and from which she wants to distance herself. Generally, an utterance may be used descriptively or interpretively. In the first case, the proposition expressed by the utterance is a true description of an actual state of affairs. In the second case, the proposition expressed by the utterance resembles another proposition that is the content of some other representation, for instance, an attributed thought. In cases of irony, Wilson and Sperber (1992) claim, utterances are used interpretively. To illustrate the difference, let us turn back to the sentence in example (125). (125) Paul really is a fine friend. A speaker using this sentence descriptively in an utterance is expressing the proposition that Paul really is a fine friend, which, in principle, can be judged against external reality as to whether it is true or not. Suppose, now that John claimed that Paul really is a fine friend and Mary, thus, entertains the thought verbalised in (129). (129) John believes that Paul really is a fine friend. Mary can use the sentence in (125) interpretively and thereby not express the proposition that Paul really is a fine friend, but rather express the proposition verbalised in (129) and at the same time dissociate herself from the proposition that Paul really is a fine friend. In other words, Mary can both believe that (129) and at the same time believe that Paul really isn’t a fine friend. Thus, according to RT, interpreting irony involves second-order metarepresentation of thoughts attributed to others. In the example above this means that for Peter to understand that Mary’s utterance is ironic, he has to be able to determine the proposition that Mary believes that John believes that Paul really is a fine friend and that Mary, at the same time, dissociates herself from the proposition that Paul really is a fine friend. Characterising irony as a form of echoic interpretive use of an utterance allows for an explanation of why ironical understatements are understood as ironic although in these cases the speaker does not say the opposite of what he meant and why nonironical false-hoods are not taken to be ironic although here the speaker putatively said the opposite of what he meant. Moreover, with such a characterisation, there is no assumption that there is any simple other way in which the speaker could have expressed what he intended to. That is because the ironical speaker does not simply mean the opposite of what he says. In other words, it is a simplification of what is going on in irony to merely assume that it involves a figurative, or nonliteral meaning. What it actually involves is the echoing by the speaker of a thought she implicitly attributes to someone else and from which she, at the same time, dissociates herself. Irony as a Form of Indirect Negation In contrast to Wilson and Sperber (1992), Giora (1995) defends a view that is based on the traditional view of irony as involving negation. More specifically, she characterises irony as a form of indirect negation, which differs from direct negation in a range of features. For instance, indirect negation does not lead to the type 154 Utterance Meaning and Communicative Sense – Two Levels or One? of scalar interpretation found for direct negation. Thus, the meaning expressed by a directly negated sentence such as in (130a) can be paraphrased as in (130b), which, in principle should allow for a more specific interpretation such as in (130c). However, that is usually not what is communicated by an utterance of a sentence such as (130a). Rather, it is a member of the scale closer to the member which has been negated that is communicated to hold (130d). (130) a. The water isn’t hot. b. The water is less than hot. c. The water is cold. d. The water is warm. Giora (1995) argues that such a scalar interpretation is not found with indirect negation. Another aspect of direct negation that indirect negation does not show is its linguistic and psychological markedness. Obviously, indirect negation is not morphologically marked as such. In addition, it seems that indirect negation, and thus irony, does not have direct negation’s ‘unpleasant value’. That is, forms of indirect negation are perceived as less aggressive than direct negation.25 Note further that in Giora’s approach irony is not a case of flouting some maxim of quality. However, it does violate what Giora calls the graded informativeness condition26 since the speaker by his utterance in the context at hand introduces a proposition which is highly improbable (and as such perceived as ‘marked’) given that context. This violation leads the hearer to ‘. . . evoke an unmarked interpretation (i.e., the implicature) comparable with the marked message whereby the difference between them becomes perceivable’ (Giora 1995, p. 245). To illustrate Giora (1995)’s approach, let us look once more at example (125). (125) Mary: Paul really is a fine friend. [uttered after Peter has told Mary that Paul has repeatedly run down Mary behind her back.] With respect to the context in which Mary makes her utterance, the proposition her utterance expresses is highly improbable, however, she actually uses it to evoke a less improbable interpretation. Moreover, the proposition expressed by her utterance is not simply disregarded as not fitting the context, rather ‘. . . it is used to implicate that a specific state of affairs is far from or very much unlike the expected/desired state of affairs indicated by the comparison’ (Giora 1995, p. 245). Thus, in Giora (1995)’s approach, the ‘literal’ utterance meaning is not simply disregarded but rather is necessary in order to compute the intended irony. In terms of processing effort, this suggests that irony should be more difficult to interpret than ‘literal’ language, since it involves the determination of both a literal as well as a non-literal meaning. In contrast to the more traditional views of irony involving the notion of negation, Giora (1995) also offers an explanation as to why a speaker should choose a form 25 Cf. example (131) below. According to this condition on well-formed discourses, each proposition that is expressed in a discourse should be either as or more informative than the proposition preceding it. 26 Problematic Phenomena 155 of indirect negation over overt negation. Not surprisingly, the reasons given refer back to the differences stated for the two types of negation. Thus, in using indirect negation, the effect of scalar interpretation as well as the psychological markedness of direct negation can be avoided. In fact, Giora (1995) considers irony ‘. . . “a politeness strategy” which enables the speaker to negate . . . by using an affirmative expression’ (Giora 1995, p. 260).27 Thus, consider the example in (131), where the ironical speaker in using the affirmative (131a) conveys the rather negative (131b), which will be understood as criticism but is not as agressive as if the speaker had said (131c). (131) a. (Ironically:) What a lovely party! b. The party is far from being lovely. c. The party isn’t lovely at all! Note also that in Giora’s approach no reference is made to higher-order metarepresentation being necessarily involved in irony interpretation. In fact, Giora (1995) criticises Wilson and Sperber (1992)’s view of irony as echoic interpretive use of an utterance on several accounts. She argues that, on the one hand, the approach does not capture all cases of irony and, on the other hand, it predicts irony to be involved in cases where it actually is not present. More specifically, Giora (1995) claims that there are both instances of intuitive irony which do not involve an echoic interpretive use of an utterance and instances of echoic interpretive use of an utterance, which fail to be ironic in the context in which they occur. Thus, consider Giora’s example conversation in (132), assumed to take place between two friends who actually know the person talked about quite well. (132) A: Do you know any G. M.? B: Rings a bell. Giora (1995) argues that in this case of irony, one need not necessarily assume that the utterances made need to be attributed to another speaker. They thus fail to be echoic. However, as Curcó (2000) points out, assuming that irony involves the speaker’s echoing of a thought implicitly attributed to some other speaker actually is to narrow a characterisation of echoic use and she argues that it is not how Wilson and Sperber (1992) view the notion. Rather, ‘[e]choic utterances . . . include those cases where the speaker attributes a thought or an utterance to herself at a time different from the time of utterance . . .’ (Curcó 2000, p. 262. emphasis in the original). Thus, in the example above, the thought uttered by B is attributable to B in a context where it is not mutually known to the conversation partners involved that B knows G. M. very well. For an example involving an echoic interpretive use of an utterance which fails to be ironic, Giora (1995) gives the conversation in (133). (133) Dina: I missed the last news broadcast. What did the Prime Minister say about the Palestinians? 27 Compare this to Dews and Winner (1999)’s Tinge Hypothesis, according to which ‘. . . irony mutes the evaluative meaning conveyed in comparison to literal language. Thus, ironic criticism is less critical than is literal criticism, and ironic praise is less praising than is literal praise’ (Dews and Winner 1999, p. 1580). 156 Utterance Meaning and Communicative Sense – Two Levels or One? Mira: (with ridiculing aversion) That we should deport them. The idea is that with her utterance, Mira both reports on what the Prime Minister said and at the same time expresses her negative attitude to what she reports. Nonetheless, her utterance does not come across as ironic. In fact, this is precisely as it should be, Curcó (2000) argues, because actually, Mira’s utterance is elliptical and the proposition her utterance really expresses is (134) The Prime Minister of Israel said that we should deport the Palestinians. Thus, Mira actually uses her utterance descriptively in that she states a fact. That is, she says something true about a state of affairs in the real world, namely, that the Prime Minister of Israel said that the Palestinians should be deported. Since Mira used her utterance descriptively, according to Wilson and Sperber (1992), it cannot be understood as being ironic. Moreover, what the speaker is taken to dissociate herself from is only a subpart of the complex proposition expressed by her utterance, namely ‘that we should deport the Palestinians’. However, as Curcó (2000) points out, the fact that the speaker implicitly distances herself from the speech she reports is not sufficient to make her utterance ironic. A further claim Giora (1995) makes against Wilson and Sperber (1992)’s view concerns the processing of irony. She argues that under the ‘echoic interpretive use’ view, irony should not be more difficult to interpret than ‘literal’ language, since no norm or maxim is violated and no non-literal meaning is derived in consequence. In contrast, treating irony as a form of indirect negation, where a communicative norm is violated, leading the hearer to compute a non-literal meaning in the form of an implicature, suggests that the interpretation of ironic utterances should be more difficult than the interpretation of ‘literal’ language. Moreover, since the ‘irony as indirect negation view’ assumes that an ironic utterance’s ‘literal’ meaning is necessary for the hearer to compute the difference holding between it and the derived implicature, it suggests that in processing an ironic utterance, its ‘literal meaning’ should remain activated during the interpretation process. However, there is no reason why difficulties in processing should only arise because some communicative norm has been violated and, as a consequence, some further meaning has to be computed. In fact, and as we will see below, Wilson and Sperber (1992)’s claim that irony interpretation involves second-order metarepresentation of an attributed thought on the part of the hearer, leads to the same hypothesis as posited by Giora (1995), namely that processing of irony should be more difficult than that of ‘literal’ language. Moreover, since in Wilson and Sperber (1992)’s account there is no non-literal meaning replacing the literal utterance meaning, their approach also predicts that an ironic utterance’s literal meaning should be and remain activated during interpretation. Empirical Results Concerning Irony Interpretation As we already saw above, concerning the relative difficulty of processing irony in contrast to ‘literal’ language, Wilson and Sperber (1992) and Giora (1995) make predictions similar to that suggested by Grice’s framework. However, both Wilson and Sperber (1992)’s as well as Giora (1995)’s view are in contrast to Grice’s view Problematic Phenomena 157 of irony and metaphor, where the two phenomena are treated as types of non-literal meaning whose interpretation essentially involves similar processes. According to Grice, in both cases the literal meaning of the respective utterances is rejected as not fitting the context and the non-literal meanings are calculated as implicatures. However, there are both theoretical considerations as well as empirical evidence that irony and metaphor differ in some essential properties. On the one hand, Wilson and Sperber (1992) argue that the interpretation of irony seems to require the ability of second-order metarepresentation, while in case of the interpretation of metaphor an ability of first-order metarepresentation seems to be sufficient. Moreover, these authors assume that in contrast to irony, in cases of metaphor, utterances are used descriptively. On the other hand, Giora (1995) suggests that comprehending irony is more exacting than understanding metaphor, since the former is a type of indirect negation and negation, generally, has been shown to take more time in processing than positive utterances do (cf. Clark 1973). Thus, as concerns the processing of irony and other types of meaning, the two approaches make similar assumptions, although for different underlying reasons. Similarly, in contrast to Grice, both these approaches assume that in interpreting an ironic utterance, its literal meaning is not simply discarded. However, whereas Giora (1995) holds this assumption due to the fact that she assumes the literal utterance meaning is necessary for a hearer to compute the irony as such, Wilson and Sperber (1992) hold it because in their approach to irony comprehension there is no derivation of a non-literal meaning in the first place. Empirical investigations of these hypotheses are overall favourable to Wilson and Sperber (1992)’s and Giora (1995)’s view concerning irony interpretation in general and differences to metaphor interpretation in particular. Happé (1993), for instance, provides data that corroborates Wilson and Sperber (1992)’s view that irony interpretation involves second-order metarepresentation. Colston and Gibbs (2002) is an investigation of the interpretation of metaphor vs. irony, again, showing that there are differences between them. Finally, Giora et al. (2007) show (as do Dews and Winner 1999, Schwoebel et al. 2000) that during the interpretation of ironic utterances their so-called literal meaning is not simply disregarded and substituted by a non-literal meaning. Thus, Happé (1993) investigated whether the interpretation of irony does involve second-order metarepresentation. In order to do so, Happé (1993) chose as subject group autistic children, since they are independently assumed to have problems with higher-order metarepresentation (cf. Baron-Cohen et al. 1985). Happé (1993, p. 103) assumed that ‘. . . an inability to represent thoughts about thoughts would preclude . . . utterances which . . . are interpretations of the speaker’s thought about an attributed or desirable thought’. In other words, if autistic children cannot form second-order metarepresentations, then they should have problems identifying cases of irony if Wilson and Sperber (1992)’s assumption is correct that irony interpretation involves such second-order metarepresentation. Happé (1993) divided the autistic subjects into three groups, according to the results of tests designed to assess their theory of mind abilities. Thus, one group of subjects consisted of children who had passed all first-order metarepresentation tasks given them but none of the higher-order tasks (i.e. first-order ToM group). Another 158 Utterance Meaning and Communicative Sense – Two Levels or One? group consisted of children who also had passed the first-order tasks and in addition had scored high (6 out of 8 correct) on the second-order task (i.e. second-order ToM group). Finally, one group of subjects consisted of children who had failed the firstorder false belief tasks (i.e. no-ToM group). A last group of subjects, functioning as a control to the autistic subject groups, consisted of children with moderate learning difficulties, who had passed both first- and second-order metarepresentation tasks, but whose verbal IQ was considerably lower than that of the first- and second-order ToM groups while comparable to that of the no-ToM group. This group was included to exclude the possibility that the autistic childrens’ performance was simply due to a general intellectual impairment. The results of Happé (1993)’s experiments show that autistic children who pass the second-order metarepresentation tasks also are able to understand ironic utterances. The first-order ToM subject group – who failed at attributing false beliefs to others – do considerably worse in understanding ironically intended utterances. However, these same subjects (as well as the subjects of the second-order ToM group) seem to have no problem understanding metaphoric utterances. This result can be taken as evidence that metaphor interpretation does not involve as complex cognitive tasks as does the interpretation of irony. In contrast to both the firstand second-order ToM groups, the No-ToM group has severe problems understanding both metaphor and irony. Note that the results for the non-autistic control group showed that they did better than both the no-ToM group in understanding metaphor as well as the first-order ToM group in understanding irony, although the control group’s verbal ability was similar to the no-ToM group’s and lower than the first-order ToM group’s. This suggests that the difficulty for the no-ToM group of understanding metaphor and for the first-order ToM group of understanding irony is not due to a generally low verbal ability of these subjects. Thus, Happé (1993)’s results support Wilson and Sperber (1992)’s assumption that the interpretation of irony involves second-order metarepresentation and that the interpretation of irony differs from that of metaphor, which only involves firstorder metarepresentation.28 Similar results were obtained by Colston and Gibbs (2002), investigating the interpretation of both metaphor and irony in ‘normal’ subjects. In their first experiment, Colston and Gibbs (2002) tested identical statements which, depending on the context in which they were placed, either had a metaphoric or an ironic meaning. Subjects read at their own pace through small stories presented on a computer screen, the last sentence of which was the target sentence and was either intended as metaphoric or ironic. Following the target sentence, subjects saw a paraphrase sentence of the target, intended to capture the speaker’s meaning. Their task was to decide whether the paraphrase was good or bad and their reaction times were measured. The results show that subjects took longer to comprehend ironic than metaphoric utterances. Assuming that longer reaction times reflect a larger amount of inferencing, these results again suggest that irony is harder to interpret than metaphor. 28 However, this is not to say that the ability of second-order metarepresentation is the only relevant factor in irony understanding. Thus, Pexman and Glenwright (2007) point out that such factors as neural maturation and social learning also play a role in children’s development of irony understanding. Problematic Phenomena 159 Although the results of their first experiment suggest that processing irony involves more complex inferencing than required for interpreting cases of metaphor, reaction times as such do not tell us exactly what kinds of inferences subjects made when they interpreted metaphoric as compared to ironic utterances. Thus, in a second experiment, Colston and Gibbs (2002) investigated whether ‘. . . people actually engage in more complex metarepresentational reasoning when reading ironic remarks than they did when reading metaphoric statements’ (Colston and Gibbs 2002, p. 64). To this end, the authors had subjects read the same stories with their respective final ironic or metaphoric utterances as in the first experiment. After reading the stories, subjects were asked to rate their agreement with a number of inference statements. These statements were designed to probe for the various metarepresentational inferences people are assumed to draw when interpreting ironic or metaphoric utterances. Thus, one statement probed subjects’ assessment of the beliefs of the speaker when interpreting ironic or metaphoric utterances. Another statement was intended to show whether the notion of pretense plays a role in understanding irony or metaphor.29 The third statement was designed to reveal whether subjects recognised some echoing by the speaker of a thought attributed to someone else. Colston and Gibbs (2002) hypothesised that if irony interpretation in contrast to metaphor interpretation involved higher-order metarepresentation, subjects should give higher ratings to the inference statements referring to such higher-order reasoning after having read stories ending in ironic utterances than after stories ending in metaphoric utterances. This hypothesis was confirmed. In addition, Colston and Gibbs (2002) also investigated whether there are differences between understanding utterances that involve irony and such that involve both metaphor and irony. The authors predicted that metaphoric irony should be more difficult to understand than ‘simple’ irony. The idea is that there is a ‘. . . mesh between these two types of figurative meaning . . .’ that leads to longer processing times in cases where both metaphor and irony are involved. They also note that this view is compatible with the assumption that different processing modes are required in metaphor vs. irony interpretation, as suggested, e.g., by Wilson and Sperber (1992), Sperber and Wilson (1995) (i.e., descriptive vs. interpetive use of utterances). The results of that experiment confirmed their prediction. That is, interpreting cases of metaphoric irony took subjects longer than interpreting cases of simple irony, suggesting that the former is a cognitively more exacting process. There are several studies which investigated whether or not in irony comprehension, the non-intended literal meaning of the ironically intended utterance is processed and if it is to what extend. Dews and Winner (1999), for instance, investigated the processing of ironic criticism (cf. 135a) and ironic praise (cf. 135b) compared to ‘literal’ uses of the same expressions (cf. 136a constituting literal praise and 136b constituting literal criticism) (examples taken from Dews and Winner 1999). (135) a. Good move! [to a person who has just done something foolish] b. We never do anything fun. 29 It has been suggested, e.g. by Leslie (1987), that engaging in pretense requires the capacity for second-order metarepresentation. 160 Utterance Meaning and Communicative Sense – Two Levels or One? [to a companion on a dream vacation] (136) a. Good move! [to a person who has just done something wise] b. We never do anything fun. [to a roommate who has been studying hard for finals] The hypothesis tested was that subjects should take longer judging the evaluative tone of the ironically intended utterances, if their unintended literal meaning, which is opposite in evaluative tone, is computed and thus interferes. Thus, subjects were presented stories on a computer which ended in a character making either a literal or ironic comment. They then had to judge the evaluative tone of the comment by pressing a designated key. Reaction times were measured. The results show that subjects both made more errors in judging the evaluative tone of ironically intended utterances and they took longer to make those judgements than they did to judge literally intended utterances. Thus, they are interpretable as supporting the hypothesis that in irony interpretation at least some aspects of the literal utterance meaning are processed. Similarly, Schwoebel et al. (2000) also investigated the interpretation of ironic criticism and ironic praise and their literal counterparts. However, since one of their aims was to pin-point the exact time during interpretation at which the ironically intended utterances are recognised as such, they measured subjects’ reading times of target utterances. In a pretest, they had subjects indicate at which expression(s) in the target utterance, the utterance was recognised as ironic. The information gleaned from this test was used to divide the target sentences into three phrases for presentation to subjects. The first phrase gave no indication of intended irony. The second phrase ended in the expression(s) that had been identified as indicating the intended irony and the last phrase, again gave no indication of irony. An example story with a target sentence and an indication of its division is given below. (137) A new professor was hired to teach philosophy. The professor was supposed to be really sharp. When Allen asked several questions, the professor offered naive and ignorant answers. Allen said: That guy/is brilliant/at answering questions. Each story was followed by a comprehension true/false question which subjects had to answer by pressing one of two designated keys. Accuracy of answers to these questions as well as the reading times for each of the three phrases of target sentences was measured. The results show that subjects took longer to read the critical (second) phrase when it followed an ironically biasing story, than when it followed a literally biasing story. Schwoebel et al. (2000) interpret these results as showing, once again, that the ‘literal’ meaning of an ironically used utterance is processed, even when it is not intended. Moreover, the results indicate that the ironic interpretation of an utterance is accomplished ‘online’, that is, while the critical utterance is being processed. More recently, Giora et al. (2007) tested in how far contextual information available when processing an ironically intended utterance influences the interpretation process. More specifically, it was investigated what effect an expectation of irony Problematic Phenomena 161 on the part of the subjects had for their interpretation of critical utterances. In addition, the authors also were interested in the question of when the ironic interpretation of an ironically intended utterance becomes available to subjects. More generally, the experiments reported are aimed at supporting Giora’s graded salience hypothesis and the predictions it makes concerning the interpretation of irony. Thus, according to the graded salience hypothesis, apart from conventional irony, ironic interpretations of utterances are not salient. As we saw earlier, Giora assumes of salient meanings that they are always activated, even in contexts in which they are not intended. For cases of intended irony, this means that the salient meanings of the expressions used in the ironically intended utterance should be activated although they are inappropriate in the given context. In experiments 3 and 4 especially, it was tested whether given a strongly supportive context, an intended ironic interpretation would be determined without activation of the non-intended but salient utterance meaning, built up on the basis of the salient meanings of its parts. Materials consisted of small stories ending in the target sentence, which, depending on the preceding context, was either literally or ironically interpreted. To raise the expectation for irony, one set of materials actually only included stories ending in ironically intended utterances. An example context together with the ironically intended target utterance is given below (taken from Giora et al. 2007). (138) Sarit worked as a waitress in a small restaurant in central Naharia. The evening was slow, and even the few customers she did wait on left negligible tips. She didn’t think that the elderly man who walked in alone and ordered just a couple of small sandwiches would be any different. Indeed, after making her run back and forth throughout the meal, he left, and she collected his pay for the meal from off his table and found 2.5 NIS tip! When she showed her friends how much she got, Orna commented: “That was real noble of him!” The availability of the non-intended but salient utterance meaning and the intended but non-salient ironic meaning was elicited by presenting subjects with probe words related to one or the other of the respective utterance-level meanings at different inter-stimulus intervals (ISI).30 The results show that even in contexts where subjects can reasonably be taken to expect ironic comments from the speaker, it seems the target ironic interpretation takes longer to determine than the unintended, contextually incompatible but salient utterance meaning. In fact, even after an ISI of 1000 msec, probe words relating to the ironic interpretation of an utterance were more slowly responded to than probe words related to the salient utterance meaning. Overall, the results support the graded salience hypothesis and more generally those accounts, which assume that the literal (or rather salient) meaning of an ironically intended utterance is not simply discarded, rejected or suppressed. Moreover, they show that determining the ironic interpretation of an utterance is a timeconsuming process, which is not shortened even in highly supportive contexts.31 30 In the case of the example above, probe words were generous (literally related), stingy (ironically related) and sleepy (unrelated control). 31 This is also shown by the results of Giora et al. (2009)’s study, which investigated the influence of providing an ‘ironic situation’ on the interpretation of the target utterance as ironical. 162 Utterance Meaning and Communicative Sense – Two Levels or One? Summary Although the two post-Gricean approaches to irony discussed above differ in various details from one another, both assume that in order to correctly interpret an ironically intended utterance, that utterance’s ‘basic’ meaning – where this might in fact already involve figurative meaning aspects such as metonymy or metaphor – has to be processed. In RT this is so because the ‘would-be expressed’ proposition plays an integral part in irony, as it is interpretively used by the speaker to convey his dissociative attitude. In Giora’s approach, the utterance’s ‘basic’ meaning needs to be determined first as it is necessary to compute the intended irony. Contrary to Grice, who treated metaphor and irony as on a par, both postGricean approaches to irony discussed above assume that the understanding of metaphor and irony differs. Thus, whereas in the case of metaphor, it is the ‘basic’ meaning of the metaphor vehicle the metaphoric interpretation is based on, in case of irony it is the ‘basic’ utterance meaning that forms the basis for an ironic interpretation. In fact, as we saw in section 4.1.1, metaphor is actually taken to be a phenomenon of the level of utterance meaning, whereas irony clearly belongs to the level of speaker meaning. Generally, the two post-Gricean approaches to irony we looked at both assume that irony processing is cognitively more exacting than metaphor understanding. This assumption is supported by the results attained in experiments investigating the processing of metaphor and irony both in healthy subjects (e.g Colston and Gibbs 2002, Giora et al. 2007) as well as in autistic children (e.g. Happé 1993, MacKay and Shaw 2004). 4.1.3 Conversational Implicatures Generalised vs. Particularised Conversational Implicature – Theoretical Approaches As already mentioned, Grice characterised conversational implicatures as being inferences that are drawn on the basis of what is said, but which are not entailed by what is said. Thus, conversational implicatures are communicated in addition to what is explicitly said. In this sense, they constitute implied meaning aspects of what is meant by the speaker. On the Gricean view, then, it makes sense to claim that conversational implicature is a phenomenon that belongs to a level of meaning distinct from the level of what is said. On the one hand, what is said actually forms the basis on which CIs are drawn, on the other hand, CIs differ from the kinds of implicit (or in RTs terminology explicit) meaning aspects that may be argued to form part of utterance meaning, as the former arise due to the existence of particular conversational maxims and the general cooperative principle. As we already saw, Grice differentiated two types of conversational implicature, namely particularised vs. generalised conversational implicature. Recall that the difference between these two types of CI is that the former depends on the context in a way the latter does not. Thus, PCIs are computed only if the context is such that they are necessary to make sense of why the speaker said what he did. GCIs, on the other hand, occur unless the context is such that it prevents them from Problematic Phenomena 163 occurring. That is, while PCIs clearly are recognised as belonging to the level of what is meant by a speaker, this seems not so clear for GCIs. Thus, with respect to speaker intentions, PCIs have to be intended to arise, whereas GCIs seem to be independent of speaker intentions. In fact, a speaker might actually have to make sure ostensively that a hearer does not compute a particular GCI, if he does not intend the hearer to do so. Due to these characteristics of GCIs, they have received a treatment different from that of PCIs. Thus, Levinson (2000) suggests that GCIs actually do not arise at the level of speaker meaning or communicative sense, but rather that they are already involved in the determination of the truth-conditional content of the speaker’s utterance, that is, in determining what is said or utterance meaning. Chierchia (2004) takes this idea further in that he actually claims GCIs are grammatically induced meaning aspects (see also Chierchia et al. 2010). RT treats some types of what Levinson calls GCIs as part of the explicature of an utterance and some as genuine implicatures. In fact, RT argues against viewing the former meaning aspects as implicatures at all. This, in turn, is in contrast to Borg (2009) as well as Burton-Roberts (2006), who both take GCIs to be implicatures and as such not contributing to an utterance’s truth-conditional content, albeit, as we will see, for different reasons. Levinson (2000) argues that GCIs belong to a level of meaning distinct both from the level of utterance meaning as well as the level of communicative sense as discussed above.32 Whereas he takes the latter to be the level at which an utterance’s utterance-token-meaning is determined, he argues that there are phenomena, like GCIs, which should not be treated as once-off inferences, like PCIs are. This is because PCIs and GCIs are so different in nature, the former only being triggered in particular contexts, the latter always triggered unless the particularities of the context prevent them from being inferred. This property makes GCIs a kind of default inference, or, as Levinson puts it, a preferred interpretation. Thus, Levinson argues for a level of meaning he calls utterance-type-meaning33 , that is, a level of meaning that depends on the structure of language, but not on any particular context. For example, he takes GCIs, being one particular type of utterance-type meaning, to be the result of a number of heuristics which do not take into account such matters as speaker’s intentions or encyclopaedic knowledge of the domain talked about. Thus, utterance-type-meaning is also distinct from the level of utterance meaning, since contextual information does not have a role to play in the generation of such default meaning aspects as GCIs. Generally, utterance-type meaning can be taken to be a preferred interpretation for an utterance, which may be overridden. This level of meaning is based ‘. . . on general expectations about how language is normally used.’ (Levinson 2000, p. 22). To point out the significance of this level of meaning, Levinson argues that it is the basis for a distinction between convention of language and convention of language use. ‘Why is it that I can introduce myself with My name is Steve, but not I was given the name Steve, that I can express sympathy with you with 32 In fact, Levinson refers to the traditional two-fold distinction between sentence meaning and speaker meaning, of which he argues that it cannot properly accommodate the notion of GCIs and other related phenomena. 33 Note that he uses the term not in the sense Grice did. 164 Utterance Meaning and Communicative Sense – Two Levels or One? I am sorry but not conventionally with That saddens me; that I express outrage with Really! but not with In truth!; that I can say I am delighted to meet you but not idiomatically I am gratified to meet you; that I can choose a pastry by saying I’d like that one but not I’d admire that one, and so on.’ (Levinson 2000, p. 23) Thus, Levinson claims that a theory that does not assume a level of utterance-typemeaning cannot account for the regularity, systematicity and recurrence of particular types of pragmatic inference. Although particularly interested in GCIs, Levinson mentions other types of phenomena that may be considered as belonging to the level of utterance-type-meaning, for example, standardised inferences and short-circuited implicatures. These two phenomena differ from GCIs in that their determination is not based on heuristics and, moreover, in that the coming about of these particular types of utterance-type-meaning depends on routinisation, which notion, once again, can be taken to reflect (amongst other measures) frequency of use of a particular expression in a particular (or rather what gradually comes to be considered a normal) situation with a particular reading. Moreover, he takes the level of utterance-typemeaning to capture idiomatic uses of expressions as well as partially conventionalised readings. Thus, this may be the level at which collocations of particular expressions are associated with the concepts they are normally used to refer to (e.g., white wine). Levinson conceives of the level of utterance-type-meaning as a level at which ‘. . . sentences are systematically paired with preferred interpretations’ (Levinson 2000, p. 27). What is retrieved at the level of utterance-type-meaning are, e.g., schemas of particular sentence constructions as well as the preferred or default readings associated with particular, more or less complex expressions. Note that Levinson’s level of utterance-type-meaning plays a role in the determination of the utterance meaning of an utterance. That is, in the absence of a particular context, or in the absence of particularities in a given context, the interpretation process will determine on a reading for a particular utterance that can be considered normal or a default interpretation and in that sense, a preferred interpretation. Only if the context of utterance is such that it is not compatible with a default interpretation will the interpretation process determine a reading that is ‘special’ in the sense that it does not constitute a preferred or default interpretation. This suggests, of course, that in cases where there is no given context at all, a default interpretation will be determined, either on the basis of former uses of the expressions at hand or, in the case of GCIs, on the basis of some underlying heuristics. More specifically, Levinson assumes that GCIs result from the interaction of three general heuristics that underlie the interpretation of utterances. Thus, the Qheuristic (‘What is not said, is not the case.’) derives from Grice’s first submaxim of Quantity: ‘Make your contribution as informative as is required’. This heuristic gives rise to such GCIs as scalar and clausal implicatures. This can be shown using the familiar example below, where the GCI derived from the uttering of the sentence in (139a) is the proposition spelled out in (139b). (139) a. Some of the students went to the party. b. Not all the students went to the party. Problematic Phenomena 165 Thus, since all is the stronger lexical item on the scale that some belongs to, and since the speaker did not use that stronger element, the hearer is licensed to infer from that non-use that the speaker actually was not in a position to use it. Thus, although, semantically, the meaning of some is taken to be compatible with the meaning of all, the pragmatic inference, due to the Q-heuristic, is such that an interpretation of some as meaning ‘all’ actually is not warranted. The I-heuristic (‘What is expressed simply is stereotyptically exemplified.’) can be related to Grice’s second submaxim of Quantity: ‘Do not make your contribution more informative than is required’. This heuristic makes use of the fact that particular linguistic expressions are stereotypically used in particular situations, where the expression at hand does not fully encode what characterises that situation. However, we are able to use a minimal amount of expressions to describe a particular situation, because our interlocutor will, due to this second heuristic, determine on a stereotypical and much richer interpretation. Thus, in the example below, an utterance of (140a) normally is understood as expressing the proposition spelled out in (140b). (140) a. Peter broke a finger, yesterday. b. Peter broke one of his fingers, yesterday. The third heuristic Levinson identifies is the M-heuristic (‘What is said in an abnormal way is not normal.’), which goes back to Grice’s maxim of Manner. This last heuristic is, in a sense, in opposition to the I-heuristic, in that it ensures a nonstereotypical interpretation for a non-simple description of a particular situation. Whereas an utterance of the sentence in (141a) is normally understood as describing a situation in which John directly caused Peter’s death, this is not the case for an utterance of (141b). The idea is that the speaker employed a marked (complex) expression (cause to die) instead of a possible unmarked expression (to kill), indicating, due to the M-heuristic that the unmarked expression was not warranted, as it would have led the intepreter, due to the I-heuristic, to interpret the utterance as describing a ‘normal’, that is direct killing-situation. Thus, an utterance of the sentence in (141c) usually is understood as describing a situation in which John did not directly cause the death of Peter. (141) a. John killed Peter. b. John caused Peter to die. c. John caused Peter’s death indirectly. Generally, Levinson’s utterance-type-meaning seems similar to what Ruhl (1989) called default interpretation and tried to explain using the notion of stereotype. However, Ruhl did not assume that there are any default readings that arise only on the basis of particular conversational maxims or heuristics. Rather, from what he says about default interpretation, it exclusively relies on stereotypical uses of a particular (simple or complex) expression, where stereotypical may be taken to capture significantly high frequency uses. In fact, one might ask whether it is actually possible to differentiate between types of utterance-type-meaning being determined on the basis of heuristics and such as are determined on the basis of former uses of the expression 166 Utterance Meaning and Communicative Sense – Two Levels or One? in question. In other words, the question is whether the results of the interpretation process in case of particular utterances giving rise to GCIs can only be accounted for by applying to some underlying general heuristics or whether these are preferred interpretations that have emerged over repeated uses of the expressions in question. This latter view seems to be endorsed by Borg (2009, p. 79), who writes that ‘[a] GCI can be recovered just by knowing that a speaker who says “some” often conveys some but not all . . . this information could be gathered as the result of exposure to past conversational exchanges . . .’. So far, Levinson’s approach to GCIs can be taken to be Neo-Gricean in that he, similarly to Grice, assumes a number of underlying guidelines that lead to the derivation of various types of (generalised) conversational implicature. However, Levinson’s theory of GCIs goes further than Grice’s account of conversational implicatures in general, in that the former argues that GCIs are involved in the determination of what was said or the utterance meaning of an utterance. More specifically, he claims that the same processes that derive GCIs are responsible for the resolving of references, fixing of indexicals and disambiguation, which are necessary to arrive at the proposition expressed by an utterance. In other words, rather than being derived on the basis of what is said and thus playing no truth-conditional role, Levinson assumes that GCIs are necessary to arrive at what is said and that they do play a truth-conditional role. This observation leads Levinson to posit what he calls Grice’s circle and, eventually, a particular view on the interaction of semantic and pragmatic processes which is characterised by what he calls pragmatic intrusion into semantics. Thus, he differentiates two semantics components, as it were, namely Semantics 1 and Semantics 2. Essentially, this view follows the traditional definition of semantics, which is supposed to state the truth conditions for sentences and whose output therefore should be truth-evaluable propositions. Semantics 1, then, comprises the processes that produce the semantic form of a sentence from its surface structure. However, as we saw before and as Levinson agrees, the semantic representations resulting from semantic composition are not fully propositional, that is, they are not truth-evaluable. Thus, before semantic interpretation (e.g., Semantics 2) of the sentence in question can begin, the sub-propositional form has to be enriched. That is, particular pragmatic processes that take into account the context in which a particular sentence is uttered, have to provide contextually derived material to enrich the semantic form of the sentence to the proposition expressed in that particular context by that sentence. More specifically, Levinson argues that in determining the proposition expressed, pragmatic processes are inserted which are involved in the derivation of generalised conversational implicatures. These processes are supposed to guarantee the resolving of reference, fixing of indexicals and disambiguation and thus result in a truth-evaluable proposition, the proposition expressed. Once the proposition expressed is determined, semantic interpretation of that proposition can take place. Thus, what we have between Semantics 1 and Semantics 2, is an intrusion of a number of restricted pragmatic processes. What is problematic about this assumption is that, contrary to Grice’s original definition of conversational implicature, GCIs are taken to contribute to the truth-conditional content of an utterance (since they contribute to the proposition Problematic Phenomena 167 Semantics lexical meaning semantic composition sub-propositional SEM 1: sentence meaning pragmatic intrusion fully propositional saturation processes GCIs Pragmatics basis for further pragmatic inferences SEM 2: what is said PCI speech acts ... + narrow context + broad context what is meant Figure 4.1: Levinson’s view expressed by an utterance, or what is said, which then is interpreted by Semantics 2). Furthermore, since GCIs contribute to what is said by an utterance, we run into trouble with Grice’s original characterisation of CIs being inferred on the basis of what is said. This latter problem is what Levinson calls Grice’s circle: in order to infer CIs from a particular utterance, we need as a basis what is said by that particular utterance, which in turn is, at least partly, determined by processes that are employed to derive GCIs. As a solution to Grice’s circle, Levinson proposes that the processes which derive GCIs actually are local processes. That is, GCIs are not derived only once the interpretation process has yielded a full proposition (which, in fact, would not be possible if they are taken to be essential contributions to that proposition), rather, the processes that derive GCIs apply locally. In other words, the process that derives GCIs is triggered by particular expressions or constructions and it does not depend on a full proposition to apply. Note that Levinson’s assumptions concerning the processes or stages involved in the interpretation of utterances are actually quite similar to what we have seen already. Thus, the interpretation of some utterance starts with the semantic composition of the semantic form of the complex expression uttered. This semantic form is then enriched by pragmatically determined aspects of meaning to yield a full proposition which can then be checked as to its truth or falsity. Whether or not one calls the component responsible for semantic interpretation ‘Semantics 2’ seems to simply have to do with whether one is interested in somehow upholding the traditional view of semantics being concerned with full propositions. In a sense, RT also makes reference to this traditional view of semantics by differentiating between linguistic semantics and real semantics. While the former refers to the semantic component in the language faculty, the latter refers to the ‘language of thought’. Thus, the role of linguistic semantics is to provide, on the one hand, the meanings encoded by linguistic expressions and on the other hand, the rules of composition for a combination of those meanings. In contrast, it is in real semantics that the linguistic-semantically composed and pragmatically enriched semantic forms of ut- 168 Utterance Meaning and Communicative Sense – Two Levels or One? terances are interpreted.34 Whereas Levinson assumes that GCIs are based on pragmatic heuristics but ‘intrude’ into the semantic form of the utterance giving rise to them prior to its semantic interpretation, Chierchia (2004) assumes that scalar implicatures (SI) at least actually are computed by the grammar.35 However, similarly to Levinson, Chierchia (2004) also views SIs as default interpretations and assumes that they are locally drawn. More specifically, Chierchia (2004) argues that in computing the meaning of an utterance, both its ‘plain’ (without potential SIs) as well as its ‘strengthened’ meaning (with potential SIs) are computed. In particular, in determining the strengthened meaning of a given utterance, the set of alternative propositions the speaker might have expressed but didn’t plays a crucial role: ‘. . . the relevant set of alternatives is constituted by propositions built up by using the other items on the scale.’ (Chierchia 2004, p. 42). Thus, for a sentence such as given in (142a), the corresponding logical form is as in (142b) and the set of that proposition’s alternatives is (partially) given in (142c), where the arrow indicates the direction in which the entailment relation holds. (142) a. John earns $200 an hour. b. earn(j, $200) , earn(j, $300), earn(j, $400) . . . } c. { . . . earn(j, $100), entailment: ← Now, if someone utters (142a) in a ‘neutral’ context, he is normally understood as not ‘plainly’ expressing the proposition given in (142b), but rather the ‘stronger’ proposition in (143), which involves a SI. (143) earn(j, $200) ∧ ¬ earn(j, $n), where $200 < n Chierchia (2004) develops a definition for the operation of functional application (FA) that allows for an incorporation of SIs into an utterance’s semantic form while this is being built up (cf. 144).36 (144) Suppose α = [ β γ ], where β is of type <a,b> and γ is of type a. Then S S k β k k γ k , if k β k is not DE k[ β γ ]kS = k β k S k γ k , otherwise In order to account for the fact that in certain contexts (so-called downward-entailing ones) SIs do not arise/survive, Chierchia posits a constraint (the Strength Condition) that ensures that the plain meaning of a given utterance will be preferred, as in such cases it will be logically stronger than the ‘strengthened’ meaning.37 One 34 Thus, this second semantics component, then, does capture the semantic relations holding between the concepts a speaker expresses in a particular utterance situation. 35 Thus, Chierchia’s proposal does not apply to GCIs in general, but only to one subtype, namely scalar implicatures. 36 The superscript S stands for ‘strong’; DE abbreviates downward-entailing. 37 Strength Condition: ‘. . . the strong value cannot become weaker than the plain value.’ (Chierchia 2004, p. 62). Problematic Phenomena 169 advantage of Chierchia’s proposal over Levinson’s is that it can account for the fact that SIs might play a truth-functional role when the sentence including a SI trigger is embedded into a larger structure without having to assume pragmatic intrusion into semantics. That is, the process that computes SIs actually is taken to be part of the grammar. Turning back to Levinson’s characterisation of GCIs: this has been criticised by proponents of RT on several grounds. First, RT actually does not differentiate between different types of implicature, that is, there is no differentiation between GCIs and PCIs. Second, proponents of RT claim that the GCIs Levinson assumes contribute to what is said or the utterance meaning of an utterance in fact are not implicatures, but pragmatic enrichments of semantic form. Third, contrary to both Levinson and Chierchia, RT denies that GCIs have a default status and, thus are automatically and locally triggered. Proponents of RT argue against Levinson’s assumption that GCIs belong to a special level of utterance-type meaning. In RT, there is in fact no differentiation made between GCIs and PCIs. This is because all implicatures are taken to be subject to the same principle – the principle of relevance – and thus to result from similar types of processes. GCIs therefore cannot be taken to have a default status and thus to belong to some level of meaning distinct from the level of meaning that PCIs belong to. Thus, Carston claims that ‘No implicatures are a matter of default inference; rather, all must be warranted by contextual relevance’ (Carston 2004c, p. 8). This is not to say that implicatures do not vary in their generality, but the more general implicatures are not accorded the status of default interpretations. In addition, Carston (2004c) criticises Levinson’s view of GCIs as contributing to what is said. She claims that there is no circularity in Grice’s definition of CI. Rather, what Levinson claims to be GCIs are in fact no implicatures at all but rather pragmatically contributed aspects of utterance meaning. The crucial point here is that these pragmatically contributed aspects are not implicatures, where the relevant characteristic of implicatures is that they are inferred on the basis of complete propositions. Thus, the pragmatic processes that Levinson takes to contribute to what is said are in fact not such that they derive implicatures, as they do not need full propositions as their basis, but rather may operate on sub-propositional forms. Thus, it is not surprising that the phenomenon Levinson calls GCI very often contributes to the truth-conditional content of some utterance, because, actually, those aspects of meaning are part of what RT takes to be explicitly expressed by an utterance. Note, however, that RT claims the determination of explicature as well as implicature – both being pragmatically informed aspects of meaning – is characterised by mutual parallel adjustment. This means that it is not necessarily the case that the explicature of an utterance is determined prior to the determination of an implicature and implicatures, therefore, may be derived before a full proposition is available. More specifically, a particular explicature may be taken to be expressed by a particular utterance due to the influence of the hearer’s expectation of a specific implicature. On the one hand, this seems to be a reasonable assumption. Considering that communication proceeds so fast and that hearers usually are rather quick in interpreting what speakers meant, it seems unlikely that a hearer ‘waits’ until the 170 Utterance Meaning and Communicative Sense – Two Levels or One? end of an utterance, before he starts inferring what the speaker may have meant. On the other hand, assuming mutual parallel adjustment makes the claims for differentiating explicature as a separate level of meaning that is different from Grice’s what is said and necessary for the derivation of implicatures appear less convincing. That is, if the temporally preceding determination of a level of meaning that can be characterised as the proposition expressed by the speaker in fact is not a necessary basis for the drawing of further inferences, then why make the point that Grice’s level of what is said is not rich enough in meaning to form that basis? If, contrary to Grice’s assumptions, the basis for CIs does not even have to be a full proposition, why does RT argue for explicatures in the first place? Moreoever, it is not clear how individual steps of the determination of explicatures and implicatures in terms of mutual adjustment are to be characterised. Unfortunately, Carston’s remarks on the point are not very clarifying either: ‘The mechanism that mediates the inferences from logical form to communicated propositions is one of “mutual parallel adjustment” of explicatures and implicatures, constrained by the comprehension strategy. The result should consist of (sets of) premises and conclusions making up valid arguments, but the important point is that the process need not progress strictly logically from the accessing of premises to the drawing of conclusions. For instance, a particular conclusion, or type of conlusion, might be expected on the basis of considerations of relevance and, via a backwards inference process, premises constructed (explicatures and implicatures) which will make for a sound inference to the conclusion.’ (Carston 2002a, p. 139) Note that it is not obvious how the comprehension strategy constrains the mechanism of mutual parallel adjustment. Moreover, and as already noted above, the very notion of explicature and the criteria by which it may be differentiated from implicatures, are not clear. Thus, RT replaces the condition for something to count as an implicature, namely that it be determined on the basis of a fully-propositional what is said, by the condition that it be not a development of the semantic form of the utterance.38 One way of testing that this is not the case is by application of the scope criterion39 : implicatures are assumed not to fall within the scope of logical operators. If some pragmatically determined aspect of meaning does fall within the scope of some logical operator, it is taken not to be an implicature, but rather part of what is said or the explicature of an utterance. Thus, consider, for instance, the sentence in (145a) (taken from Borg 2009), which is typically understood to mean (145b), unless the context prevents this. (145) a. Jill blew the whistle on poor practices at work and was sacked. b. Jill blew the whistle on poor practices at work and was sacked because of that 38 We already saw that ‘development’ cannot be understood in terms of entailment (cf. BurtonRoberts 2005). 39 Recanati (1989, p. 325) calls this scope principle. Problematic Phenomena 171 Thus, the causal relation holding between what is expressed by the co-ordinated clauses is pragmatically contributed and constitutes, in Grice’s terms, a truthconditionally irrelevant GCI. However, if the sentence in (145a) is embedded in a larger structure involving a logical operator such as a conditional, the GCI appears to be truth-conditional after all. (146) a. If Jill blew the whistle on poor practices at work and was sacked, then she is entitled to compensation. b. If Jill was sacked and blew the whistle on poor practices at work, then she is entitled to compensation. Thus, according to RT and based on the scope principle, such GCIs actually are not implicatures but rather part of what is explicitly expressed by an utterance. However, and as already mentioned, it is an assumption only that implicatures cannot fall within the scope of logical operators. Having said this, note that according to Chierchia (2004) SIs – when they are present – do fall within the scope of logical operators. However, under his approach SIs as such actually are no longer a pragmatic phenomenon, but rather are grammatically determined. Thus, the optionality of SIs is built into the grammar by the modification of the definition for functional application.40 In other words, if SIs are present, they are already part of the SF delivered by the semantics component. As such, SIs in Chierchia’s theory are not implicatures in the usual sense. Generally, Levinson’s account and RT are similar in that both assume the phenomenon Levinson calls GCIs to contribute to the truth-conditional content of an utterance. This is also true of Chierchia with respect to SIs. The major difference between the two former approaches with respect to GCIs is that while Levinson assumes of GCIs that they are implicatures, RT takes these meaning aspects to belong to the explicature of an utterance. Thus, RT argues that although Levinson’s GCIs are pragmatically derived meaning aspects, they are not implicatures (since they are subject to the Scope Criterion). In contrast, for Chierchia, SIs at least, already form part of the semantic form of an utterance as determined by the semantics component. Moreover, RT both rejects Levinson’s assumptions that GCIs belong to a separate level of meaning distinct from both utterance meaning and communicative sense as well as Levinson’s and Chierchia’s claim that they are, as it were, automatically and locally triggered. (Mostly) Empirical Evidence Concerning GCIs The claims made about the nature of the interpretation process by the two types of theory (i.e. default and local vs. contextually driven and global generation of GCIs) have actually been empirically tested and it seems that the assumption about GCIs as being default interpretations that are locally triggered cannot be verified (e.g. Noveck and Posada 2003, Bezuidenhout and Morris 2004, Reboul 2004, Breheny et al. 2006). Thus, Noveck and Posada (2003) carried out an experiment that measured reaction times as well as ERPs for subjects’ truth/falsity judgements of sentences such as the following. 40 Recall the definition of FA in (144) which is an ‘if x then y else z’ instruction. 172 Utterance Meaning and Communicative Sense – Two Levels or One? (147) a. Some houses have bricks. b. Some elephants have trunks. c. Some crows have radios. Of particular interest are sentences such as in (147b), as these are the sentences which are assumed to give rise to the scalar implicature ‘not all elephants’. Thus, they are underdeterminate in that they can either be understood logically as ‘At least some elephants have trunks.’, or as pragmatically implicating that ‘Not all elephants have trunks.’ The results show that subjects overwhelmingly judged patently true sentences as in (147a) as true and patently false sentences as in (147c) as false. However, with sentences of the type in (147b), judgements were equivocal, with some subjects judging these sentences with respect to their assumed semantic or logical meaning as true, and others judging them based on the implicature carried by the sentence as false. However, there was clear within-subject consistency, with a subject either judging underdeterminate sentences as false or as true. Moreover, what was interesting with respect to this last group of sentences was the fact that, regardless of the judgement given by the subjects, there was no difference in the ERP measured. The ERP of interest here was the N400, which is taken to indicate semantic integration.41 Whether or not a subject judged an underdeterminate sentence as true or false, the resulting N400 was flat in both cases. In contrast, both the N400s for patently true and patently false sentences were steeper, which indicates that these items take more semantic integration than the underdeterminate items. Even when concentrating exclusively on the performance of those subjects who judged the underdeterminate sentences as false – thus, presumably deriving the implicature ‘not all’ – the results remain the same. That is, with that subject group the N400 for the patently true and patently false sentences was significantly larger than the N400 for the underdeterminate sentences, showing that no particularly strenuous semantic integration of a GCI was taking place. In contrast, there was considerable difference in the time subjects took to make a true vs. a false judgement with respect to underdeterminate sentences, with false judgements taking almost twice as much time as true judgements. This result can be taken to reflect the fact that, contrary to what Levinson assumes, deriving a GCI is in fact effortful and that it does not take place as early during the interpretation process as to allow it to be reflected by an N400, signalling semantic integration.42 While Noveck and Posada (2003) use a single sentence verification task, Breheny et al. (2006) take into account the role of context for the derivation of GCIs. In particular, an experiment was carried out that tested whether or not in a neutral context – thus a context that neither facilitates nor precludes a particular scalar implicature – scalar implicatures are computed. According to RT they should not be, as their derivation depends on the context just as much as for PCIs. In contrast, a default approach such as Chierchia’s or Levinson’s would predict that the respective GCI is triggered, namely by default. With respect to the notion of ‘neutral 41 The EEG in the experiment was recorded for 1100ms, starting 100ms before presentation of the last word in each sentence. 42 Note, however, that this generalisation may not hold, since what was tested in the experiment were scalar implicatures only. Thus, it might be that other types of GCI behave differently and might lead to different results (cp. Garrett and Harnish 2008). Problematic Phenomena 173 context’, Breheny et al. (2006) point out that a neutral test concerning the ‘some X’ construction using isolated sentences (as was done by Noveck and Posada 2003) actually is not possible, due to potential topic effects. Thus, in many languages, there is a tendency to give old information sentence initially. Therefore, presenting an isolated sentence such as (147b) to subjects might trigger contrastive assumptions and thus lead to the inference of the implicature. (147b) Some elephants have trunks. This tendency, however, is rather mild for English, since here word order is fixed. Therefore, the experiments were carried out in Greek, which has flexible word order, thus giving rise to ‘. . . a far stronger tendency . . . for grammatical subjects in sentence-initial position to be construed as old information and objects in a non-initial, post-verbal position to be construed as unrelated to a contextual issue’ (Breheny et al. 2006, p. 446). Thus, consider the English translations of example stimuli below, where the ‘some x’ phrase occurred either sentence initially or sentence finally; the former triggering the topic effect for the critical phrase, whereas the latter did not. Moreover, to generally control for potential order effects, Breheny et al. (2006) designed stimuli that included the explicit trigger only for a some but not all reading. (148) a. Some of the consultants had a meeting with the director. The rest did not manage to attend. b. Only some of the consultants had a meeting with the director. The rest did not manage to attend. c. The director had a meeting with some of the consultants. The rest did not manage to attend. d. The director had a meeting with only some of the consultants. The rest did not manage to attend. The phrase the rest at the beginning of the sentence following the sentence with the ‘some x’ construction was used to force the some but not all reading. The assumption was that if GCIs are context-driven rather than interpreted in default, then for a sentence including a ‘some x’ construction and presented in a truly neutral context, the GCI should only arise once the forcing phrase is encountered, leading to longer reading times for that phrase. Breheny et al. (2006)’s results favour RT’s approach to GCIs in that they are interpretable as showing that GCIs are not triggered in neutral contexts.43 Thus, subjects took longer reading the rest in the non-topic implicit condition as compared to all other conditions. Thus, contrary to Chierchia’s and Levinsons’ assumptions, the experiments mentioned above could neither verify the assumption that the processes that create GCIs apply locally nor that they apply by default, that is, automatically. Rather, the results seem to support a context-driven appoach, such as RT. However, Borg (2009) argues that it is not all that clear whether that really is the case. On the one hand, the time delay found by Noveck and Posada (2003) for the processing of scalar 43 As before, this particular experiment tested only scalar implicatures. 174 Utterance Meaning and Communicative Sense – Two Levels or One? implicatures may pose a problem for a general explanation in terms of a relevancetheoretic comprehension procedure. On the other hand, it seems that RT cannot explain the differences found concerning adults’ vs. children’s recovery of scalar implicatures (Noveck 2001). Recall that RT actually treats scalar implicatures not as implicatures but rather as part of what is explicated by an utterance of a particular sentence involving scalar expressions. Borg argues that . . . if the recovery of enriched scalar terms is held to be effortful and timeconsuming, so too should the recovery of other kinds of pragmatically enriched explicatures. Yet it seems unlikely that RT theorists would want explicature recovery in general to be marked by time-delay (Borg 2009, p. 76). Another problem for treating scalar implicatures as part of the explicature of an utterance is the fact that children give utterances involving scalar expressions much less often a pragmatic interpretation than adults (e.g., Noveck 2001). Given that ‘. . . both adults and children are in possession of the right reasoning processes to generate both explicatures and implicatures (i.e. both possess the same relevancedirected mechanism)’ (Borg 2009, p. 76), it is not clear why these differences should arise. This question becomes even more pressing, considering that children perform similarly to adults when it comes to employing pragmatic enrichment processes in overcoming semantic underdeterminacy, which also results in enriched explicatures for utterances. Borg (2009) offers an alternative explanation of GCI data, and scalar implicatures in particular, which is based on her minimal conception of semantics (see section 3.2.3). Due to Borg’s minimal characterisation of semantics, it is clear that GCIs cannot be semantic. However, she also recognises the fact that there is some intuitive difference between GCIs on the one hand and PCIs on the other. Thus, GCIs should not be treated as on a par with once-off inferences, ‘. . . since they do not rely on access to the current speaker’s state of mind’ (Borg 2009, p. 79). The difference between PCIs and GCIs, then, is captured by the fact that they result from distinct kinds of cognitive processing. More specifically, Borg (2009) suggests that during the interpretation of an utterance, the output of the minimal semantic component serves as input to two further systems: the general pragmatic system that derives PCIs and a more limited system that recovers GCIs. Of this latter system she assumes that it makes use of statistical facts concerning what speakers intented when they used particular expressions on former occasions. Borg (2009, p. 79, footnote 20) suggests that ‘. . . GCIs are a kind of abstraction from . . . PCIs: an agent starts by grasping a pragmatically enhanced reading of a scalar term as a PCI but later they come to grasp such readings simply through knowing how words are commonly used’. Thus, similarly to Levinson, Borg assumes that GCIs belong to a level of meaning that is distinct both from the level of utterance meaning as well as the level of what is meant. However, in contrast to Levinson, Borg neither assumes that GCIs arise on the basis of particular heuristics nor that they arise locally. Moreover, Borg (2009) has to concede that with such an approach, the truthconditional effects of GCIs on utterance meaning have to be treated as rather apparent than real. However, the approach does offer an explanation as to why we Problematic Phenomena 175 intuitively take GCIs to have a truth-conditional effect on utterance meaning. That is because these meaning aspects are, in a sense, automatically derived, both independently of the actual context as well as the actual speaker intentions in making a particular utterance. Furthermore, Borg (2009) argues that her account of GCIs also fits the experimental results mentioned before. Thus, the time delay noted for the processing of scalar implicatures is explained by the assumption that they are inferences that are drawn after the truth-conditional content of an utterance has been computed, so, contrary to Levinson’s assumption, they are not local.44 Moreover, the fact that the younger a child is, the less likely it is to interpret scalar implicatures pragmatically is explained by the assumption that the younger a child is, the less exposure it will have had to ‘usual uses’ of scalar terms, thus lacking the necessary premise for an automatic recovery of GCIs that is free of speaker intentions. The independence of speaker intentions of GCIs is also central to Burton-Roberts (2006)’s discussion of the notion of cancellation. Thus, recall that traditionally, one of the properties of CIs is that they are assumed to be cancellable. As BurtonRoberts (2006) convincingly argues, this property actually only holds of GCIs. That is, a speaker cannot explicitly cancel a PCI, the reason being that PCIs are always intended. They only ever arise in particular contexts. Thus, consider the following small conversation. (149) Max: Do you ever speak to Charles? Ann: I never speak to plagiarists. In that particular context, Max will take Ann to have conveyed both that Charles is a plagiarist and that she never speaks to him. The question is, whether she could actually explicitly cancel these PCIs without seeming contradictory. Burton-Roberts (2006) argues that she could not add ‘I’m not suggesting that Charles is a plagiarist, mind you – in fact, I love talking to him’. Similar points must be made with respect to RT’s category of explicature, of which we already saw that they may be clarified but not cancelled.45 Incidentally, Burton-Roberts (2006) suggests that RT’s assumption that (at least some) GCIs actually are part of an utterance’s explicature is the reason why RT has to assume that the latter are cancellable in the first place – because they may involve such pragmatically determined meaning aspects as Grice called GCI and of which we independently know that they have the property of being cancellable.46 However, we already saw that there is a good reason against assuming that explicatures generally are cancellable and that is that they are taken to be the proposition the speaker intended to express. Burton-Roberts (2006)’s consequence is to assume that GCIs are a ‘real’ phenomenon and that they are implicated rather than explicated by an utterance. Thus, Burton-Roberts (2006) assumes that GCIs are different from both explicatures as well as PCIs in that the communication of the latter two by a speaker 44 For experimental results that support such a view, see the discussion of Huang and Snedeker (2009, in press) below. 45 For a similar suggestion with respect to the level of what is said and using the term ‘revisability’, see Korta (1997). 46 Note, however, that originally the idea was that it is a characteristic of all pragmatically informed inferences that they are cancellable. 176 Utterance Meaning and Communicative Sense – Two Levels or One? is always taken to have been intended. In contrast, GCIs arise independently from and regardless of the speaker’s intentions. In characterising GCIs, Burton-Roberts (2006) refers to Gazdar (1979)’s notion of im-plicatures, that is, potential implicatures that only become actual implicatures of an actual utterance, if they are not inconsistent with contextual information. Thus, potential GCIs are assigned automatically to the relevant linguistic expressions. However, they only become actual GCIs when these expressions are uttered and if there is no evidence that the speaker did not want to communicate them. Such evidence may be provided either by the speaker explicitly cancelling the potential GCI or by contextual information that is inconsistent with the potential GCI. In those cases where the speaker does want to convey a particular GCI, she does not have to do anything special, as the hearer will assume that the GCI is part of what the speaker intended to communicate since the speaker did nothing to prevent him from making that assumption. Similarly to Borg (2009), Burton-Roberts also assumes that GCIs are implicatures, that is, they are drawn on the basis of the utterance meaning of a speaker’s utterance. They are cancellable in principle, however, if they are not cancelled – either explicitly or by the context – they will become actual implicatures and as such form part of what the speaker is taken to have intended to convey. Arguably, that is why they seem to have a truth-conditional effect in complex sentences. However, as Borg (2009) points out, they do not in fact contribute to the truth-conditions of the uttered sentence, but rather to speaker meaning. Thus, there is no circularity in Grice’s definition of CIs. Nevertheless, GCIs are special in that they are, in some sense, drawn automatically, although this does not happen locally. Note that the results of experiments reported by Huang and Snedeker (2009, in press) can be taken to support such a view, at least as far as scalar implicatures are concerned. In Huang and Snedeker (2009), the authors report experiments making use of the visual world paradigm. Their subjects heard stories, which where accompanied by a visual display. The stories where about four characters (two girls and two boys) and two types of objects, which were distributed amongst the characters. For instance, one girl might end up with a proper subset of one type of object (e.g., socks), whereas the other girl was given all of the instances of the other type of object (e.g., soccer balls). After hearing the stories, subjects were given instructions such as the following. (150) Point to the girl with two/three/some/all of the socks. Subjects’ gaze direction and time were measured while they were given instructions and carried them out. Note that when the subjects hear the expression girl, the potential referent is restricted to two possible entities. Once subjects heard a numeral, the intended referent was immediately disambiguated, as the girls in the stories never had the same amount of objects. The same happened with the quantifier all, since there was always only one girl with all the instances of one type of object. The critical condition, however, was that with the quantifier some. The hypotheses Huang and Snedeker (2009) wanted to test are the following. If some is interpreted ‘semantically’ (meaning ‘some and possibly all’), the intended referent would only be disambiguated at the final noun of the sentence. If, however, some is locally given its ‘pragmatic’ interpretation (‘some but not all’), the intended referent should also Problematic Phenomena 177 be immediately disambiguated (it cannot be the girl with all of the instances of one object type). Thus, either subjects show no difference in the time they take to fix on the intended referent in the different conditions, thus suggesting that they locally retrieve the interpretation of some as ‘some but not all, or they take longer in the some condition, showing that they initially interpret some semantically as ‘some and possibly all’ and thus have to ‘wait’ in order to disambiguate the intended referent. The results show that the potential pragmatic interpretation of some is not computed locally. Thus, while subjects take approximately 400ms from the onset of the quantifier to lock their gaze on the intended referent in the two, three and all conditions, they take about 800ms in the some condition. However, the time at which they identify the intended referent is not clearly the time at which they interpret the final noun in the sentence. That is, it is possible that subjects fixed on a referent before the disambiguating information from the final noun in the sentence becomes available. The experiment reported in Huang and Snedeker (in press) was designed to further investigate this possibility. Thus, the time between onset of the quantifier and disambiguating expression was increased by using compound noun pairs. (151) Point to the girl with two/three/some/all of the ice cream sandwiches/cones. The results of this experiment clearly showed that although scalar implicatures are not computed as part of initial sentence processing, they are drawn online, before the disambiguating information becomes available.47 While these results, once again, disconfirm Chierchia’s and Levinson’s assumption that GCIs are locally retrieved default interpretations, Huang and Snedeker (2009, in press) note that they are not totally unproblematic for RT either. Recall that in RT drawing pragmatic inferences are assumed to involve as little cognitive effort as possible while at the same time resulting in as many cognitive effects as possible. However, as Huang and Snedeker (2009) point out, in their experimental task, it seems that little is gained by drawing the critical inference, since ‘. . . the referent of the quantified phrase was always lexically disambiguated, making the implicature unnecessary for full comprehension’ (Huang and Snedeker 2009, p. 409). Nevertheless, subjects seem to have drawn the inference, suggesting that it involves low cognitive costs. Thus, Huang and Snedeker (2009, p. 409) caution that ‘. . . these data suggest that Relevance theory may need to acknowledge the possibility that scalar implicatures have a preferred status relative to other pragmatic inferences’. Note that the time-course of scalar implicature interpretation observed by Huang and Snedeker (2009) may allow for an alternative interpretation of the results attained by Breheny et al. (2006). Thus, recall that in the critical ‘neutral context’ condition, the scalar expression occurs sentence-finally in the first sentence, followed by the ‘forcing phrase’ sentence-initially in the second sentence (cf. 148c), repeated below. (148c) The director had a meeting with some of the consultants. The rest did not manage to attend. 47 Cp. Storto and Tanenhaus (2005) for similar results concerning the inclusive/exclusive interpretation of or. 178 Utterance Meaning and Communicative Sense – Two Levels or One? Breheny et al. (2006) predicted that if GCIs are interpreted locally by default, the reading time for the forcing phrase should not be any different from that in the other conditions, as the GCI would be drawn immediately on the encounter of some. In fact, the reading time for the forcing phrase in the ‘neutral context’ condition turns out to be longer than in the other conditions. Breheny et al. (2006) interpret this result as showing that the GCI was in fact only drawn at the time the forcing phrase was encountered. More importantly, they assume that it is the forcing phrase that triggers the drawing of the scalar implicature. However, although Huang and Snedeker (2009)’s results confirm that the drawing of scalar implicatures is not a local process, nevertheless, it does seem to take place automatically although delayed with respect to the onset of the scalar expression. Thus, it is possible that the longer reading time for the forcing phrase – which, in terms of presentation immediately followed the phrase involving the scalar expression – simply reflects the delay in drawing the scalar implicature, rather than indicating that the implicature was actually triggered by the forcing phrase.48 While most of the studies discussed so far concentrated on scalar implicature, Garrett and Harnish (2007) investigated the interpretation of expressions argued by Levinson to give rise to GCIs based on the I-heuristic. More specifically, the authors concentrated on three types of implicit meaning aspects – which, following Bach (1994a) they call implicitures – namely ‘location’ (cp. 152a), ‘possession’ (cp. 152b) and ‘time’ (152c) implicitures. (152) a. It’s raining. b. John cut a finger. c. I’ve had breakfast. In a first experiment, Garrett and Harnish (2007) investigated the strength of the three types of impliciture associated with particular complex expressions out of context. Subjects were presented stimulus sentences auditively. After each stimulus sentence, a single question word was visually presented, with two probe words printed below. The question word was appropriate to the respective impliciture type. Similarly, the probe words were such that one was the word generally used to make explicit the presumed impliciture and the other a word that is not part of the impliciture but syntactically and semantically comparable to the impliciture probe word. Thus, subjects heard ‘Somebody said: It’s raining.’, followed by the question word WHERE? on a screen with two possible answers from which to choose, in this case HERE and THERE. Both the choice subjects made as well as the time they took to make it were noted. Moreover, in addition to the impliciture sentences, other similarly structured sentences which, however, did not give rise to implicitures were tested as a control. Results showed that for impliciture sentences subjects overwhelmingly chose the impliciture probe word as the appropriate answer to the 48 Compare Garrett and Harnish (2009, p. 100) who make a similar general point concerning the interpretation Breheny et al. (2006) give their results: ‘Is the [scalar implicature] processing following the explicit trigger [i.e. only K. B.] faster than it is following the implicit trigger? If the latter is delayed (but does occur), the interpretation of the increased readings times at the forcing region [i.e., the rest K. B.] would be affected.’ Problematic Phenomena 179 question posed and they did so quickly. Thus, ‘. . . [t]he results confirmed the special status of the implicitures’ (Garrett and Harnish 2008, p. 81). In a second experiment, the authors used the materials from the first, however, the critical sentences were now placed at the end of small stories, intended to either enable or cancel the respective impliciture that had been elicited out of context before. In this experiment, the stimulus texts were presented visually and subjects read through them at their own pace. Reading times for stimulus texts were measured as was response time for question answering as in the first experiment. There were no statistically significant differences in the reading times for the two contexts or the target sentences in the respective contexts. Response times, however, differed in that subjects responded more quickly to the probe questions in enabling contexts than they did in the cancelling contexts. Overall, accuracy of responses was very high for both contexts. These results suggest that the impliciture was computed in both contexts, but had to be cancelled in the cancelling context. More generally, the results of this experiment once again suggest that ‘. . . contextual penetration of sentence interpretation should be distinguished from the interpretation that relies on sentence form . . .’ (Garrett and Harnish 2008, p. 84). Summary In summary of the section on conversational implicatures, the following points can be made. It is useful to differentiate GCIs from PCIs, both for theoretical as well as empirical reasons. A theoretical differentiation is based on two intuitive facts about GCIs, which make them different from PCIs, namely, that they seem to arise independently of a particular context as well as regardless of speaker intentions. However, as we saw, both these intuitive facts have to be treated with caution. Thus, on the one hand, there are arguments that scalar implicatures at least do need a ‘special context’ to be triggered, if information-structural aspects such as the topic effect are to be counted among contextual properties. Similarly, a point can be made that GCIs have to be consistent with speaker intentions in order to ‘survive’ or become ‘actual’ implicatures of an utterance, in which case they will be taken to have been intended by the speaker after all (cp. Burton-Roberts 2006). As already mentioned, assuming such a behaviour of GCIs might explain the fact that they seem to have a truth-conditional effect on utterances that give rise to them, without necessarily having to assume that they arise locally and are already part of the utterance meaning of an utterance. Empirical investigations of GCIs also suggest that they are ‘special’. Across a range of experiments using different testing methodologies, it has been shown that, on the one hand, the influence of GCIs on the interpretation process is not immediate but rather delayed, but on the other hand, they ‘arise’ even in cancelling contexts. While the former result suggests that GCIs are not locally drawn default interpretations, the latter result suggests that although their inference is effortful and time-consuming, GCIs still differ from PCIs in that they seem to influence the interpretation process even in those cases where they are not intended by a speaker. As concerns the coming about of GCIs, we saw that Levinson assumes that they are based on a number of heuristics, whereas Borg and Bach assume they are, in a sense, abstracted or standardised from PCIs as a result of their frequent occurrence 180 Utterance Meaning and Communicative Sense – Two Levels or One? in a particular reading. As was argued above, this latter view is straightforwardly compatible with the different results found for children and adults concerning the drawing of scalar implicatures. Moreover, Garrett and Harnish (2007)’s experimental results also favour a standardisation view of GCIs, although they are cautious that this may be due to their particular interpretation of Levinson’s I-heuristic. Recall also that Levinson himself characterises the level of utterance-type meaning as subsuming such phenomena as standardised inferences, although he suggests that GCIs are distinct from those. Overall, I would suggest, the experimental results concerning the processing of putative GCIs support the idea that there might be a special level of meaning at which sentences are paired with preferred interpretations and to which GCIs, amongst other phenomena, belong. However, the fact that results of experiments testing SIs and so-called I-based implicatures differ from one another, does suggest the possibility that the phenomena collectively called GCI might in fact be distinct and thus exhibit different properties.49 4.1.4 Speech Acts The notion of speech act captures the fact that usually in using language we carry out specific actions. Thus, the speaker of (153) does not only utter a complex expression and thus expresses a specific meaning, but, more importantly, under certain conditions he also enters into a specific kind of commitment, namely to do what he said he would do. (153) I promise to come. Speech acts in general are assumed to be a phenomenon that can be identified at the level of what is meant/communicative sense; at a level, that is, where criteria are taken into account that are specific for general human interpersonal interaction. Thus, criteria are considered that may influence the hearer’s interpretation of what the speaker meant – and thus, what his intentions were in making a particular utterance – which are not necessary for figuring out what the utterance as such means. Recall the example from Bierwisch, repeated below as (154), which, depending on the context of utterance, may be used to make a promise, statement or as a threat. Regardless of which speech act this is taken to be a token of, the hearer is able to understand what the utterance as such means. Moreover, as Bierwisch observes, although (153) involves a performative verb (or illocutionary force indicating device), the sentence may be felicitously used in a situation in which the intended speech act is not one of promising. Thus, the kind of speech act that a speaker intends to carry out with his utterance, very much depends on the utterance situation. Although there are linguistic indicators that are conventionally associated with particular speech acts, it is questionable whether this potential should be considered as being part of the semantics of such expressions. Moreover, in order to carry out a speech act of promising, it is not even necessary to use the expression promise at all. Thus, as we have already seen, an utterance of (154) made in the appropriate circumstances may also count as a promise.50 49 This point will be taken up in section 5.2.1. This does not, of course, hold true for all types of speech act. Thus, for the action of naming to be successfully carried out, an utterance has to be made by the appropriate person in appropriate 50 Problematic Phenomena 181 (154) I will come tomorrow evening. The situation is similar as regards the correlation of particular sentence types with speech acts, giving rise to the notion of indirect speech act. Thus, the three basic sentence types declaratives, interrogatives and imperatives are usually associated with the speech acts of asserting or stating, asking or questioning and ordering or requesting, respectively. If the sentence type of an utterance matches the intended illocutionary force of that utterance, the speech act carried out is said to be a direct speech act. If, however, there is no such match, the speech act is said to be indirect. In (155), for instance, the speaker overtly asks a question, but covertly performs a speech act of requesting. (155) A: Can you pass the salt? B: Yes. [B passes A the salt] Thus, the utterance meaning of (155) can be described as asking for the truth or falsity of the related affirmative proposition ‘You can reach the salt’. However, the speaker’s intentions would not be adhered to, if in answer to (155), B merely said yes and did not also actually pass the salt. Thus, A most likely intended to make a request by his utterance. The content of his actual utterance guides the hearer to what it is specifically, that A is requesting. This is so because what the speaker seems to be questioning is one preparatory condition for requesting of the hearer that he pass the salt. One possible analysis of indirect speech acts, proposed by Searle (1975), assumes that such speech acts have two illocutionary forces associated with them. Whereas one of them is characterised as the literal, but in this case secondary illocutionary force (in the example above: questioning), the other is the non-literal, but primary illocutionary force (here: requesting). Thus, we also find the notion of literal and non-literal meaning aspects applied at the level of what is meant/communicative sense, with respect to the notion of indirect speech act (e.g. Capone 2004, Huang 2007). Searle furthermore assumes that in the interpretation of indirect speech acts, some inferencing is involved. He suggested that the latter, similarly to conversational implicatures, are based on the Gricean principle of co-operation and the maxims of conversation. Interestingly, quite a number of what are called indirect speech acts have become conventionalised to such an extent that naive speakers, on a first blush, may not even be aware that they actually are not direct. The degree of conventionality becomes apparent, if one compares, e.g. (155) with the sentences below, which, although they express the same meaning as (155), cannot be used to request of the addressee that he pass the salt.51 (156) a. Are you able to pass the salt? b. Do you have the ability to pass the salt? circumstances, where that utterance has to include the expression christen or name. 51 Thus, note that whereas in (155) it would be possible to add please, thereby overtly indicating the intended speech act, this is not permissible in the examples in (156). 182 Utterance Meaning and Communicative Sense – Two Levels or One? To account for such facts, Morgan (1978) introduced the idea of short-circuited implicature: although the implicature from the literal illocutionary force of an utterance to the non-literal, intended illocutionary force is in principle calculable, in cases of conventionalised indirect speech act, the implicature is not in fact (fully) calculated. Thus, Morgan, as Busse later on, uses the notion of Gricean implicature to account for the process of conventionalisation (see also König 1988, Rolf 1995). So, traditionally, indirect speech acts are treated as conversational implicatures, albeit as to some extent conventionalised ones. Nevertheless, similarly to the traditional treatment of metaphor and irony in terms of conversational implicature, the assumption is that the direct speech act is determined first and the indirect speech act afterwards and on the basis of the direct speech act if that does not fit the situation at hand. However, as for metaphor, it has been shown also for indirect speech acts that they are not necessarily determined after potential direct speech acts. Thus, Shapiro and Murphy (1993) showed that processing of indirect speech acts compared to direct speech acts is not necessarily secondary and optional. Similarly, in a study investigating the comprehension of indirect speech acts in aphasic subjects, Hirst et al. (1984) found that subjects with anterior left-hemisphere damage did comprehend the indirect speech act, although they were not able to determine the direct speech act on the basis of the syntactic structure of the utterance. In their first experiment, Shapiro and Murphy (1993) tested the traditional view concerning the interpretation of indirect speech acts, which they call the serial model, as well as an alternative model, which they call the parallel model. As we saw before, according to the serial model, indirect speech acts are derived after the direct speech act has been computed and judged as not fitting the context of utterance. The former, thus, depends on the latter and is only optionally derived. In contrast, the parallel model assumes that direct and indirect meanings of an utterance are processed in parallel. That is, in terms of processing, indirect meanings are not based on direct meaning in the sense that the former are derived only after the latter has been. Moreover, this model assumes that accessing a potential indirect meaning is not optional, but rather obligatory. Subjects were presented sentences of which they were asked to decide whether they had a plausible direct meaning. There were four stimulus conditions, namely, plausible direct (+D) and implausible direct (-D) meaning and plausible indirect (+I) and implausible indirect (-I) meaning. The items were designed such that each could be understood as having a direct and/or an indirect meaning or neither (e.g., +D-I, +D+I, -D+I and -D-I). (157) +D-I: Where is Sandy this weekend? +D+I: Don’t you think this is exciting? -D+I: Can you stop whistling? -D-I: Can text books tell time? The serial model predicts that the availability of a plausible indirect meaning should play no role in subjects’ decision on the plausibility of the direct meaning. That is, reaction times in conditions (+D-I and +D+I) should be equally fast. The parallel model, however, leads to the following predictions. Although subjects were specifically asked to judge whether critical items had a plausible direct meaning only, Problematic Phenomena 183 if the critical item has a plausible indirect meaning, this should be automatically processed, leading to longer reaction times in the (+D+I) condition as compared to the (+D-I) condition, as the subject will have to decide which of the two available interpretations is the direct one. The results of the experiment support the parallel model. Thus, subjects took longer to judge whether a target sentence had a plausible direct meaning if that sentence also had a plausible indirect meaning, showing that the indirect meaning interfered. Thus, even though in the experiment there was no context given to bias an indirect meaning interpretation and subjects were specifically asked to concentrate on direct meanings only, they nevertheless processed plausible indirect meanings and actually could not ignore them. This shows that indirect meanings are not secondary to direct meanings and they are not optional but rather obligatory. Hirst et al. (1984) investigated further the question of whether processing of the direct speech act is necessary for the understanding of the indirect speech act. Their subjects were anterior aphasics, who generally seem not to be able to use knowledge of syntax in their interpretation of utterances. Thus, ‘. . . they may not be able to parse a “Can you . . .” utterance as a question since it requires sophisticated syntactic competence to distinguish a question from a declarative sentence’ (Hirst et al. 1984, p. 28). However, anterior aphasics might make use of contextual information as well as of their world knowledge and knowledge of the individual words in order to interpret a particular utterance. Recall once more that the serial model holds that the processing of an indirect speech act depends on the prior processing of the direct speech act, which is then judged as not fitting. This model predicts, then, that anterior aphasics should not be able to interpet an utterance as an indirect request, when they are not able to process the direct question. In contrast, the parallel model predicts that anterior aphasics should be able to interpret an utterance as an indirect request, employing context, even if they are not able to interpret the direct question, employing context, because the two interpretations are independent of one another. Subjects watched colour-videotaped episodes, with two characters involved in different kinds of activity in different situations. In each episode, one of the characters asked a ‘Can you X?’-question, to which the other character responded either by starting to carry out some action or giving a simple verbal yes/no answer. All in all, four different responses were possible: appropriate or inappropriate questionanswering and appropriate or inappropriate action carried out. Subjects were asked to judge whether the response to the ‘Can you X?’-question was appropriate. The results show that the anterior aphasics did understand indirect requests, since in those cases they indicated at a rate above chance that an action would be appropriate in response. However, in those cases where ‘Can you X?’ was intended as a direct question, anterior aphasics judged the appropriate ‘yes’ answer as actually appropriate in only 34% of the cases, indicating that they did not understand the ‘Can you X?’-question as a genuine question. Thus, Hirst et al. (1984)’s results suggest once again that the prior processing of the direct speech act is not necessary for the understanding of an indirect speech act. To summarise, it seems that the interpretation of indirect speech acts neither requires the prior determination of the direct speech act made by the speaker of a 184 Utterance Meaning and Communicative Sense – Two Levels or One? particular utterance, nor is it optional. That is, if an indirect meaning is available, it will be processed even in situations in which there is no context which could be said to have triggered it (cf. Shapiro and Murphy 1993). In a way, then, indirect speech acts can be compared to familiar metaphors, although in contrast to the former, the latter are not based on utterance meaning. That is, it seems that like familiar metaphor, indirect speech acts of the type investigated by the studies cited here, might be assumed to be conventionalised to such an extent that they are processed in parallel to the potential direct speech act expressed.52 In fact, the particular characteristics of such, in a sense, conventionalised indirect speech acts can be captured by assuming that they, like GCIs are part of Levinson (2000)’s level of utterance-type meaning. Thus, like GCIs, conventionalised indirect speech acts are based on utterance meaning, they arise independently of the particular context of utterance and they have to be cancelled if they are not intended. 4.2 Differentiating What is Said from What is Meant As we already saw, the different approaches to utterance interpretation that assume at least two separate context-dependent meaning levels (e.g., Grice, Bierwisch) assume that the one level (i.e., what is meant/communicative sense) is, in some sense, based on the other (i.e., what is said/utterance meaning). However, it is not clear how exactly this relation is to be explicated. Thus consider again the example (67). (67) A: Would you like to join us for lunch? B: I had a very large breakfast. EX: I had a very large breakfast today. CI: B doesn’t want lunch. It seems obvious that what B actually wants to convey in uttering I had a very large breakfast is that he does not want to have lunch. However, this is not explicitly expressed in his utterance of I had a very large breakfast. Rather, it is the hearer’s task to arrive at the speaker’s intended meaning and that is only possible via a number of pragmatic processes applying to the meaning of the complex expression B uttered. Thus, although in this example, B can be taken to have meant what he said, what he meant is not exhausted by what he said. Moreover, in order to determine the sum, as it were, of what B meant, the hearer first has to determine the meaning of the utterance that B made, in order to derive potential conversational implicatures that the speaker might have wanted to convey. Note that examples such as the one above contrast with examples, where it is not totally clear which role the speaker’s actual utterance plays in the process of determining the meaning she intended to convey. Thus, consider another, wellknown example of conversational implicature (taken from Levinson 1983). 52 However, the studies cited did not test whether indirect speech acts are also initially processed even when there is contextual information available that biases the interpretation towards a direct one, a behaviour characteristic for familiar metaphors. Differentiating What is Said from What is Meant 185 (158) A. I do think Mrs Jenkins is an old windbag, don’t you? B. Huh, lovely weather for March, isn’t it? B’s utterance in (158), at first blush, seems totally irrelevant to the topic that A introduced with his utterance. However, this example can be analysed along the lines that B’s utterance might in fact be relevant in that he may have conveyed the proposition that he does not like to talk about people behind their back or that the person A just talked about, is standing right behind A. However, the question, for both these potential interpretations, is which role, if any, the actual content of B’s utterance plays in the determination of what B actually meant. What can be said about this example, regardless of how to answer that question, is that the meaning of B’s utterance can be determined, in a sense, independently of what he intended to convey by it. That is, arguably, A will be able to conclude what B intended to communicate by his utterance regardless of whether his utterance appropriately describes the state of the weather at the time of utterance. More specifically, even if B says what he does on an uncomfortably rainy and windy day, A will be able to figure out what B primarily intended to convey was something other than an irony (as this is not relevant in the situation). Thus, B’s utterance may or may not be taken to be part of what he meant by making it. Lastly, consider an example of what might be assumed to be an idiom of German. (159) Wir stehen hinter dir ‘We stand behind you’ The sentence in (159) has a very salient non-literal meaning out of context amounting to ‘we support you’. A likely context in which the sentence in (159) can be felicitously uttered is one in which it qualifies as a promise on the part of the speaker. It is rather unlikely that (159) with the reading ‘we support you’ would be used by a speaker to, e.g., warn the addressee. Having said this, the sentence in (159) can of course be used to issue a warning, however, the meaning associated with it would probably be something along the lines of ‘We are watching what you’re doing’, another nonliteral meaning of the sentence. Thus, in a case like (159), recognising the speaker’s intention in making the utterance, can actually contribute to understanding the utterance meaning itself. The fact that it is not clear how what is said/utterance meaning relates to what is meant/communicative sense has been argued to show that actually no differentiation between these two putative levels of meaning is possible. Thus, Borg (2004b) and Cappelen and Lepore (2005) argue against the claim put forward by Bach (1994a) that indirect speech reports capture what is said by a particular utterance. Rather, indirect speech reports seem to pick up on all sorts of aspects associated with a particular utterance, thus, again constituting evidence that what is said cannot be strictly differentiated from what is meant. In contrast, Recanati (2004) assumes that what is said should be differentiated from what is meant and he proposes that the difference be captured by distinguishing two general types of pragmatic processes, where the one type leads to what is said by an utterance and the other to what is meant by the speaker of that utterance. Similarly, Bierwisch (1980) claims that a differentiation of utterance meaning from communicative sense is necessary, although 186 Utterance Meaning and Communicative Sense – Two Levels or One? for different reasons, namely the fact that it is possible both to understand what a speaker said without necessarily knowing what he meant by his utterance and vice versa. 4.2.1 What is Said/What is Meant and Indirect Speech Reports Thus, Borg (2004b) as well as Cappelen and Lepore (2005) agree on the claim that indirect speech reports cannot, across the board, be taken as reflecting what was said (in the Gricean sense) by the reported utterance/speaker. It should be noted that Carston (2002c) concurs in this view. In fact, the idea that starting from an utterance’s meaning, or what is said, one can recover the semantic meaning of the sentence used to make the utterance, is not an idea that is generally held by contextualists.53 Borg (2004b) shows that amongst the possible set of such indirect speech reports of the utterance of a particular sentence are such which clearly are not semantically related in content to the semantic meaning of the sentence originally used. Thus, consider the sentence in (160), where the sentences given in (161) exemplify possible indirect speech reports for a particular utterance of (160) by some person Jim in different possible contexts54 . (160) Blair lives at No. 10. (161) a. Jim said that Tony Blair lives at No. 10 Downing St, London, UK. b. Jim said that the current Prime Minister lives at No. 10. c. Jim said that that man lives there. d. Jim said that Baby Leo’s father lives in No. 10. e. Jim said that the most right-wing Labour leader to date secured power where other, more left-wing, predecessors had failed. f. Jim said that he knows about British politics. There are several points to note here. First, although the reports all include the expression x said that . . ., this does not ensure that what is reported here is anything close to what the theoretic notion of what is said tries to capture. In this connection, recall the experiments mentioned above on the difference between minimal proposition (involving something close to Borg’s liberal truth conditions), proposition expressed (involving actual truth-conditions) and purely implicated meaning, where the results showed that naive speaker-hearers do not understand and use the term say in the way Grice intended it in his theoretical considerations. Moreover, contrary to what is predicted by Grice’s characterisation of the levels of what is said and what is meant and their interaction, the results of the experiments mentioned suggest that interpreters do not generally differentiate between what a speaker said 53 As mentioned, e.g. Bach (1994a) holds this view. It should be noted, however, that in order for the intended correlation between what is said by an utterance and an indirect speech report thereof to hold, certain restrictions on the form of the indirect speech report apply. The critique remains, though, that such restrictions obscur the fact that in reality, indirect speech reports may take a multitude of forms. 54 Examples are taken from Borg (2004b) and, due to their subject matter, are somewhat dated. Differentiating What is Said from What is Meant 187 and what he implied with his utterance. In my opinion, this is clearly to be seen by examples such as (161e) and (161f), which would seem odd at the least, if they were intended as indirect speech reports of what Jim ‘said’ in the theoretical sense. They seem to involve uses of the term say that might, e.g., be paraphrased as follows. (162) a. Jim meant that the most right-wing Labour leader to date secured power where other, more left-wing, predecessors had failed. b. Jim intended to show that he knows about British politics. This, however, would indicate that what we have here, if anything, is a report on what the ‘reporter’ thinks the speaker meant or intended by making the particular utterance he did. Thus, it does not capture what is said but rather what was meant. The fact that for Borg all the different mentioned indirect speech reports are possible/appropriate lead her to assume that there is no clear-cut boundary between the levels of what is said and what is meant. Thus, whether some particular meaning aspect counts as part of what is said or what is meant may differ from occasion to occasion. To repeat, Borg, takes what is said to be part of speaker meaning and as such, it is totally pragmatically determined and therefore one cannot expect that from knowing what was said by a speaker one can deduce the meaning of the sentence the speaker uttered. Thus, there is only a one-way relationship between sentence meaning and what is said, namely such that the latter is based on the former. The question, again, is what exactly it means for what is said to be ‘based’ on sentence meaning (cf. RT’s ‘development’). This question is especially interesting here, as Borg argues that the minimal semantic content of a sentence, its sentence meaning, need not be part of what the speech act means, that sentence is used to carry out. That is, semantic content need not be a part of speech act content. However, if what is said is part of speaker meaning, where the latter can be assumed to correspond to speech act content, and sentence meaning is not part of speaker meaning, then what exactly is the relation that sentence meaning and what is said are in? Moreover, from Borg’s discussion it rather seems as if it were not at all clear whether what is said actually can be taken to be a level of meaning distinct from what is meant and if that is the case, it is actually questionable whether we can state how we get from sentence meaning to what is said. Now, in Cappelen and Lepore (2005)’s approach, the speaker generally is not much of an authority when it comes to what it was she said when making a particular utterance. As already noted, Cappelen and Lepore (2005) state that indirect speech reports cannot capture the technical notion of what is said, however, they seem to think that they do capture what the speaker said in a wider sense. Thus, they list a few examples of reports on the so-called ‘Smoking Gun’ utterance55 made by 55 ‘When you get in these people, when you get these people in, say: “Look, the problem is that this will open the whole Bay of Pigs thing, and the president just feels that,” ah, without going into the details . . . don’t, don’t lie to them to the extent to say there is no involvement, but just say this is sort of a comedy of errors, bizarre, without getting into it, “the president believes that it is going to open the whole Bay of Pigs thing up again, and ah because these people are plugging for, for keeps and that they should call the FBI in and say that we wish for the country, don’t go further into this case.” Period. That’s the way to put it, do it straight.’ 188 Utterance Meaning and Communicative Sense – Two Levels or One? President Nixon, all of which they claim are appropriate and report on what the speaker said. Here are just three of them. (163) a. Nixon told Haldeman to tell the CIA to tell the FBI not to pursue their investigation into the Watergate burglary. b. Nixon wanted the CIA Director, Richard Helms, to thwart the FBI’s probe of the Watergate burglary by saying it was a CIA operation. c. Nixon told Haldeman to break the law. These reports exemplify Cappelen and Lepore’s wide understanding of what is said. In fact, we already noted above that possible indirect reports of a speaker’s utterance may pick up on what is said by a speaker but also on what was meant. Cappelen and Lepore do not distinguish these aspects within speech act content, since they assume that there is no possible way of doing so. From the discussions so far and within different frameworks, it is clear that it is indeed difficult to differentiate between the two levels. Nevertheless, I find the argument they give to defend their particular view of indirect reports stating what a speaker said not entirely convincing. Here are a couple of objections we have encountered . . .. Did Nixon really say what we attribute to him? Did he, strictly speaking say it? Did he literally say it? Let’s see: We know that this particular utterance was the cornerstone of the impeachment case against Nixon. [. . .] Imagine the absurdity of a defense of Nixon that he didn’t strictly speaking, ask the CIA to block the FBI investigation. (Cappelen and Lepore 2005, p. 196-7) It goes without saying that it would have been absurd to defend Nixon on the lines that he did not strictly speaking request the CIA to block the FBI investigation, but this does not get rid of the discrepancy there is between what a speaker may convey or mean by making a particular utterance and the meaning the utterance as such can be attributed. Note also that this is not where the problems with indirect speech reports end. Thus, apart from possibly capturing what is said and what is meant by (the speaker of) an utterance, they may pick up on aspects of an act carried out by a speaker he himself may not be aware of. Thus, (163c) picks up on the fact that the CIA’s interfering with an FBI investigation constitutes an illegal act. In this particular context, it is reasonable to assume that Nixon was aware of that fact, but still, neither did he tell Haldeman (in the sense of ‘order’, which seems to be intended here) to break the law, nor is it clear whether it is appropriate in this scenario to asumme that (163c) constitutes part of what Nixon meant by his utterance. Another example may make this point more obvious: a German hiker meets a British hiker on tour through Scotland. Towards evening they get to a bigger town, go to a grocery store to get some food, etc. and then, since the weather is nice, decide to have their dinner on a bench in a picturesque public place with nice views. While they unpack their proviant, the German says (164) to the Briton, handing him a can of German beer. (164) Here. Have a beer. Differentiating What is Said from What is Meant 189 What the German does not know is that in many public places in Britain consuming alcohol is against the law. In this context, would it be appropriate to report the German has having invited or urged the Briton to break the law? Probably not. However, one might object that this scenario does not compare to the Nixon scenario since we can assume of Nixon that he did know that if Haldeman did what Nixon told him to, he would be breaking the law. But even if we change the context in our example such that the German actually knows about this particular law, I would content it still is unappropriate to report what he meant, let alone ‘said’, as ‘He invited/urged the Briton to break the law.’ Of course, if the Briton excepts the beer and drinks it, it follows that he does break the law, due to the logical relation holding between carrying out this particular action in Britain (i.e., consuming alcohol in a public place) and breaking the law in Britain. However, it is questionable, whether this is part of what the speaker intended with his utterance.56 Generally, the problem with indirect speech reports can be traced back to the apparent vagueness involved in the expression say. Thus, recall once again the results of Nicolle and Clark (1999)’s first experiment, which had three conditions, asking subjects to choose a paraphrase of the speaker’s utterance that best reflected either what the sentence meant the speaker used, what the words meant the speaker used in his utterance, or what the speaker communicated by his utterance. The results across the different conditions did not significantly differ, suggesting that subjects did not differentiate between utterance level meaning and speech act content. However, these results do not show that in interpreting natural language utterances, people do not build up a level of context-sensitive meaning distinct from speech act content. Rather, what they show is that the expression say is vague as it can be used to refer to both these levels. Moreover, as the use of say in some of the indirect speech reports cited above shows, it can even be used to pick up on an entailment of what the speaker actually said/meant, although this entailment may not actually be what the speaker intended to convey or what was foremost in his mind when making the reported utterance. In fact, the original speaker of some reported utterance may not even have been aware of the reported entailment (cf. Gross 2006). Thus, indirect speech acts neither exclusively capture the technical what is said by an utterance, nor do they simply capture what the speaker intended to convey or meant by his utterance. However, that fact as such is no evidence that no differentiation between the levels of what is said and what is meant is possible or necessary. 4.2.2 Primary vs. Secondary Pragmatic Processes Recanati (2004) assumes that a differentiation of types of pragmatic processes is possible and that the distinction allows to differentiate between the levels of what is said/utterance meaning and what is meant/communicative sense. Thus, he differentiates between so-called primary and secondary pragmatic processes. Primary pragmatic processes are characterised by the fact that they take place at a subpersonal level57 and determine what is said by an utterance. Such processes take 56 57 Cp. Gross (2006) for similar worries. The terms sub-personal and personal level are taken from Dennett (1969). 190 Utterance Meaning and Communicative Sense – Two Levels or One? place unconsciously; we are not aware of them and thus cannot reflect upon them. They comprise such processes as the resolving of reference, fixing indexicals, disambiguation and free enrichment (and possibly other pragmatic processes). The important characteristic primary pragmatic processes have in common is that they do not require a fully propositional form to apply to. Rather, they (may) apply to the sub-propositional output of the linguistic process of semantic composition, which yields the semantic form of a sentence. The final and conscious output of these unconsciously proceeding processes then, is what is said, which is of a fully propositional form and equatable with the truth-evaluable content of an utterance. Note, however, that Recanati’s notion of what is said differs from Grice’s traditional characterisation in that he assumes processes like free enrichment to contribute to what is said. Moreover, the primary processes are such that they do not take into account speaker intentions. In contrast, secondary pragmatic processes take place on a personal level; they are accessible by our consciousness and we can reflect upon them. Thus, Recanati claims, normal interpreters are usually both aware of what has been said and of what has been implied. However, while they are consciously able to recapitulate how they arrived at what was implied by some utterance on the basis of what has been said, they cannot ‘calculate’ how they arrived at what was said (or the utterance meaning) on the basis of the surface sentence used to make that utterance. Thus, they are not aware of the primary pragmatic processes applying to the semantic form of the sentence uttered; it is only the final result of the application of those processes that is consciously available to them. Moreover, they are able to recapitulate the inferential connection between what is said and what is implied. The common characteristic of secondary pragmatic processes is that they do need a fully propositional form to apply to. That is because from the fact that p, they proceed to inferentially derive further propositions (implicatures). Moreover, secondary processes do take speaker intentions into account. Note that Recanati’s characterisation of primary and secondary pragmatic processes as being derived on the basis of sub-propositional and fully propositional forms, respectively, implies that there is a temporal ordering to their application – hence their names – primary processes take place first, whereas secondary processes take place afterwards. Recanati’s view is similar to that held by Ruhl (1989), who differentiates between conscious and unconscious cognitive processes. As the names suggest, while the former are consciously available to us, the latter are not. Thus, Ruhl for example suggests that the reason why we assume of a particular expression to have a certain literal meaning is that we are not conscious of the fact that there are processes in our minds, which abstract over uses and thus create default interpretations. However, an expression might be used with a ‘non-default’ (non-salient) but literal reading nonetheless. With respect to Recanati’s differentiation between primary and secondary pragmatic processes, it is important to recall that he rejects what he calls the ‘minimalist’ view (following the Gricean view) of what is said, where what is said is the minimal proposition derivable form the semantic form of an utterance. Thus, as already mentioned, he argues that the minimalist what is said is not even consciously available to a normal hearer-speaker. Thus, for B’s utterance in (165) Differentiating What is Said from What is Meant 191 (165) A: Would you like to join us for lunch? B: I had a very large breakfast. Recanati predicts that a normal hearer-speaker if asked to state what B actually said with his utterance, would come up with what Carston called the explicature of the utterance, namely ‘The speaker had a very large breakfast today’. Thus, the assumed minimal proposition ‘The speaker had a very large breakfast prior to the time of utterance’ is not available to the hearer, as it is the result of sub-conscious pragmatic processes, enriching the semantic form, but not the final result of all such processes. In fact, it is questionable whether this is a level of meaning that interpreters derive at all during interpretation. Recanati’s notion of ‘what is said’ is that it is the actual conscious output of the application of primary pragmatic processes to the semantic form of an utterance and the only consciously available ‘proposition expressed’ by the utterance of a particular sentence58 . Semantics lexical meaning semantic composition sub-propositional Pragmatics actual truth-conditions fully propositional sentence meaning primary pragmatic processes + narrow context basis for further pragmatic inferences what is said secondary pragmatic processes + broad context what is meant Figure 4.2: Recanati’s view We already saw that this view is supported by the results of experiments carried out by Gibbs and Moise (1997). Thus, the results show that apparently subjects did not have anything like a minimal proposition available, rather, when asked what a speaker actually said with his utterance, they would choose the enriched version of the traditional, Gricean what is said, namely one corresponding to RT’s explicature. Note that, in the last experiment in Gibbs and Moise (1997), which we have not mentioned yet, it was tested whether in a relevant context, subjects would actually choose a rather minimal proposition as being ‘said’ by a speaker. The results showed that they did, however, in this case what they chose still was what was said by the respective utterance in Recanati’s enriched sense. It just happened to coincide with what minimalists would refer to as what is said, that is, the minimal proposition, due to the very specific contexts of utterance. Recall, however, that Gibbs and Moise (1997)’s results were criticised on the grounds that they presupposed of the subjects that they make use of the differentiation between what is said in the enriched sense 58 This is why Recanati’s approach is called the availability based approach. See above for the formulation of the availability principle. 192 Utterance Meaning and Communicative Sense – Two Levels or One? and what is implicated, when, in fact, it might be that they are guided by some totally different criterion. Thus, Nicolle and Clark (1999)’s study was geared to show that naive speakers do not normally differentiate between an utterance’s explicature and its communicative sense in the technical understanding. Rather, what seems to matter to them, albeit not consciously so, is the way in which an utterance achieves relevance in a particular context. Thus, if what is predominantly relevant in a given context is a communicated implicature, naive interpreters might pick the implicature as what has been said by the speaker. That is, they do not share the technical understanding of such notions as ‘say’ and ‘communicate’ presupposed by theorists. Thus, while Recanati’s differentiation between primary pragmatic processes and secondary pragmatic processes was supposed to capture the different levels of meaning explicature and what is meant, respectively, it seems that such a differentiation actually is not possible, at least if speakers’ intuitions are assumed to be a guide when characterising these different levels of meaning. However, it is quite possible that, in a particular context of utterance, the naive interpreters are not aware of a particular proposition on which they consciously base further inferences they draw. Moreover, even though speakers may not be consciously aware of a level of meaning that corresponds to the minimal proposition of an utterance that does not mean that this level may not play a role during interpretation after all. Note also that Recanati’s characterisation of primary and secondary pragmatic processes actually cannot be applied to differentiate between explicatures and implicatures (cf. e.g. Carston 2002a). This is because RT claims that implicatures need not be drawn on the basis of a full proposition. Rather, the fact that RT assumes mutual parallel adjustment of explicatures and implicatures rules out the claim that processes leading to the latter type of pragmatic inference need a full proposition to apply to. Furthermore, RT assumes that speaker intentions already play a role for the determination of an utterance’s explicature (i.e., the proposition the speaker intended to express). Also, due to mutual parallel adjustment, there cannot be any inherent temporal ordering to the application of pragmatic processes leading to explicatures or implicatures. Moreover, RT does not assume that the processes retrieving implicatures from what the speaker said necessarily are part of a personal level, which is characterised by allowing for conscious reflection. The assumption is that there is one pragmatic system that has the function of recovering the speaker’s intention in making a particular utterance, however, this system is taken to be a subpersonal system, not necessarily requiring conscious reflection on what the speaker’s intention(s) might have been.59 Carston (2002a, p. 145) gives the following example to make clear the difference in assumptions about its interpretation according to Recanati’s view in contrast to RT. (166) A: Do you want something? B: Yes. I’ve run out of paper. B answers A’s question in the affirmative. Thus, it is likely that A, having 59 For a thorough discussion of Recanati’s differentiation of primary and secondary pragmatic processes see Carston (2007). Differentiating What is Said from What is Meant 193 processed this first part of B’s reply, at this point forms the ‘. . . anticipatory assumption schema [B wants ]’ Carston (2002a, p. 145). The second part of B’s reply, then, provides a completion for that schema – ‘B wants some paper’. Now, according to RT this completed proposition happens to be an implicature of B’s second utterance, as it is not the result of a ‘development’ of the semantic form of B’s second utterance. Note, however, that if B’s second utterance had been ‘Some paper’ or ‘I want some paper’, then A’s completed assumption schema would rather have counted as an explicature of that utterance. This fact is unproblematic for an approach to verbal interpretation as RT, since it assumes that there is only a single pragmatic system responsible for the derivation of both explicatures as well as implicatures. However, according to Recanati’s differentiation between different types of pragmatic processes, the completed assumption schema [B wants some paper] would have to be viewed as the result of one distinct type of pragmatic process in the case where it turns out to be an explicature, and of another distinct type of pragmatic process in the cases where it is an implicature of B’s utterance. ‘According to Recanati’s view . . ., the understood answer, that is the completed schema, ‘B wants some paper’, must be achieved by fundamentally different kinds and levels of processes in the two cases . . .. However, this is not supported by intuitions about the two possibilities, and, in the absence of compelling arguments, it is difficult to see why we should adopt this view.’ (Carston 2002a, p. 145) A similar critique is provided by Borg (2006)’s review of Recanati (2004). She points out that Recanati’s category of primary pragmatic processes also includes such processes as free enrichment, which are characterised by the fact that they are optional and actually do apply to full propositions. She argues that this fact blurs the usefulness of a distinction between primary and secondary pragmatic processes. Thus, consider the following example. Jack asks Jill whether she wants to go and have something to eat and she replies: (167) I’ve had breakfast. Amongst other things, according to Recanati, it is a primary pragmatic process that constrains the timescale within which Jack will interpret Jill to have had breakfast. The point here is that the process that constrains the timescale might deliver different results. It could lead to Jack interpreting as what is said by Jill that ‘Jill has had breakfast today’. If that is the case, then a further implicature, derived via a secondary pragmatic process will be necessary to provide Jack with a relevant answer to his question: ‘Jill does not want to go and eat’. However, nothing speaks against assuming that what the process constraining the timescale results in leads to the following what is said by Jill: ‘Jill has had breakfast recently enough to make eating currently undesirable’, which would be relevant enough to provide an answer to Jack’s question. Thus, in this latter case, no further inferences would have to be derived. ‘Yet, if both interpretations are possible, then it looks as if something which counts as an implicature to one hearer might count simply as what 194 Utterance Meaning and Communicative Sense – Two Levels or One? is said to another, in which case we don’t seem to have anything like a determinate notion of the literal meaning of an utterance.’ (Borg 2006, p. 3)60 As we saw above, Recanati’s characterisation of primary pragmatic processes as being sub-personal and secondary pragmatic processes as being personal is not helpful here either, as speakers, contrary to Recanati’s assumption do not seem to have concrete and static intuitions about the difference between the theoretical notions of what is said and what is implicated. That is, the experiments carried out on the distinction did not verify Recanati’s availability principle – speakers in interpreting an utterance do not necessarily and consciously differentiate between what a speaker said and what he merely implicated.61 4.2.3 What is Said/What is Meant and Distinct Knowledge Systems Recall that Bierwisch differentiates between three distinct levels of meaning, expression meaning, utterance meaning and communicative sense. This differentiation reflects his view that there are three distinct knowledge systems involved in producing what he calls ‘communicative linguistic behaviour’, namely, knowledge of language, everyday knowledge and knowledge about social interaction. That is, the basic meaning SEM (E) of an expression E, be it simple or complex, is determined by the language system alone. The meaning of an utterance exemplar t of E, M (t) is determined by what is contributed to it by SEM (E) on the one hand, and the context of utterance C, which in turn is determined by our everyday knowledge, on the other. Both context and utterance meaning are mental representations that are determined by the conceptual system. The third level of meaning, the communicative sense CS (t) of an utterance exemplar t of E is constituted by the contextually determined meaning M (t) of that utterance exemplar, plus knowledge about social interaction in general which determines the specific interactional situation IAS in which the utterance takes place. As mentioned already, in contrast to Grice, Bierwisch does not assume that utterance meaning necessarily forms part of communicative sense. Generally, he claims that the differentiation between the two levels is required since it is possible for a hearer to interpret correctly what the speaker meant by what he said, without knowing what was actually said. Bierwisch cites the example of the American prisoner of war in Italy, who recites the following line of a German poem with the intention of showing that he is German: (168) Kennst Du das Land wo die Zitronen blühen. ‘Do you know the country where the lemons bloom.’ It can be argued that in this case, if neither speaker nor hearer are speakers of German, they might not understand the meaning of the utterance of (168), but the 60 Where the phrase literal meaning here should be interpreted as the utterance meaning or what is said by an utterance. 61 See also Garcia-Carpintero (2006) for another critique of Recanati’s availability principle. Differentiating What is Said from What is Meant Semantics (simple) expression meaning semantic composition sub-propositional reference resolution fixing indexicals disambiguation ... utterance meaning + narrow context basis for further pragmatic inferences conversational implicature speech acts + broad context ... Pragmatics does not dependent on underspecified (complex) expression meaning Conceptual System fully propositional 195 communicative sense Figure 4.3: Bierwisch’s view on the relation of utterance meaning and communicative sense hearer of (168) might nevertheless discover what the speaker’s intention might have been in uttering it. Note that the reverse can be the case as well, namely that it is possible for a hearer to understand an utterance without necessarily grasping what the speaker of that utterance meant by it. Thus, imagine a situation where A is helping B to prepare for an exam. They have both learned the subject matter together and additionally, A has asked B exam-like questions for B to check how far he has progressed in learning. As A takes leave of B he utters the sentence in (169). (169) I will come tomorrow evening. Irrespective of B’s understanding of what A has said, he might nevertheless be unsure whether in uttering (169) A wanted to promise that he will learn together with B tomorrow, or whether he threatened that he will get angry if by tomorrow evening B has not advanced in memorising the subject matter. It should be noted, however, that although an argument can be made that in principle the interpretation of the two levels of meaning what is said and what is meant is independent of one another, more often than not, determining the level of what is meant does depend on knowing the proposition expressed by the utterance made or is at least influenced by what the hearer assumes was expressed by the utterance made. In fact, this can be taken to be the normal communication situation. Bierwisch criticises further that aspects of meaning at the level of communicative sense are usually viewed as directly related to linguistically determined meaning. For instance, speech act theory is taken to be ‘. . . an extension of the theory of meaning in natural language’ (Wunderlich 1977, p. 243). However, as Bierwisch points out, this view obscures the fact that there is a basic distinction to be made between language and communication. Thus, Bierwisch shows that in order to explain such a phenomenon as performative utterances, it is not sufficient to assume that they have to be analysed in terms of felicity conditions and that the notion of truthconditions is not applicable here. As he puts it: ‘. . . under certain conditions a sentence having an explicit performative formula as its topmost clause specifies a 196 Utterance Meaning and Communicative Sense – Two Levels or One? truth condition that can be satisfied by the speech acts in which the sentence is used.’ (Bierwisch 1980, p. 12). (170) I promise to come. The truth condition specified by (170) can be paraphrased very loosely as ‘The speaker promises to come’. This condition is fulfilled only if the speaker actually performs the speech act of promising in using the sentence in (170), that is, if the interactional setting of the utterance of (170) is such that it qualifies as the setting necessary for the performance of an act of promising. That is, whether an utterance of an explicit performative such as (170) constitutes the speech act of promising, depends on the interactional setting in which it is uttered. This view also explains why it is possible to use a sentence that includes an explicit performative formula without actually performing a speech act. (171) A: Why are all my parties such a success? a. B: I promise to come. In this exchange, the interactional setting for B’s utterance can be assumed to be such that it does not qualify as a setting necessary for the felicitous performance of an act of promising. What is crucial for examples such as the above is that there is a difference to be made between meaning as it is determined by linguistic and contextual factors and the interpretation of a linguistic utterance as a specific type of social interaction. Therefore, ‘Speech act theory is a branch of the theory of communication, viz. that involving linguistic utterances, rather than a part of the theory of language.’ (Bierwisch 1980, p. 3). Turning back to the idea that speaker intentions play a particular role at the level of communicative sense, it should be noted that in figuring out why a speaker made a particular utterance, hearers make use of information resulting from their conceptualisation and evaluation of the social relationship between themselves and the speaker. Thus, although determining a speaker’s intention in making a certain utterance involves making assumptions about what the speaker had in mind, the assumptions made also crucially depend on how the hearer, possibly unconsciously, judges his relationship to the speaker. This idea is captured in Schulz von Thun (1982)’s communication-psychological approach by assuming that there actually are four sides to a message (utterance) that have to be taken into account: the level of content (‘Sachebene’), the aspect of relation (‘Beziehungsebene’), the aspect of self-revelation (‘Selbstkundgabe’) and the aspect of appeal (‘Appellseite’). Thus, consider the example in (172) (adapted from Schulz von Thun 1982) of how meaning is determined by the social setting in which communication takes place. The context for the small verbal exchange in (172) is the following: A man (A) and a woman (B) are sitting in a car, which the woman is driving. (172) A: Du, da vorne ist grün. ‘Listen, it’s green ahead.’ B: Fährst Du oder fahre ich?! ‘Are you driving or am I?!’ Differentiating What is Said from What is Meant 197 The four sides to A’s utterance, as B interprets them, might be (crudely) paraphrased as follows: (173) Level of content: the traffic lights are green Aspect of relation: You need my help. Aspect of self-revelation: I’m in a hurry. Aspect of appeal: Drive faster! Schulz von Thun (1982) points out that it is the interpreter’s free choice to react to any of the four sides of the message. In the example in (172), the side that B is apparently verbally reacting to is the aspect of relation. However, in the situation stated, she might also react to the aspect of appeal of A’s message non-verbally at the same time by actually driving faster. Her verbal reaction and the way it is shaped might only turn against the aspect of relation that in her opinion is conveyed by A’s message. In the case where A actually did not intend his utterance to be interpreted on the aspect-of-relation side as ‘You need my help’, the question is if we would like to say that the communication between A and B was unsuccessful, although B actually did react as intended to one side of the message by driving faster. That is, assuming that successful interpretation of a speaker’s meaning involves determining the speaker’s intention in making an utterance, does not capture the complexity of such a task, as we have to assume that there are various sides to verbal interaction, highlighting different aspects of the overall intention of the speaker. More specifically, we might want to say that in making an utterance, the speaker does not necessarily have only one intention, but rather several, possibly on different levels and with different prominence. Be that as it may, the important point to note here, once again, is that the type of information apparently necessary for determining what a speaker might have intended with his utterance seems to be distinctly different from the knowledge neccessary to determine what the speaker’s utterance as such means. This hypothesis is also corroborated by findings concerning autistic childrens’ understanding of the speaker’s intention in making a particular utterance in contrast to their understanding of what the utterance as such means. Thus, MacKay and Shaw (2004) tested autistic children’s understanding of various types of figurative language as well as the likely intentions with which they were employed by the speakers. Although the autistic children did understand what the utterances meant more or less well, they had considerably more problems in determining why a speaker should use a particular type of figurative language in the first place. In this connection, recall RT’s assumption that the determination of explicatures and implicatures actually is subject to mutual parallel adjustment, where assumptions about what the speaker meant by his utterance seem to influence how that utterance is interpreted in the first place. Although viewing the evolving of the interpretation process in these terms is not unproblematic – as it is not clear on which basis, then, to differentiate explicatures from implicatures and why to assume a richer level of utterance meaning than originally proposed by Grice in the first place – what it does capture is the fact that people do seem to understand utterances differently if they have a grasp on what the speaker might possibly have intended in making that utterance, than when they do not have a definite grasp on those intentions. For 198 Utterance Meaning and Communicative Sense – Two Levels or One? instance, autistic people tend to interpret language literally and a possible reason for that might be that they do not understand why a speaker said what he said. In other words, it might be that autistic people have problems considering a speaker’s possible intentions and the fact that the speaker actually persued a particular goal in making a particular utterance. However, while there may be differences between autistic and ‘normal’ people in determining what is meant by a speaker of an utterance, this fact alone does not tell us whether there really is a difference in what the two groups take to be the utterance meaning of that utterance. As we saw, one cannot rely on intuitions when it comes to spelling out what the speaker said vs. what he meant by his utterance. In MacKay and Shaw (2004)’s study, subjects were asked both for aspects of the background against which a particular utterance took place, as well as the reason the speaker might have had to make this particular utterance. For example, in the hyperbole condition, subjects where asked what they thought how many CDs the speaker of ‘I’ve got thousands of CDs’ actually has. In other words, they were asked to infer from the utterance the situation it was supposed to describe. Since the subjects did not take into account the possibility that the speaker might want to boast in making such an utterance, most of them assumed that the speaker does have a very large number of CDs. This is different for the control group of normally developed children, who judge the situation differently.62 However, that does not tell us how the two groups in fact interpreted the utterance as such. It might be that that interpretation does not differ, but that the difference in judgements simply is due to how these children differ in their evaluation of the situation in which such an utterance is made and which purpose it has. Having said this, I do not want to deny that hearers come to the interpretation task with more or less specific assumptions about what the speaker might want to convey in a particular discourse situation. It is just that I am reluctant to call these assumptions implicatures, since it seems to me that these kinds of inferences are different in kind from what is usually understood by the term implicature. Thus, in contrast to ‘real’ PCIs, such assumptions are not (necessarily) intended by the speaker to be inferred by the hearer and they are not (necessarily) the product of an inferencing process that is based on the meaning of a prior utterance on the part of the speaker. In fact, it seems to me that generally what a hearer’s assumptions in a particular utterance situation influence is the final interpretation of what he takes the speaker to have meant by his utterance (cf. the differences in judgement of ’normal’ and autistic children above). Recall once again Schulz von Thun (1982)’s idea that a message has several sides to it and the interpretation of one of them is based on the assumptions a hearer has as to the speaker’s opinion of him. For example, if B thinks that A thinks B is a looser, then it is possible that in a situation in which A, B and several others are planning a cycling tour they want to go on together and A says (174) to B about the potential tour they are just discussing, B might understand him to have meant 62 Thus, of the answers given by children in the control group, only two involved numbers higher than 50 (namely 70 and ‘hundreds’). In contrast, in the subject group, eight children gave numerical responses over 50. In fact, their answers involved such unlikely quantities as ‘million’, ‘zillion’ and even ‘infinity’. Differentiating What is Said from What is Meant 199 something like ‘You won’t be able to make it’ or ‘This is too difficult for you’. (174) This is a very difficult tour. In other words, particular assumptions hearers bring to the interpretation process may influence how they understand why a speaker said what he said, but it is not going to change their understanding of what it in fact was the speaker said. Thus, if in this particular situation what A wanted to convey – apart from the fact that he considers the contextually relevant tour as difficult – was in fact the idea that he thinks doing this tour will involve very careful planning (and that he specifically addressed himself to B because he actually thinks B is good at organising things), one might argue that communication between A and B was only partly successful in that B did understand what A said but was wrong about his reasons for saying what he did. Thus, as we saw above, assumptions the interlocutors have concerning other interlocutors’ opinions about one another may influence the interpretation of what a speaker meant by his utterance. Moreover, assumptions concerning the social setting in which a particular utterance takes place influence what the speaker will be taken to have meant with his utterance. Thus, Capone (2004) mentions the example of a teacher using (175) in addressing his favourite pupil Michelangelo who is prompting a classmate. (175) I saw you. Capone (2004) says of this example that even though the teacher might actually admire the fact that Michelangelo tries to help his fellow pupil, in the particular social setting in which his utterance occurs, it is clear that part of what the teacher means in saying (175) is that he wants Michelangelo to stop prompting.63 Note that the assumptions the hearer forms in a particular discourse situation are also crucial in explaining irony. That is, on a first blush, in cases of irony, the utterance made by a speaker does not seem to fit with the situation in which it is made. This is not to say that the hearer has clear-cut expectations of what a speaker is going to say, however, whatever the speaker does in fact say, the hearer will try to integrate with his conception of the discourse situation. This can be seen especially well in such cases of irony in which what the speaker said might actually also be taken to be (part of what is) meant by her, but where the meaning expressed by her utterance nevertheless in some way contrasts with the situation it is used in. (176) I love children who keep their rooms clean. [Said by a mother having just looked into the messy rooms of her kids.] Thus, the speaker of (176) can be taken to use the sentence she does descriptively, but at the same time her utterance in the situation it occurs in will be interpreted as 63 In fact, Capone (2004, p. 7) goes further in that he assumes that the utterance as such is going to be understood differently from its use in another context: ‘. . . this example nicely instantiates the view that the context is the total social setting in which the speech event takes place, the meaning of an utterance being determined by its place in an interactional sequence.’ It should be clear that I disagree. 200 Utterance Meaning and Communicative Sense – Two Levels or One? ironic, since she does not voice her negative attitude to the messiness of the rooms directly. In sum, the important point to note is that interpreting a particular utterance as ironic, as used to carrying out a specific speech act or as giving rise to particular PCIs involves contextual information beyond that needed to determine the utterance meaning as such. Moreover, in the majority of utterance interpretation situations, the proposition taken to be expressed by the utterance of a speaker is involved in determining what the speaker meant in making that utterance. However, assumptions about the intentions behind or the reasons for making the utterance the speaker did, do not influence a hearer’s interpretation of the meaning of an utterance as such, it only influences her interpretation of what she takes the speaker to have meant. 4.3 Summary One of the aims of the present chapter was to review the phenomena traditionally assumed to arise at the level of what is meant/communicative sense to see whether for their characterisation it is necessary to assume that they are based on a fully propositional utterance meaning. As we saw, for metaphor at least, this does not seem to be the case. Rather, arguments were given for assuming that metaphor actually belongs to the level of utterance meaning. Thus, on the one hand, metaphorical meaning usually is associated not with entire utterances, rather, it seems that it is simple expressions (or at least expressions below sentence level) that may receive a metaphorical interpretation. On the other hand, it was argued that metaphor can be treated along the same lines as metonymy (e.g., Dölling 2001) and other types of figurative language use assumed to arise at the level of utterance meaning (e.g., Sperber and Wilson 2008). Moreover, it seems that metaphor interpretation proceeds independently of the intentions the speaker had in making the utterance that includes the metaphor (e.g., Borg 2001). In addition, metaphor seems to differ in some important aspects from irony, which type of figurative language use was acknowledged as belonging to the level of what is meant/communicative sense. Thus, in contrast to metaphor, the interpretation of an utterance as ironic does seem to require a prior determination of a kind of ‘basic’ utterance meaning (e.g. Giora et al. 2007). Furthermore, interpreting an ironic utterance is cognitively more exacting than interpreting an utterance involving metaphor (cf. Happé 1993, Colston and Gibbs 2002, MacKay and Shaw 2004). What is similar in non-familiar metaphor and irony interpretation is that the ‘basic’ meaning the respective figurative meaning is based on stays activated to some degree and for some time during interpretation. What differentiates both metaphor and irony from conversational implicature is that in cases of the former the meaning understood differs from some ‘basic’ meaning, whereas in cases of the latter the inferred meaning aspects are added to what is understood. Thus, Grice assumed that conversational implicatures are based on some ‘basic’ utterance meaning and added to it. Generally, CIs originally were assumed to be both speaker intended and cancellable. However, as we saw, not all CIs exhibit these two features to the same degree. Thus, whereas PCIs only arise in (broad) contexts in which they are clearly speaker intended, GCIs are initially computed regardless of whether they are speaker intended or not. As a Summary 201 consequence of this (in)dependence on speaker intentions, PCIs actually are not cancellable, whereas GCIs are. What is special about GCIs is that if they are not cancelled, they consequently are assumed to have been intended by the speaker after all. Thus, what makes CIs and irony similar and differentiates them from metaphor is that both are based on some full utterance meaning and both are (ultimately) speaker intended.64 Another phenomenon undisputably taken to arise at the level of what is meant are speech acts. Thus, determining the speech act intended by a speaker in making a particular utterance allows the hearer to interpret how what the speaker uttered is to be taken. Of particular interest for our purposes are indirect speech acts, since they are traditionally characterised as involving a basic, literal illocutionary force and a derived, non-literal one. Arguments were given for not treating indirect speech acts in terms of conversational implicatures, since their determination – contrary to what was traditionally assumed – does not seem to necessarily involve a prior determination of a potential but non-fitting direct speech act. The second aim of this chapter was to find an answer to the question of what types of information have to be available to the interpretation process to determine phenomena taken to belong to the level of what is said and such taken to belong to the level of what is meant. To this end, we looked at some of the arguments given for and against assuming two distinct levels of context-sensitive meaning. That is, even though particular phenomena were characterised as being based on some level of what is said/utterance meaning, there are arguments that there actually are no clear criteria which allow us to differentiate between meaning aspects that are part of what is said/utterance meaning and such as are part of what is meant/communicative sense. Thus, Borg (2004b) as well as Cappelen and Lepore (2005) argue against the accuracy of indirect speech reports as a tool to capture what is said or the utterance meaning of the reported utterance. However, while their reservations are absolutely warranted, the fact that indirect speech reports cannot capture the technical notion of what is said by an utterance is not sufficient to show that a differentiation between what is said and what is meant is not in principle possible. Incidentally, the same point holds true concerning the results of experiments aimed at testing whether people differentiate between the respective two levels of meaning. As already mentioned, what the experiments in question showed is that naive speaker-hearers understand the expression say differently from its technical characterisation. Moreover, they show that people do not necessarily consciously reflect on the supposed difference. However, again, these facts are not by themselves sufficient to show that there actually are no two levels of context-sensitive meaning that can be differentiated from one another. Recanati (2004) proposes a differentiation between two general types of pragmatic processes, which is supposed to capture the difference between the two levels of what is said and what is meant. Whereas the sub-personal and thus, unconsciously operating primary pragmatic processes result in what is said by an utterance, the personal and thus consciously operating secondary pragmatic processes lead to what is meant by an utterance. Recanati (2004)’s differentiation of pragmatic processes in 64 Note, however, that the discussion of GCIs will be taken up once more in chapter 5, leading to a somewhat different characterisation. 202 Utterance Meaning and Communicative Sense – Two Levels or One? terms of whether they are consciously available predicts that naive speaker-hearers should be aware of the difference between what is said and what is meant (e.g., his availability principle). However, the results of experiments aimed at verifying this prediction show that it is inaccurate. Thus, a differentation of what is said from what is meant cannot rely on speaker intuitions. Bierwisch (1980) also assumes that a differentiation between what is said and what is meant is necessary and he makes reference to different knowledge systems that seem to be involved in determining the different levels of meaning. Thus, while the processes leading to utterance meaning make use of information provided by general background knowledge, the processes leading to communicative sense rely on information drawn from knowledge about social interaction. Generally, the two knowledge systems are independent of one another and so are, strictly speaking, the two meaning levels for the determination of which they are employed. However, more often than not the utterance meaning of what some speaker said plays a distinctive role in figuring out what the speaker meant in saying what he did, i.e., the communicative sense of his utterance. Nevertheless, there still is a difference between what an utterance means and what a speaker means or intends in making a particular utterance and experimental studies with autistic subjects corroborate the view that it is quite possible to understand the former without grasping the latter (cf. MacKay and Shaw 2004). Thus, generally, and following Bierwisch, the argument can be made that the contextual information the processes leading to what is said draw on is taken from general background knowledge, whereas the contextual information the processes leading to what is meant draw on is taken from a knowledge base about social interaction. However, this again raises the question as to the role of speaker intentions in interpretation, as reasoning about these is usually taken to draw on the latter type of knowledge system, i.e., that on social interaction. This issue will be taken up in chapter 5 (section 5.2). Chapter 5 Varieties of Meaning, Context and the Semantics/Pragmatics Distinction Recall that this thesis started out from the observation that although the terms literal meaning and non-literal meaning are extensively used in the semantics and pragmatics literature, usually there is no clear indication of what kinds of meaning aspects these two terms are actually assumed to pick out. Moreover, although there seem to be some sort of standard characterisations for the two terms, they are not used consistently with those characterisations and – what is worse – those characterisations can be shown to be inappropriate for the description of the phenomena the two terms are intuitively taken to pick out. Since, traditionally, the two terms were used to differentiate semantics and pragmatics from one another, this is a very unsatisfactory state of affairs. Thus, in this chapter, I will turn to alternative characterisations of literal meaning and non-literal meaning, before finally formulating my own (section 5.1). However, while my alternative characterisation of literal meaning and non-literal meaning captures the kinds of meaning aspects these notions are usually taken to refer to, it does not allow the two terms to figure in the characterisation of the differentiation between semantics and pragmatics either. Thus, in section 5.2, I will turn back to another notion traditionally used in that differentiation, namely that of context-(in)dependence. While in section (5.2.1), I will offer a proposal concerning the particular type of contextual information made use of by the process of free enrichment, in section (5.2.2), I will defend a view of the semantics/pragmatics distinction that actually does not make crucial reference to the notion of context-(in)dependence, but rather to the nature of the processes intuitively taken to belong to the individual systems. 5.1 Towards an Alternative Characterisation of (Non-)Literal Meaning When thinking about literal (vs. non-literal meaning) in natural language, it is interesting to look at an expression, which, supposedly and intuitively, expresses 204 Varieties of Meaning, Context and the Semantics/Pragmatics Distinction just that ‘meaning’ as its literal meaning. Thus, Israel (2002) gives a short but revealing survey of the development of the contribution made by the expression literally to the meaning of utterances in which it occurs. Whereas in its earliest usages in English it did refer to the meaning or sense of expressions taken to be, in some sense, ‘basic’ (cf. 177), the conditions of usage for literally where extended over time. (177) All those passages are not to be literally understood. Thus, Israel (2002) cites later examples, where literally seems to be used to indicate the quality of a speaker’s commitment to his utterance. This applies to examples as in (178), where literally is used to indicate that the speaker is committed to the truth of a literal interpretation of his utterance. (178) Most of these kids cannot think, they literally cannot come in out of the rain. However, it applies also to examples as in (179), where literally is used to emphasise a fitting choice of words, although the intended interpretation clearly has a derived character and thus cannot be taken to be literal in the sense of ‘basic’. (179) [In his music videos he] literally brings words to life; one of his favourite techniques is to superimpose song lyrics on a background image. Whereas in examples (178) and (179) literally is used to distinguish different possible interpretations of an utterance, there are also uses of literally, where it is used to differentiate between speech acts a speaker might carry out in making the utterance he does. (180) You wouldn’t understand. I don’t mean that as an insult, I mean it literally. Thus, the expression literally can be used to refer to the meaning or sense of expressions in their literal form; it can also be used to indicate the speaker’s commitment to the truth of a literal interpretation of his utterance and it can be used to pick out, arguably, what is in some sense considered to be the basic speech act amongst the different possible speech acts which can plausibly be performed by one and the same utterance. In addition, the expression literally can also be used to indicate the appropriateness of the use of expressions that will receive a figurative interpretation in the respective utterance. Turning back to the terms literal and non-literal meaning; in chapter 1 I claimed that these terms are used to refer to types of meaning arising at different levels of meaning as well. In fact, in chapter 2 I argued that their traditional characterisations are not consistent and that the notions actually do not capture the data they were intented to. A crucial point made in that chapter was that lexical meaning should actually be viewed as underspecified and that if there is some meaning aspect that might usefully be termed literal, this should not be looked for at the level of expression meaning. Furthermore, in chapters 3 and 4, a range of meaning aspects were discussed, which do not easily fit into the traditional literal/non-literal meaning dichotomy, such as the meaning aspects contributed to utterance meaning by processes of ad-hoc concept formation or free enrichment. Such meaning aspects Towards an Alternative Characterisation of (Non-)Literal Meaning 205 differ from what we traditionally and intuitively understand as non-literal meaning. At the same time, we would not necessarily want to characterise them as literal. A similar point can be made concerning the notion of sentence non-literality identified by Bach. That is, traditionally, we take non-literal meaning to capture such figurative language use as found in metaphor or metonymy, where it seems that the non-literalness is associated with a particular expression in the utterance made. Generally, the question arises, how these different meaning aspects are related to one another and in particular, how they relate – at least intuitively – to the notions of literal meaning and non-literal meaning. Thus, in this section I will first discuss two proposals for alternative characterisations of literal meaning and non-literal meaning, one of which specifically tries to capture the relations to the various meaning aspects identified in chapters 3 and 4 (i.e., Recanati’s). However, what both proposals have in common is that they assume the notion of literal meaning to be applicable at the level of expression meaning and also that the lexically given meaning of an expression is a full-fledged concept. As such, it is clear that they will not provide the characterisation of literal meaning I am after (as essentially context-dependent), but they are a starting point. In the final part of this section I will offer an alternative characterisation of the notions of literal and non-literal meaning, based on the data accrued in the preceding chapters. 5.1.1 Literal Meaning and Types of Non-literal Meaning Recanati (2001, 2004) offers a detailed differentiation of kinds of meaning in order to account for differences between semantic or context-free meaning, minimally pragmatically enriched meaning and what we intuitively take to be non-literal meaning. For the characterisation of different types or aspects of meaning, he assumes that the pre-theoretic intuitions of naive speakers should be preserved. Thus, the t-literal meaning1 of an expression is its literal meaning understood as its conventional meaning. The occurrence of an expression inherits the expression’s t-literal meaning, but in most cases the expression’s actual meaning also depends on contextual features. In the cases where the departure from the t-literal meaning of an expression is only minimal, Recanati uses the term m-literal meaning. Mliteral meaning is not a case of non-literal meaning as traditionally understood, rather, Recanati argues, it is the non-minimal departures from literal meaning (in Recanati’s terms m-nonliteralness) that can constitute non-literal meaning.2 Thus, Recanati’s differentiation between minimal and non-minimal departures from the tliteral meaning of an expression basically captures the difference between saturation processes and processes such as free enrichment. Thus, minimal departures from the t-literal meaning of an expression must be linguistically licensed, that is, the tliteral meaning of the expression under consideration must provide a variable which, when that expression is actually used, is provided a contextually determined value. Thus, in a sentence that involves indexical expressions, such as (181) below, the 1 t stands for (expression) type In this respect Recanati’s view differs significantly from Borg’s, since the latter considers all meaning aspects that go beyond the m-literal meaning as non-literal. She does not differentiate between non-minimal meanings that are perceived by the hearer as deviating from some underlying primary meaning and such that are not. 2 206 Varieties of Meaning, Context and the Semantics/Pragmatics Distinction semantics or t-literal meaning of the indexical can be said to license ‘. . . the search for a contextual value’ (Recanati 2004, p. 69). (181) He is thirsty. The situation is different in case of the process of free enrichment, which adds meaning aspects to the minimal proposition without there being any variable in the semantic form of the expression. Thus, cases of free enrichment constitute cases of non-minimal departures from t-literal meaning. However, non-minimal departure from the t-literal meaning of an expression, while being a necessary condition for non-literalness in the ordinary sense, is not a sufficient condition. Thus, Recanati makes a further distinction namely that between primary and secondary meaning. Secondary meaning (or in Recanati’s terms p-nonliteral meaning) is meaning that is derived from some more basic, primary meaning (e.g., p-literal meaning). The meaning of an utterance is non-literal in the ordinary sense, if it has a secondary character, that is, if it is derived from some prior, primary meaning (pliteral meaning). Examples of such non-literal meaning include conversational implicatures, indirect speech acts and, arguably cases of irony. Recanati defines primary meaning as not being inferentially derived from some earlier determined meaning. For example, an utterance of (181) can be used to make a statement about some particular person. For instance, in a situation where the speaker wants to make a statement about some person Paul, he can do so by uttering (181). The proposition expressed by that sentence in that utterance situation is ‘Paul is thirsty.’, but that is due to the specific contextual factors holding. It is a possible m-literal meaning of the sentence in (181), since the only departure from the sentence’s t-literal meaning is a saturation of the variable provided by the semantics of the indexical he. The speaker uttering (181) may, however, intend more than just to make a statement concerning Paul. He might make the utterance in order to convey that Paul should be offered a drink. The proposition that Paul should be offered a drink is a proposition in its own right and can therefore be said to be independent of the proposition that Paul is thirsty. However, its interpretation depends on a prior recovery of the m-literal meaning of (181). Thus, the m-literal meaning in this case is also a p-literal meaning as it forms the basis for the interpretation of the secondary or non-literal meaning that Paul should be offered a drink. Moreover, it is primary because there is no inference involved in moving from the t-literal to the m-literal meaning, or at least none the interpreters themselves are aware of. Rather, the m-literal meaning is directly determined. Recanati goes on to argue that there are meanings that involve non-minimal departures from t-literal meaning, yet are considered to be p-literal meanings nevertheless. For example we tend to interpret an utterance like (182) in a way where the marriage took place prior to the children being born. (182) They got married and had many children. However, this temporal ordering is not part of the t-literal meaning of (182). Neither is it part of its m-literal meaning, since m-literal meaning is defined as departures from t-literal meaning that are licensed by the conventions of language. Thus, although (182) is a case of a non-minimal departure from t-literal meaning, it is Towards an Alternative Characterisation of (Non-)Literal Meaning 207 not a case of non-literal meaning in the ordinary sense. It is still considered to have p-literal meaning, since its meaning is not derived from some earlier determined meaning. Recanati makes similar points regarding meaning aspects that are the result of free enrichment. As we saw above, such meaning aspects are considered to be nonminimal departures from the t-literal meaning of the expression concerned, however, they do not constitute non-literal meaning in the traditional sense, since they do not have a secondary character, that is, they are not determined on the basis of some underlying meaning that has been determined first. Thus, looking once more at example (78), the idea is that what is contributed by the process of free enrichment (or expansion in Bach’s terminology) is not non-literal in the traditional understanding, as it does not have a secondary nature. That is, ‘from this cut’ is not the result of a reanalysis of some previously determined primary meaning. (78) a. You’re not going to die. b. You’re not going to die [from this cut] Secondariness, then, is a sufficient criterion for some meaning to be viewed as nonliteral in the traditional sense. However, Recanati argues that it is not a necessary condition. What is characteristic for non-literal meaning as traditionally understood is that it is perceived by the language user as being special or deviant. This is due to the fact that non-literal meaning is an aspect of language use, thus, it comes about during the interpretation of some utterance and therefore is available to the interpreter. Recanati calls this the transparency condition and this is a defining property of any case of non-literal meaning. ‘A use of words counts as non-literal in the ordinary sense only if there is something special about that use that is, or can be, perceived by the language users themselves.’ (Recanati 2004, p. 75, emphasis in the original) Thus, there are other types of meaning that qualify as non-literal although they lack the property of secondariness. For example, in the case of metonymy or metaphor, their non-literal nature differs from the kind of secondariness identified by Recanati for conversational implicature or indirect speech acts. As mentioned above, conversational implicature are based on the utterance meaning and added to it. In contrast, in instances of metaphor or metonymy, the non-literal meaning is not added to an earlier derived utterance meaning, but rather, its computation forms part of the derivation of the utterance meaning itself. Thus, an utterance such as (18a) is non-literal although it does not have secondary meaning in the sense defined above. (18a) The ham sandwich in the corner wants some more coffee. Rather, the non-literal meaning comes about due to the deviance of the reading of ham sandwich employed in this particular utterance situation, from the readings that are conventionally associated with the expression ham sandwich. However, this non-literal reading of ham sandwich is not secondary in that it is not based on some 208 Varieties of Meaning, Context and the Semantics/Pragmatics Distinction earlier computed and literal utterance meaning. Thus, it is rather more similar to other cases of what Recanati calls meaning adjustment, e.g., free enrichment. The difference between metonymy or metaphor and meaning aspects due to free enrichment is that the latter elaborates the meaning conventionally associated with some expression (thus, Recanati’s term sense elaboration), whereas the former loosens it (i.e., sense extension). Recanati uses the notion of schema to differentiate between the two types of processes. Thus, in free enrichment the expression that is enriched is used in its ordinary meaning, that is, the schema provided by the expression fits the situation at hand and the process of free enrichment only adds more specific details. However, in the case of metaphor or metonymy, the schema provided by the expression used does not conventionally fit the situation at hand. It therefore has to be adjusted in order to fit. That is, those meaning aspects which conflict with the aspects of the situation at hand, are filtered out. However, the meaning aspects that have thus been filtered out still stay activated to some extent. It is the number of meaning aspects filtered out and the extent to which they remain activated which is responsible for whether a hearer is consciously aware of a discrepancy between conventional word meaning and the reading an expression is used in in a particular situation. To summarise, in his (2004) book, Recanati distinguishes between different kinds of meaning. Not only is there a difference between t-literal and m-literal meaning, but also a difference between p-literal and p-non-literal meaning. Moreover, not all p-literal meanings need be m-literal meanings. That is, there are some primary, underived meanings that, at the same time, constitute non-minimal departures from the t-literal meaning of the expression concerned. Thus, Recanati’s notions of mliteral meaning and p-literal meaning can be used to describe cases of what Bach calls sentence non-literality more specifically; cases, that is, in which there is a non-minimal departure from t-literal meaning without that departure constituting genuine non-literal meaning. From the examples Recanati gives in explaining this difference, it seems that the pairs of terms apply to different levels of meaning. That is, he identifies mliteral meaning as Grice’s level of what is said, thus, a type of meaning holding at the level of utterance meaning. In contrast, he takes conversational implicature and indirect speech acts to be good examples of what he calls secondary meaning (or pnon-literal). As indicated above, these phenomena are usually taken to occur at the level of communicative sense. However, Recanati also identifies cases of non-minimal departures from t-literal meaning, which, on the one hand, differ from Grice’s notion of what is said in that they include contextually provided meaning aspects that are not linguistically licensed, and on the other hand are not secondary in that they are not based on some earlier determined meaning. This type of meaning he calls pliteral. Interestingly, this type of meaning involves different meaning aspects, namely such as result from processes such as free enrichment as well as cases of figurative uses of language as metonymy and metaphor. Thus, Recanati differentiates between different types of non-literal meaning, where some are secondary in nature but others are not. There are a number of problems with Recanati’s characterisation of the various types of (non-)literal meaning. First, he assumes that t-literal meaning – that is, Towards an Alternative Characterisation of (Non-)Literal Meaning 209 t-literal H HH H m-literal (saturation processes) HH H H m-non-literal H HH H HH p-literal H H HH H HH H sense elaboration (enrichment) H H p-non-literal (indirect speech acts, CIs, . . .) sense extension H HH HH Figure 5.1: Non-literal meaning in Recanati’s classification below threshold above threshold (figurative uses) the linguistically coded meaning – is conventional. ‘. . . start with a sense of the phrase ‘literal meaning’ which is reasonably clear and raises no particular problem. In that sense, the literal meaning of a linguistic expression is its conventional meaning: the meaning it has in virtue of the conventions which are constitutive of the language.’ (Recanati 2004, p. 68) As we saw in the preceding chapters, however, it is not at all clear that a characterisation of semantic/linguistic meaning in terms of conventionality is unproblematic, especially since it implies that conventionality is an all-or-nothing property, which, as I argued above, it is not. Second, it is somewhat irritating that Recanati takes conversational implicatures and indirect speech acts to be good examples of what is usually understood as nonliteral meaning. As Stern (2006, p. 248, footnote 2) comments: ‘[o]ne might balk at classifying such inferred meanings as either literal or non-literal’. Rather, it is phenomena like metaphor or metonymy that are considered prototypical examples of non-literal meaning. What makes conversational implicatures and indirect speech acts good examples of non-literal meaning for Recanati seems to be the fact that in his classification these two meanings do not only fulfil the transparency condition, but at the same time have a secondary character. The latter cannot be said of metaphor and metonymy, whose non-literalness Recanati’s system can only capture by applying to the transparency condition. However, recall that, traditionally, nonliteral meaning is understood as being the result of a reinterpretation of an utterance. Thus, the resulting meaning, in a sense, replaces the original meaning.3 Although we saw that the traditional view is problematic, the intuition that in cases of non-literal meaning the basic meaning is in some sense replaced by the non-literal meaning, remains. In case of conversational implicatures, however, the situation seems to be different. Here, the meaning contributed by the CIs is added to the basic meaning 3 Even though, as we saw, at least some aspects of the underlying meaning remain active and thus play a role during the interpretation of a non-literal meaning. 210 Varieties of Meaning, Context and the Semantics/Pragmatics Distinction from which they are derived. That is, the CIs are not part of the utterance meaning of the utterance made, but rather they are part of the communicative sense or what is meant by the speaker of that utterance. Thus, in the cases in which CIs are drawn with respect to a particular utterance, the meaning of the utterance itself is not changed or becomes non-literal as seems to be suggested by Recanati.4 Note further that the transparency condition is problematic itself. It imputes to speakers the ability to reliably differentiate between cases of literal meaning and cases of non-literal meaning as classified by Recanati. However, at least for the phenomenon of conversational implicature we already saw that this is not the case. Thus, although CIs are originally assumed to be derived on the basis of some priorily determined utterance meaning, speakers are not able to reliably differentiate between the ‘said’ and the ‘implied’. Moreover, Recanati’s characterisation of how the nonliteralness of metonymic or metaphoric readings for some expression comes about is rather vague. Thus, especially in the case of metonymy, one might wonder what meaning aspects of the literal meaning of the respective expression are filtered out and thus lead to it to be understood as metonymic. This characterisation works better with metaphors but even in that case, especially where the metaphor vehicle and the metaphor target are of different sorts, deriving a metaphoric interpretation does not seem to simply be a case of ‘feature-dropping’. Having said all that, Recanati’s characterisation can also not capture the fact that there are cases of non-literal meaning which seem to be as easily/fast understood as literal meaning is. That is, arguably, in the case of what Giora calls familiar metaphors, there is no strenuous process of metaphoric meaning-generation, with non-fitting meaning aspects being filtered out but staying activated to a sufficient extent. Nevertheless, the respective meanings are or at least can be perceived as non-literal. With respect to the process of free enrichment, the question arises why it is not characterised as secondary, although it adds meaning aspects to minimal propositions and why its results are not perceived as deviating, although they are added to minimal propositions. Recanati’s argument is that free enrichment is a primary pragmatic process and as such not consciously available to hearers. Thus, they cannot know that it is based on some priorily determined meaning. However, as we saw, assumptions relying on hearers’ awareness of some process/meaning aspect is problematic. Note also that Recanati’s differentiation between free enrichment on the one hand and metaphor on the other in terms of sense elaboration and sense extension, respectively is seen as problematic by proponents of RT. Thus, Carston (1997) has argued for treating narrowing (Recanati’s sense elaboration) and loosening (Recanati’s sense extension) as basically two opposite directions in which one and the same process of ad-hoc concept formation might turn, especially since there are cases of concept adjustment in which both narrowing and loosening are involved. Thus, in RT free enrichment and the process leading to a metaphoric interpretation are not 4 ‘The interpretation of both conversational implicatures and indirect speech acts involves an inference from the utterance’s primary meaning to its derived meaning’ (Recanati 2001, p. 266). Towards an Alternative Characterisation of (Non-)Literal Meaning 211 viewed as distinct.5 5.1.2 Literal Meaning as ‘Minimal Meaning’ Ariel (2002) proposes to replace the traditional notion of literal meaning by three distinct types of minimal meaning. Thus, she differentiates between linguistic meaning, salient meaning and privileged interactional meaning. The reasons for abandoning the traditional notion of literal meaning are similar to the ones put forward here, most importantly the fact that the traditional characterisation of literal meaning is inconsistent. Her differentiation of meanings is based on the motivations behind the traditional concept of literal meaning, which she takes to be linguistic, psycholinguistic and interactional in nature. Thus, Ariel suggests a three-fold distinction of different types of what she terms minimal meaning on the basis of the kinds of functions that these meanings serve. Ariel claims that what we take to be the semantic, that is, linguistic meaning of an expression may not coincide with the reading that is most salient from a psychological point of view. Similarly, the interpretation of some utterance that we take a speaker to be bound by may be different both from the result of the composition of the linguistic meanings of the expressions involved as well as from the readings that are most rapidly activated during the interpretation of that utterance. Let us look at these different types of literal meaning, or rather minimal meaning in more detail. The first type of minimal meaning that Ariel distinguishes is the linguistic meaning of expressions, that is, their encoded meaning. Linguistic meaning is not wholly truth-conditional, that is, it includes meaning aspects that are not truthconditionally relevant and the truth-conditional aspects of meaning that it does provide may not exhaust the entire set of truth-conditions necessary to determine the truth or falsity of the meaning of some utterance of the linguistic expressions. Moreover, although linguistic meaning generally is characterised as being compositional, non-compositional meanings (such as for idioms) may also grammaticise and thus become linguistic meaning. From how Ariel characterises linguistic meaning, it seems that she does not take it to be characterised by underdeterminacy. On the contrary, like Recanati, she seems to assume that linguistic meanings are full-fledged readings for expressions. Thus, she argues that although linguistic meanings rarely figure in actual conversation unenriched, interpreters are able to opt for this ‘bare’ linguistic meaning. Ariel gives an example to illustrate this possibility. (183) a. S: Tire, ani hitxalti ita lifney xamesh shanim ‘Look, I started with/made a pass at her five years ago’ b. M: Ata hitxalta ita o she . . . ‘You made a pass at her or . . .’ c. S: Lo . . . ‘No . . .’ 5 See, however, Burton-Roberts (2007b)’ critique of the symmetric treatment of what do seem to be two distinct processes. Also, see below for a general critique of the way ad-hoc concept formation is characterised by RT. 212 Varieties of Meaning, Context and the Semantics/Pragmatics Distinction In this example, the construction hitxalti ita is used, which is ambiguous as it may mean ‘started working with her’ or ‘made a pass at her’. Ariel claims that the context in which this little conversation took place, made it quite clear that S intended the former reading. M, however, picks up on the second possible reading. Ariel argues that M’s interpretation of S’s utterance cannot be dismissed in the way a completely irrelevant meaning might have been. She assumes that this is the case because M’s interpretation of S’s utterance is linguistically justified, since he picks up on a ‘bare’ meaning of the construction hitxalti ita. This example suggests two things. First, that Ariel takes linguistic meanings to be full-fledged readings. Moreover, she explicitly states that the various meanings of homonymous6 and polysemous expressions are ‘. . . no doubt . . .’ (Ariel 2002, p. 394) instances of linguistic meaning. As should have become obvious from what has been said sofar, the assumption that the meanings of polysemous or homonymous expressions generally are linguistically coded is not at all uncontroversial. Second, what identifies such ‘bare’ readings as instances of linguistic meaning is the fact that in the situations they are opted for, they are not contextually appropriate. In other words, they arise or are chosen as the intended meaning by interpreters, although they do not fit the respective context. Thus, Ariel assumes they cannot be the result of interpretation ‘in context’, as it were. However, I do not find this evidence convincing, because what Ariel calls ‘bare’ linguistic meaning does not have to be taken to be a linguistic meaning at all. It might belong to the second level of minimal meaning identified by Ariel, namely salient meaning, which notion she directly borrows from Giora. Thus, as mentioned before, this type of meaning can be seen as a psycholinguistically motivated minimal meaning in that it is the meaning that is activated first, automatically and obligatorily. In the sense that this meaning is the easiest to access – or the meaning the activation of which is least costly – it may be regarded as a kind of literal meaning.7 Ariel argues that not all salient meanings are necessarily identical to linguistic meanings and vice versa. This is due to the fact that the criterion for saliency of meaning is the speed with which we process the respective meaning, rather than whether the meaning under consideration is conventional. However, recall Giora’s definition of salient meanings as being a function of several aspects, one of them actually being conventionality. Moreover, Ariel does not give a detailed explanation of what she takes the notion of conventionality to capture. She does mention, in a footnote, that her view of conventionality differs from Giora’s. Thus, whereas Giora takes conventionality of a meaning to be one property that influences the status of that meaning as to whether it is (less-)salient or not and thus (co-)determines whether that meaning will be lexically coded, Ariel differentiates between conventional uses of an expression with a particular reading and the lexicalised meanings of an expression. Conventional uses determine the salience of a meaning, but not its linguistic status. A particular reading is only linguistically conventionalised if it is lexicalised. But at this point the question arises 6 Note that Ariel (2002) uses the term ambiguous to refer to homonymous expressions. However, since the notion of ambiguity usually is understood to encompass both homonymy and polysemy, to avoid confusion, I will consistently use the term homonymous where Ariel (2002) has ambiguous. 7 However, Giora emphasises that the notion of salient meaning cuts across the traditional literal/non-literal distinction. Towards an Alternative Characterisation of (Non-)Literal Meaning 213 which aspects lead to the lexicalisation of a particular reading of some expression. Note that Ariel’s view on conventionality is similar to Busse’s, (discussed in chapter 2) who also differentiates between conventions of language and conventions of language use. Recall, however, that Busse takes the process of conventionalisation to include the process of lexicalisation proper, that is, the process that finally changes the status of some reading from being pragmatically derived to being semantically coded. As was suggested above, conventionality seems to be closely connected to the frequency of use of an expression with a particular reading. Thus, arguably, it is increased frequency of use that leads to a shortening of the inference chain ending in a pragmatic inference. The point where the inference chain becomes so obscured that we are no longer able to consciously reconstruct it, can be taken to be the point at which the reading under consideration lexicalises. Ignoring for the moment the question whether one wants to assume that semantic meanings are full-fledged readings or not, the crucial point about these assumptions concerning the lexicalisation process is that it does rely on the conventionality of a particular reading, which in turn can be taken to be determined by the frequency of use of the reading under consideration. If that is the case, then I see no reason why it should be necessary (in Ariel’s sense) to differentiate a level of linguistic meaning from a level of salient meaning. To come back to the example mentioned above, the reading ‘made a pass at her’ of the construction hitxalti ita of which Ariel claims that it cannot be dismissed in the way a completely irrelevant meaning might have been due to the fact that it is linguistically justified: this conclusion is not necessary as the respective reading may be taken to be a salient meaning of the construction. Thus, Ariel mentions studies the results of which show that during the interpretation of an ambiguous utterance, all salient meanings are initially activated, but most of them are supressed rather quickly, as soon as it becomes clear that they are not intended. However, the experimental situation is, I think, an idealised one, in that it is very likely that the most prominent of the intentions of the subjects is to comply with what they are asked to do by the experimenter. In other words, in such an experimental environment there is no other, personal reason to be less than fully co-operative, or, to put it differently, there is no particular reason why the subject should have other personal intentions that determine his interpretation of a given utterance. The idea is that if, as Ariel suggests, M in this scenario is not being fully cooperative, that is, if his relation to S, or to the topic discussed, is such that, say, he enjoys teasing S and in this situation does so by willfully misunderstanding him, then that attitude towards S and the intention that grows out of it will influence the way he interprets S’s utterance. The interpretation process will not necessarily unfold in the way that the plausibly not intended reading of the construction hitxalti ita is heavily suppressed. Recall Schulz von Thun’s characterisation of a message having more than just two sides, where this characterisation takes into account that a hearer’s interpretation of a speaker’s utterance is influenced by how the former assesses the relation holding between himself and the speaker. Moreover, speakers and hearers may have certain attitudes towards one another, which may be affected by the interactional context. So, in the example above, it seems that M has a certain attitude toward S, which he does not want to express directly, but which is expressed 214 Varieties of Meaning, Context and the Semantics/Pragmatics Distinction by his apparently willfully misunderstanding S. Or it might be that he simply wants to humorously tease S and does so by choosing to willfully misunderstand him. The point is, that if M actually has any one of the intentions mentioned, they will influence the way he interprets S’s utterance and thus may lead to the interpretation process resulting in more than one interpretation. Another problematic aspect about Ariel’s differentiation of linguistic meaning from salient meaning concerns the role the two are assumed to play in the interpretation process. Looking once more at the example above, Ariel states that the interactionally implausible interpretation of S’s utterance by M is linguistically justified, because the basis for it is a linguistic meaning of the construction hitxalti ita. Does that mean that in interpretation linguistic meanings and salient meanings are accessed independently of one another? If the main reason for M’s arriving at that particular interpretation of S’s utterance is the fact that the reading is linguistically justified rather than a salient meaning of the construction in question, then it should be possible for that reading to be chosen as the interpretation of the construction, although it is not a salient meaning. Thus, linguistic meaning and salient meanings have to be taken to play different roles in the interpretation process. I find such an assumption highly implausible. First, the interpretation process is a cognitive task that can be measured by appropriate methods. As has been shown in numerous experiments, manipulating the input to that process may affect it and the results it returns. I do not see how linguistic meaning (in Ariel’s sense) can figure in such a process without being picked up by the instruments used to measure the unfolding of that process. Second, it is not clear to me on which basis Ariel distinguishes between linguistic meaning and (less-)salient meaning. She claims that the former is a significant level of meaning as it ‘characterises the native speaker’s competence in her language.’ (Ariel 2002, p. 362). However, I am not quite sure what it means for linguistic meaning to characterise a speaker’s competence in his language. Moreover, such a characterisation of the role of linguistic meaning only makes sense if one assumes that linguistic meaning is constituted by full-fledged readings. However, as we saw, there are a number of arguments for not doing so. Thus, if linguistic meaning is taken to be characterised by underdeterminacy, then it cannot easily be taken as being reflected in a native speaker’s competence in his language. Rather, when talking about a speaker’s semantic competence, what is actually meant is her ability to use a particular expression appropriately in situations in which that expression is, more or less, standardly or conventionally used. Finally, let us turn to the third type of minimal meaning Ariel suggests: the privileged interactional meaning. This type of meaning is minimal in the sense that it may be taken to capture that content of an utterance, which the speaker is taken to be bound to. It thus constitutes the most basic level of communicated meaning. Ariel concedes, however, that there is no unique meaning representation that may be taken to invariably function as the privileged interactional meaning. Rather, what is taken as the privileged interactional meaning of some speech act may actually differ from interlocutor to interlocutor. Recall that this was also suggested by the experiments in which subjects were asked to choose among a number of alternative paraphrases the one they thought best captured what a speaker said with his utterance. Depending on the nature of the testing material and situation, Towards an Alternative Characterisation of (Non-)Literal Meaning 215 subjects did not always choose the paraphrase that captured the theoretical notion of what is said. Rather, they sometimes chose paraphrases which included what theoretically would be identified as clear cases of implicature (e.g. Nicolle and Clark 1999, Bezuidenhout and Cutting 2002). Ariel argues that this type of meaning is both distinct from linguistic meaning and salient meaning. In contrast to the two latter types of meaning, which are often subconscious, privileged interactional meaning is fully conscious. Moreover, it clearly differs from linguistic meaning in that it is a cognitive hybrid, including both coded and inferred meaning aspects, whereas linguistic meaning is clearly coded. As evidence that salient meaning differs from privileged interactional meaning Ariel suggests the differences in processing that have been found for the interpretation of homonymous and polysemous expressions. Thus, for both types of expression, early on during their interpretation, all their meanings which are equally salient are accessed. However, for homonymous expressions, their unintended meanings are then rather quickly suppressed, whereas the unintended readings of a polysemous expression remain activated for much longer. In a particular speech situation, though, the interpreter of a polysemous or homonymous expression, having determined the reading of that expression in that particular situation, may not consciously be aware of other possible readings, although they were unconsciously accessed during the interpretation process. Moreover, the interpreter is not consciously aware of the fact that the unfolding of the interpretation process differs for homonymous and polysemous expressions. Thus, privileged interactional meaning is supposed to capture a level of meaning similar to explicature. That is, it consists both of coded and inferred meaning aspects and it is a meaning level which can be evaluated regarding its truth or falsity. As already stated, Ariel admits that there is not necessarily a way of uniquely determining this particular level of meaning. Nonetheless, it is characterised as that minimal level of meaning of an utterance the speaker is taken to be committed to. This view of privileged interactional meaning as a type of literal meaning seems to be based on the traditional idea according to which what is said by an utterance is both fully propositional as well as literal in that it only consists of the lexical meanings of the expressions that are part of the utterance and of meaning aspects contributed by the traditionally assumed saturation processes. As such, it is traditionally assumed to be that level of meaning to the assertion of which the speaker of the respective utterance is taken to be bound to. However, as we saw especially in chapter 3, actually the level of meaning a speaker is taken to be bound to, is richer than Grice’s level of what is said and, crucially, it may involve meaning aspects that traditionally are classified as non-literal (such as metonymy or metaphor). 5.1.3 Nature of the Processes Determining (Non)-Literal Meaning Before turning to my own characterisation of literal meaning and non-literal meaning, I would like to summarise and discuss the various processes identified in chapter 3 and 4 involved in determining the meaning of an utterance. The point I want to make is that there seem to be no processes specifically aimed at determining what 216 Varieties of Meaning, Context and the Semantics/Pragmatics Distinction might be called the literal meaning of an expression. Thus, recall the processes discussed in connection with the level of utterance meaning in chapter 3. We saw that the processes of disambiguation, fixing of indexicals and reference resolution lead to what Grice called what is said. However, we also saw that this traditional level of meaning may not actually be of fully propositional form. Since in chapter 2 I suggested that literal meaning is to be found at the level of fully propositional meaning, the saturation and disambiguation processes leading to Grice’s traditional what is said cannot determine this type of meaning. However, recall that Bierwisch proposes the two processes conceptual shift and conceptual differentiation which, arguably, do result in literal meaning. These two processes supply meaning aspects that fix the particular reading the expression they apply to has in a particular (narrow) utterance context. Thus, they are saturation processes in that they provide the values for variables in the semantic forms of the expressions concerned. Note that, according to Bierwisch, in both cases what the processes contribute are primary or non-derived meaning aspects. That being so, conceptual shift and conceptual differentiation can be seen as processes that result in literal meaning aspects being integrated into the propositional content expressed by an utterance. However, note that similarly to Grice’s view, such a model of (non-)literal meaning interpretation would again amount to a literal first or serial model of interpretation. In fact, as the results of experiments on the interpretation of novel cases of non-literal meaning suggest, such a model may be appropriate when it comes to modelling how unfamiliar cases of non-literal meaning are interpreted. However, for cases of familiar non-literal meaning, it makes the wrong predictions. An alternative is to assume that saturation processes may already integrate meaning aspects that are intuitively non-literal, but which have a high frequency and are thus familiar and might be taken to be conventionalised to the degree that they have become part of the conceptual family from which the processes of conceptual shift or conceptual differentiation choose a member. Thus, some types of metonymic shift – such as e.g., the producer-for-product metonymy – seem to be highly productive (at least for English), allowing for unproblematic extension to new cases (cp. Frisson and Pickering 2007). That is, identifying some person as an author seems to automatically allow the use of that person’s name to refer to books produced by that person. That being so, one could assume that such highly productive processes apply virtually automatically, thus, leading to no perceptible reading time or semantic integration effects. Another possibility is that for highly frequent metonymies, the metonymic reading in fact has become part of the concept family on which the process of conceptual shift operates. For example, as soon as some person is characterised as an author, the possibility of using that person’s name for his books is made available by the general relation holding between authors and their creations. Similarly, in the case of institution-for-person metonymy, the metonymic reading seems fairly conventional, due to the general knowledge that institutions are associated with people. The fact that such readings may still be perceived as non-literal might be due to the particular semantic relations taken to hold between those readings and the socalled literal ones or it might simply be due to the fact that processes exist that create these kinds of meaning relations, allowing for the use of a particular expression in a Towards an Alternative Characterisation of (Non-)Literal Meaning 217 particular way in the first place. Thus, the process of metonymic shift may apply to a concept that is part of the set of primary readings of a particular expression and result in a metonymic interpretation for that expression in that particular situation. If the relevant expression is frequently used in that metonymic reading, the result of the process might be integrated into the concept family of which the primary concept the expression is used to express is part. In that case it is conventionalised and the interpretation process no longer has to infer it. Note, however, that frequent use of an expression with a particular metonymic reading is not a necessary condition for it to be integrated into the set of primary readings of the expression. Thus, recall Frisson and Pickering (2007)’s experiment in which subjects showed no difficulty interpreting a ‘producer-for-product’-metonymic use of an expression they had only encountered once before and not in a metonymic reading. That is, the ‘producer-for-product’-metonymy seems so regular and productive that it is enough for speaker-hearers to identify some person as an author to be able to use that person’s name to refer to his works. Thus, the particular metonymic reading is added to the set of primary readings for the expression automatically, as it were. That is the point at which a metonymic reading is accessed as quickly as a primary one. However, due to the existence of the metonymic process and the fact that people can consciously reconstruct how the process works, they are also consciously able to realise that some reading is metonymic, even if that reading is highly familiar and in that sense conventional. This idea suggests that something is left behind by the process of metonymic shift that allows the system to recognise that a particular metonymic reading for a particular expression has been generated before because it seems that this is needed to account for how such a reading can be conventionalised in the first place. If it is only created on the fly, as it where, and no trace of it is left behind, it is unclear how the system would be able to recognise that it is used often enough for it to be more efficient to get permanently stored.8 Thus, the set of potential primary meanings for an expression from which such processes as conceptual shift and conceptual differentiation choose, includes not only non-derived, basic and in that sense literal meanings, but also such as are clearly derived, non-basic and in that sense non-literal. Thus, membership in the set of primary meanings is not determined on the basis of ‘literalness’, rather, it is based on salience of meaning. That is why the set of potential readings for an expression that a pointer points to should rather be called the set of salient readings for an expression. In fact, as suggested already in chapter 2, actually what a pointer points to is not simply the set of salient readings for an expression – where that set, as we saw, has to be characterised as being dynamic – rather, it is a conceptual region, or what RT calls the encyclopaedic entry of a concept, which may be taken to be organised by salience and thus subsumes the set of salient readings. That is, the encyclopaedic entry includes ready-made readings for an expression but also material from which 8 This idea might be modelled by assuming that the particular reading determined by the interpretation process and due to a particular use of an expression does leave a trace in memory on each occasion it is determined by the interpretation process. The more such traces exist, the easier it will be to activate this particular reading. (Cp. Hintzman 1986). 218 Varieties of Meaning, Context and the Semantics/Pragmatics Distinction a contextually fitting reading may be ‘build’. In contrast to saturation processes in general, free enrichment (or expansion) is characterised as a process that operates on the fully propositional semantic form of an utterance adding meaning aspects that can be taken to have been intended to be expressed by the speaker and that are necessary in order for the resulting, enriched proposition to form the basis for further pragmatic inferences. Thus, in contrast to conceptual shift, conceptual differentiation, metonymical shift or ad-hoc concept-formation, free enrichment does not operate on meaning below the level of full propositionality. Rather, according to Bach (1994a), what is characteristic for cases where free enrichment applies is that each of the expressions involved in the respective utterance is used literally (or with primary, non-deviating meaning), however, the proposition taken to be expressed by the speaker of the utterance is ‘. . . conceptually more elaborate . . .’ (Bach 1994a, p. 135) then the unenriched proposition. That is why Bach (1994a, p. 135) calls this phenomenon sentencenonliterality: ‘. . . a sentence is used nonliterally without any of its constituents being so used’. This is in contrast to metonymy and metaphor, where the non-literality is usually associated with simple expressions. Thus, recall the examples Dölling (2001) gave for metonymic shift and metaphorical interpretation, respectively of an expression. As we saw, these processes integrate conceptual material into the inflected SF of the respective expression, using the structure provided by the met-operator. What is important for the resulting interpretation to be coherent is that the SF-parameter within the SEM of the expression concerned be fixed when the process of metonymic shift or that of metaphorical interpretation applies.9 For example, Dölling (2001) gives the respective parameter-fixed SFs (i.e. 121) for the individual metonymic uses of Lamm given in (120), where the SF-parameter within the basic SEM of Lamm has been fixed by . (120) a. Hans möchte heute Lamm essen. ‘Hans wants to eat lamb today.’ b. Maria trägt seit gestern Lamm. ‘Since yesterday, Maria wears lamb.’ c. Anna weigert sich, Lamm zu essen oder zu tragen. ‘Anna refuses to eat or wear lamb.’ (121) a. λx. FLEISCH(x) & ∃y [MAT(y)(x) & (y) (lamm)] b. λx. FELL(x) & ∃y [MAT(y)(x) & (y) (lamm)] c. λx. FELL∨FLEISCH(x) & ∃y [MAT(y)(x) & (y) (lamm)] Similarly for potential metaphoric uses of Lamm as in (123) below. Here the parameter in the SEM of Lamm has been fixed to =. (123) Annas Mann ist ein Lamm. ‘Anna’s husband is a lamb.’ 9 Or at least that they get fixed in the course of the application of such processes. Towards an Alternative Characterisation of (Non-)Literal Meaning 219 (124) λx. PERSON(x) & ∃y [INST(y)(x) & ART(y) & ∃z [ÄHNLM (z)(y) & (z) = (lamm)]] Another process that operates on the semantic form of simple expressions is adhoc concept formation. This process can be taken to determine non-literal meanings for the expression it operates on, since it is characterised as contextually adjusting the encoded meaning of an expression. However, if one assumes that semantic meaning actually is highly underspecified, it does not seem useful to characterise adhoc concept formation as a process that applies to the encoded meaning and adjusts it. That is, the encoded meaning under the assumption of underspecification is not even a full meaning; it does not constitute a full concept. The intuition underlying the assumption of a process like ad-hoc concept formation is, however, that it does apply to a full reading or concept and returns a concept that is based on that original concept. Recall also that in RT it is assumed that ad-hoc concept formation is the underlying process for meaning aspects differing in their degree of perceived nonliteralness (e.g., loose uses, hyperbole, metaphor). One possibility of capturing this intuition is to assume that ad-hoc concept formation operates in those cases, where, based on the consideration of the context, the process that determines the value of the SEM-parameter in that particular context, does not (easily) find a reading that is sufficiently general/specific enough to fit in that context. Thus, the reading ad-hoc concept formation operates on is that member of the value set of the SEM-pointer which is best compatible with the context at hand, but not general/specific enough. Recall example (66) from chapter 3, repeated below. (66) Kato (of O. J. Simpson, at his trial): He was upset, but he wasn’t upset. [He was upset’ but he wasn’t upset”] The interpretation of the two occurrences of upset in this example can be adequately described employing the idea of ad-hoc concept formation. Thus, the first occurrence of the expression upset is assigned one of the expression’s salient readings, namely the one that fits best the contextual circumstances. The second occurrence of upset, however, seems to be used by the speaker to convey a very specific concept, given the particular circumstances of utterance. To repeat, a rough paraphrase of what Kato wanted to express might be ‘O.J. Simpson was sad and angry to a certain degree, but he was not sad and angry to a degree that led him to killing his wife.’ Thus, it seems rather unlikely that the speaker entertained this particular concept prior to the particular situation he found himself in and in which he made the utterance in (66). It is even more unlikely that this concept is stored as part of the salient readings for upset for either speaker or hearer. Thus, the concept is created online, during the interpretation of the respective utterance and by specifying the concept taken to be expressed by the first occurrence of upset in the utterance. While the idea of ad-hoc concept formation works well for such examples as the one given above, it is actually questionable whether this process also is involved in what is called loose use in RT, examples of which are repeated below. (103) a. France is hexagonal. b. Holland is flat. 220 Varieties of Meaning, Context and the Semantics/Pragmatics Distinction c. I must run to the bank before it closes. In fact, at least for the examples given, it is questionable whether a specific process is necessary to explain this type of phenomenon at all (cf. Burton-Roberts 2007a). Recall that, in the above example, RT claims, the expressions hexagonal, flat and run are used to express concepts that are less strict or specific than the ones they are taken to encode. The first thing to note is that if what expressions encode actually are not full concepts, then what is encoded by hexagonal, flat and run respectively, cannot said to be a strict or specific reading, as it is no full reading at all. Second, and as noted by Burton-Roberts (2007a) for the assumed encoded meaning for these expressions, assuming that the strict or specific concepts are the only concepts the pointer in the SEM of the respective expressions points to, seems a rather arbitrary stipulation. On the contrary, the fact that the uses of the relevant expressions in the examples above only seem ‘special’ on second thoughts and only under the assumption that their salient meanings are strict, makes it even more questionable whether this assumption is in fact appropriate. In this connection, recall also Ruhl (1989)’s cautionary note not to attribute ‘non-literalness’ to a particular reading for an expression only on the basis that this reading deviates from some other reading that expression might be used to express. Note also that, at least for the expression flat, the assumed ad-hoc concept in fact does not actually seem to be so ad-hoc. That is, using flat to describe the general geographical circumstances of a country or some region seems to be a rather conventional means of making such a description. In fact, for the corresponding German expression flach, the respective reading figures in the compound Flachland (i.e., lowland, plain), indicating that the relevant concept expressed by flach is not at all ad-hoc. Thus, I would tentatively suggest that the concepts expressed by so-called loose uses of expressions in fact are nothing more then possible salient meanings, and that in order for them to be expressed on a particular occasion of use of the respective expression, nothing else happens than that the interpretation process chooses that reading from the set of salient meanings the expression might be used to express. This suggestion is only tentative, as the sole intuition that such uses are not felt to involve some effortful process of concept-formation might be misleading. Thus, the idea that cases of what is called loose use actually involve nothing more than salient readings of the respective expression needs to be empirically tested. Having said that, one might argue that even though such loose readings might be members of the set of salient meanings today, diachronically speaking, they might still be the result of some loosening process – or as argued by RT, of ad-hoc concept formation – which applied to the strict meaning and created the loose reading from that. Thus, recall that the set of salient meanings as I have characterised it also includes non-basic, derived readings. As mentioned already, the problem is that there does not seem to be a way of objectively ascertaining whether a particular meaning is literal or not. Be that as it may, the point is that for the process of ad-hoc concept formation, as it is characterised by RT, one would assume that its operation in a particular interpretation situation would involve greater cognitive effort than simply choosing a ready-made reading from a given set. My assumption, then, is that in the interpretation of the examples of loose use that RT provide, Towards an Alternative Characterisation of (Non-)Literal Meaning 221 this process does not operate. Rather, I would suggest that the respective readings involved in so-called loose uses stand in the same relation to the so-called strict readings as the salient readings of a polysemous expression generally do. Recall that in RT the process of ad-hoc concept formation is generally assumed to be at play for all kinds of more or less non-literal meaning aspects, amongst them, of course, metaphor. However, even if ad-hoc concept formation is taken to involve loosening of the denotation of an existing concept, it is questionable whether this appropriately characterises what is going on in metaphor interpretation. That is, while the idea, as we saw, works for cases such as in (106), it is questionable whether ad-hoc concept formation can be applied to cases, where the salient concept of the metaphor vehicle could not even in principle be attributed of the concept of the metaphor target, as is the case in (184).10 (106) John is a soldier. (184) Peter is a bulldozer. That is, the interpretation of bulldozer as denoting a category to which Peter is said to belong, does not seem to involve simply a cancellation of particular features of the original concept bulldozer is taken to express ‘literally’.11 Thus, Dölling (2000)’s treatment of metaphor as involving a similarity function ÄHNL that relies on a set of features M with respect to which two concepts are considered to be similar better captures what seems to be going on in metaphor, although how those features are determined remains an open question. To summarise, the nature of the processes contributing to utterance meaning that we looked at suggests that it is differences in the information used by those processes that leads to the resulting utterance meaning being perceived as either literal or non-literal. That is, there does not seem to be a distinctive set of processes resulting only in literal meaning. Taking into account the fact that potential readings for an expression differ in their grade of salience, there are processes that assign values to parameters in the semantic form of an expression – so-called saturation processes – and the readings that might potentially be that value are characterised by being highly salient, but not necessarily primary in the sense of ‘non-derived’. That is, already saturation processes might provide as a value for some variable in the semantic form of an utterance what might be called a non-literal meaning. In contrast, there do seem to be processes whose function it is to derive a particular meaning from some underlying, basic, full meaning, such as metonymic shift, metaphorical interpretation, ad-hoc concept formation and – if one accepts Bach’s argumentation – free enrichment. 5.1.4 (Non-)Literal Meaning as (Non-)Basic Meaning In this section, I want to propose an alternative characterisation of the notions literal meaning and non-literal meaning, based on the assumption that linguistically coded 10 Note that Glucksberg and McGlone (1999)’s attributive category approach faces the same problem. 11 This is fully acknowledged by Carston (2002b), who suggests a few directions in which to turn for a possible solution to the problem, but does not clearly offer one herself. 222 Varieties of Meaning, Context and the Semantics/Pragmatics Distinction meaning actually is highly underspecified and the fact that these are very intuitive notions. I start by recapitulating the different uses of the two terms in the literature. Recall that the traditional view assumes of literal meaning that it is to be found at the level of expression meaning, stored in the lexicon as the lexical meaning of simple expressions and as such entering into the process of semantic composition, which, therefore, also results in larger structures that have literal meaning. Thus, in its use as characterising lexical meaning, literal meaning refers to some basic, primary, nondeviating, full meaning. Recall further that traditionally, it was assumed that to reach the level of utterance meaning or what is said, only such processes operate that resolve reference, assign values to indexicals and disambiguate the potential semantic structures generated by semantic composition. There is a sense in which the thus resulting meaning is literal in that it does not include any derived meaning.12 Thus, this particular contextually informed meaning might be called literal in the sense of non-derived, because the processes it results from only assign values to variables provided by the semantic form of the respective expression. Therefore, in its second use as characterising a particular pragmatically enriched meaning, literal is used to refer to some disambiguated meaning that is constituted of basic, primary and non-deviating meanings plus values assigned to variables in the semantic form of the utterance by the processes of fixing of indexicals and reference resolution. Note that this amounts to the assumption that what is contributed by saturation processes in general contributes to literal utterance meaning.13 Finally, a particular speech act determined by processes leading to the level of communicative sense might be called literal, because it is taken to be the basis for the hearer’s drawing of further inferences in his interpretation of what the speaker meant by his utterance, leading to an indirect speech act. Thus, similarly to the various uses of the expression literally cited at the beginning of section 5.1, three different uses of the term literal can be differentiated, one, where it refers to a particular type of meaning associated with simple expressions, one, where it refers to some contextually enriched meaning of an utterance and one, where it refers to the direct speech act attributed to the speaker. What all three uses share is the general aspect that the meaning described as literal is in some sense basic and underived, although the meaning levels the term is applied to differ in their complexity. Similar points can be made concerning the term non-literal meaning. That is, particular meanings may be associated with a simple expression, where those meanings are derived from some underlying basic full meaning and in that sense are non-literal. However, what is crucial is that traditionally, such non-literal readings for simple expressions were assumed to only arise due to a particular use of an expression by a particular speaker in a specific context. That is, in contrast to literal meaning, the term non-literal meaning traditionally is taken as not applicable at the level of expression meaning. In addition, the term is used to refer to utterances 12 This is so because traditionally, the lexical meanings that are composed into the sentence meaning of the complex expression uttered are necessarily literal. 13 However, as I will suggest below, under the assumption that lexical meaning actually is underspecified, what is contributed to utterance meaning by such processes as conceptual shift and conceptual differentiation actually may not necessarily considered to only involve primary or literal readings of a particular expression. Towards an Alternative Characterisation of (Non-)Literal Meaning 223 where what the utterance explicitly/implicitly means is not exhausted by what the sentence used to make the utterance means (i.e., Bach’s sentence nonliterality resulting from the process of expansion). Finally, the term non-literal is also used to refer to meaning aspects arising at the level of communicative sense, namely for cases in which what the utterance meaning describes, deviates in a fairly special way from the situation at hand or, generally, from what one might reasonably expect the speaker to intend to express (irony). Moreover, when it comes to the speech act a speaker is taken to have carried out with his utterance, indirect speech acts are usually characterised as non-literal, that is, as deviating from some underlying basic or primary direct speech act. Thus, as for literal meaning, for the term non-literal meaning (at least) three different uses can be differentiated, one, where it refers to a type of meaning associated with the particular use of a simple expression, one, where it refers to some contextually enriched meaning of an utterance and one, where it refers to the indirect speech act attributed to the speaker. In addition, the term non-literal is further used to characterise the meaning arising from a particular use of a complex expression (i.e., irony). Here, what the four uses of non-literal share is the general aspect that the meaning described is in some sense non-basic or derived and deviating from some relevant alternative meaning. However, as with literal meaning the meaning levels the term is applied to differ in their complexity. To summarise, both the terms literal meaning and non-literal meaning are used to describe particular meaning aspects at different levels of meaning. The only characteristic that the thus described meaning aspects have in common is that at the level of meaning at which they occur, they have a basic (thus being literal) or a deviating (thus being non-literal) status. However, note that this common characteristic does not allow for the inference that the kinds of meaning aspects described as literal or non-literal are in any way comparable or similar. This is because what is described as basic or deviating is basic or deviating with respect to the level of meaning at which the respective meaning aspect occurs. In other words, the fact that particular meaning aspects occurring at different levels of meaning are collectively described as either literal or non-literal does not make the thus denoted categories homogeneous. Thus, on the one hand, literal meaning at the level of expression meaning refers to context-free, stored basic meanings of expressions. Literal meaning at the level of utterance meaning refers to context-dependent, complex meanings, where those meanings do not include any non-literal meaning aspects. At the level of communicative sense, the term literal refers to yet a more complex meaning representation, including inferences concerning the speech act carried out by the speaker of the utterance in question. On the other hand, non-literal meaning used with respect to particular readings an expression might have, refers to a meaning of some more or less complex expression that traditionally is taken to be determined in the particular context in which that expression is used. When Bach speaks of sentence nonliterality, he refers to a meaning that is fully propositional but not solely the product of saturation processes. Irony is characterised as involving non-literal meaning because of the particular discrepancy between what the utterance made by a speaker means and what the speaker, given the particular utterance situation, can reasonably be taken to actually mean. Finally, indirect speech acts are characterised as non-literal 224 Varieties of Meaning, Context and the Semantics/Pragmatics Distinction due to the assumption that there is a conventional fit between sentence form used and speech act carried out, which is not adhered to in cases of indirect speech acts. Thus, it seems that if one wants a characterisation of literal meaning and nonliteral meaning that captures all the different phenomena at the different meaning levels these terms are used to refer to, that characterisation is going to have to be very general and vage. However, as indicated already in chapter 2 and afterwards, there are reasons for not counting some of the meaning aspects traditionally characterised as literal or non-literal, respectively, amongst the phenomena these terms in fact refer to. Let me repeat the claims I made concerning the nature of literal meaning and non-literal meaning in chapter 2. Neither literal meaning nor non-literal meaning can be found at the level of expression meaning, if – as I argued it should – the meaning available at this level is assumed to be highly underspecified, abstract and not actually constituting any full reading at all. Note that due to the assumption of underspecification of lexical meaning in general, I do not differentiate between linguistic meaning and salient meaning in Ariel’s sense. In other words, there are no full readings for expressions stored in the lexicon and even if there were, such meanings would not be unaffected by the features that determine the salience on a particular occasion of particular readings potentially expressed by a phonological form. Contrary to the traditional view, I argued in chapter 2 that both literal meaning as well as non-literal meaning can be found at the level of utterance meaning. Thus, what we have at this level of meaning are both literal as well as non-literal interpretations or readings for expressions below the sentence level. And although non-literal meaning can intuitively be characterised as ‘deviating’ from some underlying or basic meaning, in terms of the time-course of the interpretation process, a non-literal meaning need not necessarily be derived from some underlying meaning online, if that non-literal meaning is familiar, sufficiently frequent and conventionalised to a certain extent. Thus, the type of non-literal meaning characterised so far is what Bach calls constituent non-literality. However, recall that Bach argues for another type of non-literal meaning to be identified at the level of utterance meaning, namely sentence non-literality. He uses this term for a particular type of meaning resulting from the process of expansion operating on a level of meaning that is characterised as fully propositional (in Bach’s terms: what is said plus the results of the process of completion), but not the proposition the speaker is intuitively taken to have intended to express with his utterance. Note, however, that, intuitively at least, the relation between the type of meaning resulting from expansion and the minimal proposition of the utterance that process is based on is different from that between, say, the metaphorical or metonymic meaning of a given expression and the meaning these non-literal readings are based on. Whereas in cases of metaphor or metonymy, we start out from whatever is the basic meaning and end up with a meaning that differs in a principled way from that basic meaning, in case of expansions of minimal propositions, what we have as a result is not some different meaning, but a meaning that is more precise (hence Recanati’s term sense elaboration). In other words, in case of expansion it seems that meaning aspects are added to the minimal proposition to make it more precise, whereas in case of metaphor or metonymy meaning is ‘changed’. In this respect expansion is more similar to (particularised) Towards an Alternative Characterisation of (Non-)Literal Meaning 225 conversational implicature than to cases of metaphor or metonymy. That is, (P)CIs are drawn on the basis of the proposition expressed or impliciture and are added to it rather than replacing the utterance meaning. As such, they are not cases of what is normally intuitively understood as non-literal meaning. However, expansion and the processes leading to PCIs differ in that the former integrate sub-propositional meaning aspects into the minimal proposition, resulting in the proposition expressed, whereas PCIs are themselves full propositions inferred on the basis of the proposition expressed. Recall that traditionally irony is also considered a good example of non-literal meaning. That is because originally, irony was understood as involving two meanings, as it were, namely the one ‘literally’ expressed by the utterance made and a second, opposite to the meaning expressed by the utterance made but actually intended to be conveyed by the speaker. However, neither of the two theories of irony interpretation discussed in chapter 4 adhere to this traditional view of what is involved in irony. More specifically, recall that RT’s approach to irony does not even make reference to a non-literal meaning being involved in producing or understanding this ‘figure of speech’. However, what licenses the idea of the non-literal in cases of irony is the fact that there seems to be some contrast involved between what the speaker says and the specifics of the contextual circumstances in which she makes her utterance. Having said this, if one wants to characterise non-literal meaning as a meaning that (is derived and thus) deviates from some underlying, basic meaning and replaces it, irony actually should not be classified as a type of non-literal meaning. It seems that similar points have to be made with respect to indirect speech acts. Thus, recall that the idea is that indirect speech acts are non-literal in that they are derived from some priorily determined, underlying direct speech act, which in turn is determined based on the sentence form of the utterance made by a speaker. However, recall also the studies cited above which showed that indirect speech acts do not take longer to interpret than direct speech acts and that indirect speech acts may be understood although the assumed underlying or basic direct speech act is not. This suggests that, actually, what are called indirect speech acts are not derived from some priorily determined direct speech act after all (or at least not online). Note also that the idea of indirect speech acts relies on the assumption of the so-called literal force hypothesis, according to which particular sentence types are conventionally associated with particular speech act types or illocutionary forces. If one does not accept this hypothesis, then there actually is no such thing as an indirect speech act, since all speech acts have to be determined in the social context in which an utterance is made and the sentence type used in a particular utterance is no failsave guide towards the intended speech act. Having said this, as we saw in the studies on so-called indirect speech acts, there are, of course, conventionalised speech acts such that the interpretation process that determines the speech act carried out on a particular occasion of utterance may be short-circuited, as it were. After the whole discussion concerning the standard notions of literal meaning and non-literal meaning, characterising the former simply as basic and the latter as derived or deviating from basic meaning and, in a sense, replacing it might be viewed as a rather meagre and disappointing result. However, there are a few points 226 Varieties of Meaning, Context and the Semantics/Pragmatics Distinction in favour of this rather general characterisation. First, recall that in the standard characterisations of literal and non-literal meaning specific reference was made to their role in processing. As we saw, however, the distinction between literal and non-literal meaning does not necessarily play a role in the interpretation process. That is, although one might argue that there must have been a point in time at which a non-literal meaning for an expression was derived from its literal meaning, this fact does not of itself lead to perceivable differences in the interpretation of the expression in its non-literal meaning at some later point. Having said this, as was shown by Frisson and Pickering (2007) for the producer-product metonymy, under certain circumstances even interpreting a totally unfamiliar non-literal use of a particular expression may involve relatively little effort. Second the characterisation of non-literal as derived or deviating from some basic meaning allows the term to be applied to idioms. That is, the meaning associated with an idiomatic expression is conceived of as deviating from some basic meaning in the sense that speaker-hearers are able to consciously recognise that either the idiomatic expression does not have a compositional meaning and also no ‘homophonous’ counterpart with compositional meaning, or, for cases where there do exist compositional counterparts, that the idiom in question has a meaning that is not compositional and in that sense deviating. Third, note that non-literal meaning is not secondary in Recanati’s sense. That is, for an expression to be interpreted non-literally, the meaning of the utterance it is part of does not have to be determined first. Thus, I do not consider conversational implicatures and so-called indirect speech acts as good examples of non-literal meaning. Rather, both of these phenomena seem to me to have much more in common with free enrichment/expansion, which processes Recanati also does not characterise as resulting in non-literal meaning. Incidentally, the idea of above and below threshold activation Recanati resorts to in order to explain the non-literalness without secondariness found e.g., in metaphor and metonymy is interesting with respect to less or unfamiliar non-literal uses of expressions, which, it might be argued, are easier to identify as non-literal than more conventional non-literal uses. However, such latter uses can be identified as nonliteral by speakers on second thoughts, thus, non-literalness cannot be characterised entirely in terms of above/below threshold activation. Having said that, recall that solely relying on speaker’s intuitions is problematic, as there is the danger that, if asked, a speaker-hearer will try to establish the status of a particular reading of an expression as non-literal with respect to some other reading when in fact that post hoc established connection may not actually have had any role to play in the speaker’s understanding of the reading in question before.14 More generally, the problem is that, with the notion of salience cutting across the notion of literalness, there does not seem to be a principled way of establishing whether a particular reading of an expression is basic or not (since it is not necessarily the meaning that comes to mind first on encountering a particular expression). Be that as it may, the important point to note is that with the very general characterisations given for literal and non-literal meaning the two notions still cannot 14 This point will be taken up below in the discussion of loose use and RT’s characterisation of it in terms of ad-hoc concept formation. The Nature of Context in Utterance Interpretation 227 be used in a differentiation of semantics from pragmatics. Moreover, the fact that a particular meaning aspect is classified as literal or non-literal alone does not allow the drawing of any conclusions concerning the time-course of its interpretation. In light of these facts, it seems that the characterisation of literal meaning and nonliteral meaning given above actually is not very useful, beyond, perhaps, capturing intuitions behind the notions of literal meaning and non-literal meaning and being consistent. The major problem is that there seems to be no objective criterion by which to differentiate literal from non-literal meaning. That is, it is not clear on which basis – other then pure intuition – we judge a particular meaning to be basic. In particular, as we saw above, cases of literal and non-literal meaning cannot be differentiated on the basis of the processes that bring them about. 5.2 The Nature of Context in Utterance Interpretation As we saw throughout this thesis, in the discussion about the nature of the two levels of meaning utterance meaning and what is meant and whether they actually can and should be differentiated, assumptions about the nature of the contextual information necessary to determine the two respective levels of meaning play a crucial role. Thus, one aspect in which the various views on the nature of the level of utterance meaning differ is whether the processes determining this level are taken to be sensitive to considerations about the speaker’s potential intentions in making the utterance he did. The notions of narrow vs. broad context capture this difference in that although both consist of the information about who the speaker and addressee(s) of an utterance are as well as the time and place at which it takes place, only the latter also includes speaker intentions. In addition to the concrete utterance situation – involving or not involving assumptions about speaker intentions – general background assumptions are important when it comes to interpreting utterances. Thus, utterances are intuitively judged as (involving) literal or non-literal (meaning) against a particular set of background assumptions, where this, in turn, is taken to be sensitive to the particular utterance situation. More generally, especially in approaches that assume underspecification of lexical meaning, a further important pool of conceptual knowledge for interpretation processes to draw from is the overall world knowledge, which provides the actual readings for linguistic expressions. As argued above, knowledge concerning social interaction should be taken as distinct from overall world knowledge, since the former only contributes to the level of what is meant, whereas the latter is already important for the level of what is said/utterance meaning. It should be noted that when it comes to the interpretation of ongoing discourse, the notions of both narrow as well as broad context are not sufficient. Rather, what has been called a common ground is needed, that is, the context that is constituted and updated by the ongoing discourse the participants are engaged in. Interestingly, in the approaches considered above, this particular type of context has not been explicitly mentioned, as the focus rather is on the interpretation of single utterances and not so much on the interpretation of ongoing discourse. However, as I will argue 228 Varieties of Meaning, Context and the Semantics/Pragmatics Distinction below, the principles taken to hold for the interpretation of discourse to be coherent may also play a role when looking at the interpretation of individual utterances. What is more, particular assumptions made concerning the resources used in the interpretation of discourse may also be fruitfully exploited for the interpretation of individual utterances. Thus, in section 5.2.1, I will first introduce some important aspects concerning the interpretation of ongoing discourse. Moreover, Irmer (2009)’s suggestion will be introduced that in the interpretation of discourse, one source of information that is made use of is conceptual frames that are evoked by the speaker’s use of the expressions that make up his utterance. Taking up this idea, I will offer a proposal of how such implicit meaning aspects as taken by RT to be contributed by the process of free enrichment are determined without the need to take recourse to assumptions about the speaker’s intentions in making a particular utterance. Recall further that the different approaches to the nature of utterance meaning or what is said also differ in their conceptions of whether or not the semantics component of the language faculty should be taken to be concerned with context-dependent meaning and if so, to which extend. An interesting question to ask is why particular semantically oriented approaches to utterance interpretation should assume of the semantics component that it be sensitive to specific contextual information in the first place. I will take up this question in section 5.2.2 and argue that it has to do with assumptions about what types of operations semantics generally is taken to be involved in. 5.2.1 Context and the Interpretation of Implicit Meaning Aspects Free Enrichment and Implicit Meaning Aspects Let us begin with the idea within RT that free enrichment and, say, metaphorical interpretation are simply two different directions taken by one and the same process of ad-hoc concept formation. Actually, this view is glossing over some important differences between the two processes and the resulting meaning aspects. Thus, whereas in cases of metaphor what is being operated on is the concept associated with a simple expression, in case of free enrichment what we have is the addition of a conceptual element to the semantic form of the utterance, without there being a linguistically mandated slot for it. More specifically, the process of free enrichment is usually understood to operate on fully propositional forms.15 In fact, it seems that the conceptual elements contributed by free enrichment are actually standardised interpretations for the particular form the utterance takes. As such, they may be considered as belonging to the level of meaning proposed by Levinson for GCIs: utterance-type meaning. Garrett and Harnish (2008)’s experiment on particular types of what they call (following Bach) impliciture meaning aspects 15 Recall that Recanati assumes of free enrichment that it is a primary pragmatic process, where one of the latter’s characteristic is that it operates on sub-propositional forms. However, Recanati, in a sense, is forced to this treatment of free enrichment, since he takes the process to be subpersonal and thus not available to consciousness. See, however, Carston (2007) for a criticism of this view. The Nature of Context in Utterance Interpretation 229 corroborate this assumption. Thus, it is not only the case that speakers in saying ‘X broke a finger’ very often actually intend to express ‘X broke one of X’s fingers’, rather, even if they make use of this particular sentence in a context where it is clear that it is not a finger of X that is broken, this particular interpretation arises nevertheless. Thus, it seems to have a special status in that it arises despite not being intended. Recall that one of the arguments that lead RT to assume that the meaning aspects contributed by free enrichment are part of the explicitly expressed utterance meaning is the fact that they may play a truth-conditional role. However, as already mentioned, this fact may be captured in a way other than assuming that they have to be speaker-intended to begin with. Thus, although the term GCI suggests that what these inference are is some special type of CI, in fact, what they are is something that has been abstracted away from genuine CIs.16 In Bach’s terms, they have become standardised and as such arise independently of speaker intentions. However, because they are standardised, they have the ability to contribute to the truth conditions of the utterance they are part of in those cases where they clearly are intended. In other words, speakers (unconsciously) can make use of the fact that a particular utterance-type is standardly associated with a particular meaning aspect when they explicitly want to communicate that meaning aspect. They can rely on the fact that hearers will access the meaning aspect in question, once the respective utterance-type with which it is standardly associated is used/instantiated and that hearers will retain it as indeed intended, unless evidence to the contrary arises. As we saw, Burton-Roberts (2006) suggests that it is a defining property of GCIs only that they are cancellable. If this is so then it would supply a theoretical test for whether a particular meaning aspect taken to be contributed by free enrichment might be treated in terms of GCI: it has to be cancellable in principle. Unfortunately, the cancellation test only works for a subset of what are usually considered GCIs (e.g., by Levinson). Crucially, the test does not yield the desired results for the implicatures based on what Levinson calls the I-heuristic and which Garrett and Harnish (2008) tested. In fact, Burton-Roberts (2006) cites the following example to support the view that explicatures are not cancellable. (185) a. I haven’t had breakfast. b. I haven’t had breakfast today. c. ??I haven’t had breakfast, but I did today. Thus, it seems he does not count such I-based implicatures among GCIs. However, as Garrett and Harnish (2008)’s results indicate, these phenomena do show the main characteristics of GCIs as stated by Burton-Roberts (2006), namely arising even in cancelling contexts and arising when the utterances that give rise to them are decontextualised. Note also that for the positive variant of the above example, the cancellation of the implicit meaning aspect (IMA) does seem to be possible. (186) a. I’ve had breakfast. b. I’ve had breakfast today. 16 Cp. Borg (2001). 230 Varieties of Meaning, Context and the Semantics/Pragmatics Distinction c. I’ve had breakfast, but not today. Similar points can be made for the other types of the – in Levinson’s terms – I-based implicatures tested by Garrett and Harnish (2008). Thus, in the positive variants, cancellation of the IMA seems possible, but not so for the negative variants. (187) a. It is(n’t) raining. b. It is(n’t) raining here. c. It is raining, but not here. d. ??It isn’t raining, but it is raining here. (188) a. John broke/didn’t break a finger. b. John broke/didn’t break his own finger. c. John broke a finger, but not one of his own. d. ??John didn’t break a finger, but he did break one of his own. Thus, both the positive as well as the negative variants given in the (a) examples go along with the respective IMA explicitly spelled out in the (b) examples. The (c) examples show that it is possible to suspend the specific IMA standardly associated with the respective utterance-types in case of the positive variants. Nevertheless, note that intuitively, the (c) examples are understood as still expressing some IMA, although that meaning aspect is much more general than the one standardly associated with the respective utterance-type.17 Thus, looking back at (186c) the speaker is understood to have expressed that he has had breakfast before the day of the utterance, in (187c) the speaker is understood to express that it is raining somewhere other than the location where he makes his utterance and in (188c) the speaker is understood to express that John did break a finger, but that it wasn’t one of his own. Note, however, that whether or not a particular IMA is cancelled seems to depend very much on the terms in which the putative cancellation is couched. Thus, consider Korta (1997)’s example of an impliciture cancellation, of which he assumes that it is quite acceptable. (189) I haven’t eaten breakfast, but I do not mean that I haven’t eaten breakfast this morning. In fact, I have never eaten breakfast. Thus, what seems to be going on in this example is similar to what we saw for (186c), namely, the IMA today is cancelled, resulting in an understanding of the speaker’s utterance as expressing that he has not had breakfast prior the time of utterance, in other words, he’s never had breakfast. That is why the IMA in this particular case cannot be cancelled by the sort of continuation offered by Burton-Roberts (2006) (i.e., but I did today), as this would be contradictory to how the sentence is understood without the IMA. The same holds for the other examples stated above. That is, (187c) and (188c) are odd, because if the respective IMAs are cancelled, then continuing the respective sentences in the way it is done by Burton-Roberts 17 This fact might actually be taken to show that what we have here is in fact not genuine cancellation. See, however, below for a general critique of the cancellation test. The Nature of Context in Utterance Interpretation 231 (2006) leads to contradiction in both cases. Thus, if the IMA here is cancelled in (187c), what we get is an interpretation where it is not raining anywhere. If the IMA john’s own is cancelled in (188c), the interpretation would be that John did not break any finger. Moreover, trying to apply the cancellation test to scalar implicatures as in (190) and (191) below shows that even with those examples, that is, the ones that BurtonRoberts (2006) wants to establish are cancellable and thus genuine GCIs, the test turns out to be problematic. (190) a. Some of the students went to the party. b. Some but not all of the students went to the party. c. Some of the students went to the party, in fact all of them did. d. None of the students went to the party. e. ??Some of the students didn’t go to the party, in fact all of them did. f. ?? It is not the case that some of the students went to the party, in fact all of them did. Thus, with the scalar expression some, it isn’t actually possible to have a negative variant of the sentence in (190a), in which the implicature-triggering scalar some occurs. Therefore, the negation (190d) of (190a) cannot be said to cancel an implicature, because no scalar implicature is triggered in the first place. Trying to formulate a negative variant of (190a), as is done in (190e), which potentially does give rise to a scalar implicature shows that in such a case a continuation which is supposed to cancel the implicature makes this sentence contradictory as well. More specifically, (190e) is acceptable only if considered an example of metalinguistic negation. Consider now an example involving a numeral. (191) a. John has got three children. b. John has got exactly three children. c. John has got three children, in fact, he has four. d. ?John doesn’t have three children, he has four. Here it is easier to formulate a negative variant of the positive sentence that gave rise to the implicature. However, as in the example before, the negative sentence actually does not give rise to the scalar implicature, so, again, there can’t be any cancellation here. Nevertheless, in the case of (191d), if this is understood as involving genuine truth-functional negation rather than metalinguistic negation, it again is contradictory. Generally, what is problematic about the idea of cancellation of GCIs is that the phenomena categorised as GCIs differ in some crucial aspects and this fact actually makes it difficult to apply one overall theoretical test to decide whether a particular phenomenon should be treated as a GCI or not. Thus, it seems that if one wants to verify whether a particular putative IMA is characterised by the fact that it arises even when it is not intended by the speaker and that it arises out of a particular context of utterance, one has to resort to psycholinguistic (and possibly to neurolinguistic) experimental methods. Having said this, one would still like to have a theory of how such standardised IMAs arise and what exactly is going on in 232 Varieties of Meaning, Context and the Semantics/Pragmatics Distinction those situations in which it is clear that they are not intended by the speaker. In other words, if one wants to assume a level of utterance-type meaning, the question is how this figures in the actual interpretation process and on which basis such standardised meaning aspects are integrated into (and, as the case may be, deleted from) the semantic form of the utterance at hand. Discourse Interpretation and Information from Conceptual Frames One possibility is to assume that in interpreting a particular utterance, what the hearer makes use of is information stored in frames that are evoked by the expressions the speaker used in his utterance (e.g. Baker et al. 1998, Nissim et al. 1999, Baker et al. 2003, Fillmore et al. 2003, Ruppenhofer et al. 2010). Irmer (2009) suggests this idea in his analysis of bridging inferences18 within the framework of Segmented Discourse Representation Theory (SDRT) (e.g., Asher and Lascarides 2003, Lascarides and Asher 2007). I will briefly describe Irmer (2009)’s approach below. Thus, consider the example in (192). (192) a. John was murdered yesterday. b. The knife lay nearby. On a first blush, there is nothing in these two sentences that makes their sequence overtly coherent. Nevertheless, they are understood as forming a text. More specifically, the knife in (192b) is understood as potentially referring to the instrument used by the unmentioned murderer in his act of murdering. Understanding knife in (192b) as referring to the potential instrument used in the murdering event mentioned in (192a) is the result of a bridging inference. The question is how such a bridging inference is drawn. In particular, what is crucial here is the question on which kinds of information such inferences may be based. One likely source of information is the so-called common ground against which verbal communication takes place. This consists of knowledge about the situational context, the discourse context as well as (relevant) world knowledge. The situational context involves the people participating in the discourse, the time and place at which the discourse takes place and the objects present (as they may be potential referents for indexical expressions). The discourse context includes the propositions expressed or conveyed by preceding utterances. Depending on the general topic of the discourse, aspects of world knowledge relevant to that topic are also taken to form part of the common ground. This last point is especially interesting in the present discussion, as the question arises how what is taken to be relevant information is identified and becomes part of the common ground. Irmer (2009)’s suggestion, at least with respect to bridging inferences, is to assume that during the interpretation of an utterance, specific pieces of conceptual knowledge drawn from frames are integrated into the semantic form of the utterance at hand. Before looking at how this works in more detail, let us first turn to how such frames are characterised in the framework Irmer (2009) makes use of, namely Frame Semantics. 18 Note incidentally, that in Levinson’s classification of implicatures, bridging inferences actually are treated as a type of I-implicature. The Nature of Context in Utterance Interpretation 233 Definition: A Killer or Cause causes the death of the Victim. Core Frame Elements: FE Killer description The person or sentient entity that causes the death of the Victim Victim The living entity that dies as a result of the killing Instrument The device used by the Killer to bring about the death of the Victim Cause An inanimate entity or process that causes the death of the Victim Means The method or action that the Killer or Cause performs resulting in the death of the Victim inherited FE sort Agent sentient Patient sentient Instrument physical entity Cause Means state of affairs Non-core Frame Elements: Beneficiary, Manner, Place, Purpose, Time . . . Lexical Units: annihilate.v, annihilation.n,. . . murder.v, murderer.n, . . . Figure 5.2: The Killing frame As the name suggests the leading idea of Frame Semantics is that world knowledge is organised in frames. More specifically, it is stereotypical situations that frames are mental representations of. The ideas of Frame Semantics are implemented in the on-line lexical resource FrameNet, which currently consists of about 11 6000 lexical units, 960 semantic frames and 150 000 annotated sentences.19 . In FrameNet, each semantic frame consists of a number of elements, the frame elements (FEs), some of which are taken to be crucial to the identification of that particular frame (i.e., conceptually necessary core frame elements); some of which are peripheral or extra-thematic (i.e., not conceptually necessary non-core frame elements). Peripheral FEs mark such notions as Time, Place, Manner, Means, Degree, and the like. They do not uniquely characterize a frame, and can be instantiated in any semantically appropriate frame. [. . .] Extrathematic frame elements situate an event against a backdrop of another state of affairs, either an actual event or state of the same type, . . . or by evoking a larger frame within which the reported state of affairs is embedded . . .. (Ruppenhofer et al. 2010, p. 27) 19 For an illustration of a semantic frame, consider the Killing frame in (5.2) evoked by the lexical unit murder used in (192.) 234 Varieties of Meaning, Context and the Semantics/Pragmatics Distinction Each frame has a definition, which is an informal description of the concept the frame represents. Moreover, the lexical units associated with the particular frame are listed. A lexical unit is an unambiguous sound-meaning pairing, thus, polysemous expressions consist of a number of lexical units, each of which might potentially be associated with a different frame. If a particular lexical unit is used in an utterance, it is said to evoke the frame it is associated with. In addition to the information listed above, the relation of a given frame to other frames is indicated.20 Thus, a particular frame might inherit some or all of the elements of a more abstract frame; a particular frame might be a subframe of a more complex frame, etc. Note also that the more abstract frames might actually not have any lexical units associated with them. Their function is rather to relate the more concrete frames to one another. The core FEs roughly correspond to the semantic roles involved in the eventuality conceptually represented in the frame. However, as can be seen from example (192a), not all core FEs need necessarily be expressed in a sentence about an instance of the particular eventuality. On the one hand, this might happen for structural reasons (e.g., 192a is a passive sentence, thus, the subject of the corresponding active sentence is not syntactically obligatory), on the other hand, the unexpressed core FEs might be implicitly understood or inferrable from the context. To explain how such brigding inferences as in (192) are drawn, Irmer (2009) proposes to integrate the core FEs of the frame evoked by the verb used into the semantic form build up for the respective utterance. Thus, whereas in SDRT the basic semantic form for the first sentence in (192) looks as in (193), with the integration of the core FEs from the Killing frame evoked by murder, it looks as in (194). e, j (193) named(j, john) murder(e) patient(e, j) e, j | x1 , x3 e : Killing (194) killer(e, x1 ), x1 =? victim(e, j), named(j, john) instrument(e, x3 ), x3 =? For reasons of space, I do not want to introduce the formalism of SDRT in too much detail. However, a few notes on the graphical representation used in that theory: in the upper part of the horizontally divided boxes the discourse referents introduced by the respective utterance are given, whereas in the lower part conditions holding for those referents are stated. The conditions might involve properties of discourse referents and their relations to other referents. The vertical line used in the upper part 20 For reasons of space, these relations are omitted in 5.2. The Nature of Context in Utterance Interpretation 235 of the box in (194) is introduced by Irmer (2009) in order to differentiate between what he calls regular (i.e., linguistically introduced) and weak discourse referents (i.e., introduced through evoked frames). In this particular case, the Victim discourse referent introduced by the Killing frame has already been identified with the linguistically given discourse referent j, as this is taken to have the property of being a Patient and the thematic role Victim actually is a specific type of Patient.21 The Killer and Instrument roles, in contrast, remain unspecified (indicated by the notation x1 =?, x3 =?). However, in (192), the second sentence linguistically introduces a discourse referent that may be identified with the Instrument role. This is because the expression knife evokes the Weapon frame with the core-FE Weapon. However, and as Irmer (2009) points out, although knives can be used as weapons, they are not necessarily designed as such. ‘The only knowledge we can use is that there is no clash of sorts: both knives and killing instruments are physical entities’ (Irmer 2009, p. 203). Nevertheless, identifying the knife introduced in the second utterance with the weak discourse referent x3 that is a murder-instrument from the first utterance is plausible and results in a consistent interpretation. Having said this, generally, what is even more important for the identification of discourse referents to be at all possible is that a particular discourse referent introduced earlier is actually accessible for the newly introduced discourse referent. Without going into the details here, accessibility is established subject to the discourse relation assumed to hold between the utterances under consideration.22 Free Enrichment and Information from Conceptual Frames As already said, Irmer (2009) restricts the elements used from the frames to those that are characterised as core FEs. In fact, he only concentrates on the core FEs of frames evoked by the main verb in the respective utterances. However, all content words evoke frames and generally it can be assumed that the information in the individual evoked frames in some way or other enters into the interpretation of the utterance at hand. In addition, there is a case to be made for allowing core as well as non-core elements of evoked frames to figure during the interpretation process. In that case, the information provided by the evoked frames may be considered to form the minimal background, as it were, against which the utterance in question is interpreted. In this connection, recall that Irmer (2009) introduces a two-way distinction between regular and weak discourse referents. However, as he himself suggests, a more fine-grained differentiation between discourse referents with respect to their salience might be useful. A differentiation of discourse referents based on their salience in a particular context would allow for an assumption concerning the operation of the processes resolving underspecification of discourse referents: only such weak discourse referents down to a particular degree of salience will be tried to get resolved, identified with others, etc. Turning now to the IMAs taken to be contributed to an utterance’s meaning 21 In this particular case, this relation is due to the inheritance relation between the Killing frame and its parent frame Transitive action, which, amongst others, has the two core-FEs Agent and Patient, which are inherited by the Killing frame as the core-FEs Killer and Victim, respectively. 22 In the case of (192), this being Background. 236 Varieties of Meaning, Context and the Semantics/Pragmatics Distinction by the process of free enrichment: recall once again that the meaning aspects contributed are not in any way linguistically mandated; they are unarticulated constituents. Recall further the examples Wilson and Sperber (2000) use to argue against Stanley (2000)’s hidden indexical approach. (93) I’ve often been to their parties, but I’ve never eaten anything [there]. (94) I must wash my hands. I’ve eaten [using my hands]. Wilson and Sperber (2000) use the above examples to argue that if one assumes a hidden indexical approach, one would be forced to posit such variables in the semantic forms of each of the second sentences in the respective examples above. I already indicated in chapter 3 that I do not think this argumentation goes through. This is because in these examples what we have is actually small texts and the IMAs assumed to be expressed by the speaker in the respective second sentences of these texts can be taken to ‘arise’ due to the particular discourse context in which the respective utterance takes place. The question, of course, is, what is meant by ‘arise’. Here, I think, we can follow Irmer (2009)’s suggestion that during the interpretation of utterances, information from evoked frames is integrated into the semantic form build up for that utterance. Thus, let us concentrate on (93) first, starting with the first utterance. Unfortunately, FrameNet does not provide information on the frame(s) associated with the copula, however, intuitively the sense expressed in (93) could be paraphrased with go to, which does have an associated frame in FrameNet, namely the Attending frame, with the core-FEs Agent and Event. The noun party, in turn evokes the Social Event frame, with the core-FEs Attendee and Social event. Assuming that these latter two roles can be traced back to the roles Agent and Event23 , respectively, by some inheritance relation, the more specific roles can be identified with the more general roles and integrated in the representation of the utterance’s meaning. Note that in addition to its core-FEs, the frame Social Event has quite a number of non-core FEs, amongst them Refreshment as well as Place. Turning now to the second sentence in (93), the verb to eat, in turn, evokes the Ingestion frame (see figure 5.3), with its core-FEs Ingestor and Ingestibles, as well as, amongst others, the non-core FEs Time, Place and Manner. Intuitively, the place at which the eating-event expressed in the second utterance takes place is interpreted as identical to the place of the respective party. This intuition can be captured by assuming that the two non-core FEs place of the two respective frames are identified, although not further specified.24 Moreover, it seems that the information that a party stereotypically involves refreshments is necessary to understand the discourse in (93) as coherent in the first place. Thus, the core-FE Ingestibles from the Ingestion frame evoked by the second utterance is related to the non-core-FE 23 Unfortunately, FrameNet as yet has not specified the inheritance relations of the Attending frame to other frames. 24 This identification is based on the principle introduced by Cohen (2006) Equality by Default: Two discourse referents are assumed to be equal, unless it can be proven that they are distinct. As this is a very general principle and as such prone to overgeneralisation, it should be mentioned that there exist quite a number of more specific principles that constrain the applicability of this particular principle. (See, e.g., Irmer (2009) for more information on the interaction of this principle and more specific ones.) The Nature of Context in Utterance Interpretation 237 Definition: An Ingestor consumes food or drink (Ingestibles), which entails putting Ingestibles in the mouth for delivery to the digestive system. This may include the use of an Instrument. Sentences that describe the provision of food to others are NOT included in this frame. Core Frame Elements: FE description Ingestible The Ingestibles are the entities that are being consumed by the Ingestor Ingestor The Ingestor is the person eating or drinking inherited FE sort Substance n. s. Agent Sentient Non-core Frame Elements: Degree, Duration, Instrument, Manner, Means, Place, Purpose, Source, Time Lexical Units: breakfast.v, consume.v,. . . drink.v, eat.v, . . . Figure 5.3: The Ingestion frame Refreshments of the Social Event frame evoked by the first utterance.25 Thus, it is the background information that a party might involve refreshments that makes it appropriate to express some sort of contrast by using the expression but, although overtly what is talked about in the individual sentences (parties in the first, eating in the second) is not necessarily related. This can be seen when considering, e.g., Asher and Lascarides (2003)’s characterisation of the discourse relation Contrast. Generally, the authors argue that the relation of Contrast between two utterances α and β holds, if β involves one of the expressions but, however or in contrast. This condition is trivially fulfilled by (93). In addition, Asher and Lascarides (2003) argue further that the two utterances be structurally similar while semantically dissimilar or contrasting. Arguably, the condition that the semantic content of the two utterances be contrasting is only fulfilled by assuming that attending a party normally involves partaking of refreshments. Let us now turn briefly to the interpretation of (94). Intuitively, the relation between the two utterances is one of Explanation. Thus, the second utterance in this small discourse is understood as providing an explanation for the proposition expressed in the first. If the speaker’s hands introduced by the first utterance are identified as the instrument used for eating, the explanation becomes plausible. In fact, the Ingestion frame evoked by the verb to eat does include Instrument 25 Although this is not specified in FrameNet, the idea would be that Ingestibles and Refreshments are related with respect to their semantic sorts. The problem here is that the two frames the respective FEs are part of are not in any inheritance relation, which actually makes it difficult in FrameNet to relate them to one another other then by intuition. 238 Varieties of Meaning, Context and the Semantics/Pragmatics Distinction amongst its non-core FEs. Needless to say, as in example (192), identifying the aforementioned speaker’s hands as the instrument used in the eating-event requires recourse to world knowledge that it is possible for a person to use his hands as the instrument in an eating event. Let us now turn to examples involving IMAs that consist of only single sentences. For example, the speaker’s use of breakfast in (186a) arguably again evokes the Ingestion frame, where one of the non-core frame elements (FEs) is the time at which ingestibles are ingested.26 (186a) I’ve had breakfast. In fact, according to FrameNet, in this particular case, the expression to have also evokes the Ingestion frame. The fact that the sentence is in the past tense may lead to a partial specification of the non-core FE Time as having taken place at some point before the time of utterance. By taking into account knowledge about usual practices in the culture at hand, namely that it is quite usual for people to have breakfast once every day, the relevant time span may be restricted to the morning hours of the day of utterance. Consider now the interpretation of an utterance of (188a) – another example of a meaning aspect, according to RT, due to free enrichment. (188a) John broke/didn’t break a finger. In an utterance of (188a), using the expression break evokes a frame Experience bodily harm with the core FEs Body part and Experiencer, while finger evokes a frame Observable Bodypart with the core FEs Body part and Possessor. Obviously, the referent of finger can be identified as playing the role of body part, whereas John is an accessible discourse referent and an appropriate value for the Possessor and the Experiencer-variable, where the former is identified with the latter, the latter being retained in the representation of the utterance’s meaning as it is more specific.27 Consider next example (78), repeated below. (78) a. You’re not going to die. b. You’re not going to die from this cut. Here, the expression to die evokes the death frame, which has as one of its non-core FEs the cause of death. From the situation in which this utterance is said to be made, the cut the addressee has suffered might potentially be the cause of death, if it is severe. Now, it seems that what the speaker expresses is that the addressee’s cut is not going to be the cause of his death. Incidentally, this remainds me of another example, mentioned by Bach (2000) and involving the positive variant of (78a). The situation is one where an ontologist 26 Note that FrameNet only lists the verb to breakfast, which, however, does evoke the Ingestion frame. 27 Once again it should be noted that the two frames Experience bodily harm and Observable Bodypart are not in any inheritance relation. However, the informal characterisation of the FE Experiencer in the former frame entails that the referent filling this role be a possessor of the referent filling the FE body part. The Nature of Context in Utterance Interpretation 239 is talking to one of his patients about that patients’ cancer. The patient wants a frank prognosis and the ontologist utters (195a). (195) a. You’re going to die, but not from this cancer. b. You’re going to die. c. You’re going to die from this cancer. Here it seems that the speaker is alluding to the common sense knowledge that all living beings must die, the knowledge that death might have an identifiable cause as well as the common sense knowledge that having cancer is a potential cause for death. It is likely that in the particular utterance situation assumed the addressee will interpret the first part of (195a) as expressing (195c). That is, the verb to die again will evoke the death frame and, due to the particular situation in which this utterance takes place, the non-core FE element cause will get specified as the addressee’s cancer. However, this is taken back in the second part of the ontologists’ utterance. In other words, the fact that a death might have a cause remains, but in this situation the cause is not specified, or rather, what is expressed is that the patient’s cancer is not going to be the cause for his death. Finally, an utterance of (187a) is normally understood as (187b), because the verb to rain evokes the Precipitation frame amongst whose core-FEs are time and place. (187a) It is(n’t) raining. (187b) It is(n’t) raining here. Since the sentence used is in the present tense and progressive aspect, the time variable of the Precipitation frame can be filled by the time of the context of utterance, which can be assumed to be given anyway as it is needed for the resolution of deictic expressions. The same goes for the place variable. Since the context of utterance provides a value for that place variable, this will be used, unless there is some information in the discourse background that makes such an identification inappropriate. A further interesting point to note concerns a potential interaction of the expressions used in an utterance when it comes to deciding which lexical unit to assume in those cases where an expression is polysemous. Thus, recall once more example (186a), in which the expression to have is used. Actually, in the FrameNet database, there are several entries for this verb, each of which is associated with a different frame. (186a) I’ve had breakfast. have.v have.v have.v have.v have.v Have associated Inclusion Birth Ingestion Possession 240 Varieties of Meaning, Context and the Semantics/Pragmatics Distinction However, as already indicated, I think it is likely that the noun breakfast is associated with the Ingestion frame as well and this fact will influence the choice of the appropriate lexical unit for the expression to have (i.e., that also evoking the Ingestion frame). The same can be said for Ruhl (1989)’s example (52). (52) The thief took the jewels. The verb take is highly polysemous and, in addition, occurs as part of quite a number of phrasal verbs. take.v take.v take.v take.v take.v take.v take.v take.v take.v Has as requirement Taking time Capacity Taking Ride vehicle Ingest substance Removing Bringing Conquering In one of its numerous readings, take is associated with the Taking frame. The co-occurrence of the two expressions take and thief, where the latter evokes a frame Theft, which actually inherits from the Taking frame, might lead the interpretation process to assume that the lexical unit the expression take in this particular utterance is used to express, actually is the one that is associated with the Taking frame. Having analysed the explicature/impliciture examples above with the help of (more or less) general frames and their elements, the question arises where the notion of a level of utterance-type meaning comes in. Note that the examples I have analysed were mostly such that there was no evidence that the IMAs were not appropriate. As a last point, I want to make a suggestion as to what happens if a sentence like (188a) is interpreted in a context in which it is clear that it is not John’s finger that is broken. (188a) John broke/didn’t break a finger. As we saw from the experimental results obtained by Garrett and Harnish (2008), it seems that the specific IMA ‘his own finger’ is initially interpreted, even though the forgoing context does not support it and it, thus, has to get cancelled. A first thing to note is that actually, the dispreferred but appropriate reading in which John breaks someone else’s finger involves a sense of break that differs from that used in the preferred reading, thus evoking another frame, namely the frame cause harm with the core-FEs Agent, Body part, Cause and, crucially, Victim. Whereas in the Experience bodily harm frame it is the Experiencer that is identified as the Possessor of the Body part under discussion, in the case of the cause harm frame, it is the Victim that is identified as the Possessor of the Body part. The question is how the preferred reading of (188a) can be explained. Note that in this particular case, co-occurence of the expression to break with finger is not of any help, as the frame Observable bodypart evoked by the latter is plausible with The Nature of Context in Utterance Interpretation 241 both the relevant readings of break. However, an important fact that I have not elaborated on is the fact that discourse interpretation is taken to be subject to certain constraints or principles.28 One of these constraints is Asher and Lascarides (2003)’ meta-rule ‘maximise discourse coherence’, which consists of four parts, one being ‘resolve underspecifications’ and the one of special interest in the present discussion. Thus, note that if to break is assumed to evoke the frame cause harm, the Possessor core-FE of the Observable bodypart frame evoked by finger would have to get identified with the Victim core-FE of the cause harm frame, where this actually remains underspecified in an utterance of (188a). In contrast, if the frame Experience bodily harm is used, the Observable bodypart’s Possessor core-FE will be identified with the former frame’s core-FE Experiencer which is actually assigned the value john. Thus, if the frame Experience bodily harm is used rather than the frame cause harm, more underspecifications will get resolved, which makes this interpretation preferrable.29 In a sense, then, the level of utterance-type meaning – at least as far as the kinds of meaning aspects are concerned we looked at here – boils down to the preferred resolutions of unspecified potential discourse referents and their relations. A consequence of this view is that GCIs – at least those traditionally taken to be based on Levinson’s I-principle – actually are not implicatures at all but rather IMAs, integrated into the semantic form of an utterance during its interpretation. This characterisation, of course, only holds if the primary characteristic of implicatures generally is taken to be that they arise on the basis of propositions. If the primary characteristic is rather taken to be the idea that implicatures arise on the basis of particular conversational maxims or principles, then the IMAs discussed above might still be characterised as implicatures, as their integration into the semantic representation of an utterance, as we saw, is guided by overarching principles on general discourse interpretation. Be that as it may, note that the approach sketched here captures Recanati’s intuition that free enrichment is a primary pragmatic process, although one characteristic of these processes is that they do not need a proposition as a basis and free enrichment normally is taken to be based on propositions. In fact, with an approach that assumes that such IMAs are the result of the integration into the semantic form of an utterance of information from frames evoked by individual expressions used in an utterance, the process of free enrichment is no longer characterised as applying to propositions. Consequences What makes an approach to utterance interpretation using frames interesting is the fact that it allows to keep the semantics of the expressions used in an utterance minimal, while still allowing particular meaning aspects to figure in the interpreted 28 as, e.g., the principle Equality by Default mentioned below. Note also that the preferred reading involves a model of the world that is more minimal than the model for the dispreferred reading, as the former involves a smaller number of discourse referents than the latter (i.e., the FE victim). As Irmer (2009, secion 5.2) points out, the notion of minimality of models plays an important role in discourse interpretation. Thus, the preferrence for a particular reading of an ambiguous expression may be traced back to the fact that the model for that reading is more minimal than the model for the alternative reading(s). 29 242 Varieties of Meaning, Context and the Semantics/Pragmatics Distinction utterance, where those meaning aspects may or may not get specified and where they are defeasible. Moreover, assuming that expressions evoke particular frames that provide a restricted number and type of elements as potential discourse referents restricts the search space for values for newly introduced discourse referents in an enfolding discourse. In addition, and as is also argued by Irmer (2009), introducing such variables for potential discourse referents does not mean that they have to get specified in the course of utterance or further discourse interpretation. That is, it is quite possible that such variables will remain unspecified. Conversly, as we saw above, it is always possible to clarify the exact value assigned to some framerelated element to the point where the element may in fact stay underspecified after all.30 Crucially, such an approach allows to characterise the process called free enrichment as contributing meaning aspects to the meaning of an utterance, where those meaning aspects are not the values of semantically provided slots. However, free enrichment can still be seen as a type of saturation process and thus, in a sense restricted, in that it provides values to variables introduced into the utterance context by frames that are evoked by the expressions used in a particular utterance. Moreover, free enrichment can be assumed to actually not rely on a consideration of the speaker’s potential intentions in making the utterance he did. In a sense, then, assuming that the interpretation process makes use of information from frames is to eliminate hidden indexicals from the semantics of expressions to reintroduce them into the conceptual information used by the interpretation process. However, this seems to be indicated anyway by the fact that background information is generally assumed to play an important role in determining even what is traditionally called the literal meaning of an utterance. Moreover, since frame-related elements are integrated into the semantic form of an utterance while this is built up by the interpretation process, one would actually expect those elements to have a truth-conditional effect. However, the exact effect they will have on a particular occasion depends on the specificity of the actual value assigned to them. 5.2.2 Context, Semantic Interpretation and the Semantics/ Pragmatics Distinction As we saw, especially in chapter 3, approaches to utterance interpretation differ in their conception of the semantics component in the language faculty. In particular, one cause for dissonance is the question of whether what the semantics component delivers should be taken to be propositions or only sub-propositional semantic forms. Approaches assuming that semantics only delivers sub-propositional forms do so based on the ideas that, generally, semantics only deals with context-independent 30 Thus, recall the example sentences such as repeated below, where the preferred interpretation of the respective IMA is taken back but the general element this IMA is taken to be the value of is still felt to be present. (i) It is raining, but not here. (ii) I had breakfast, but not today. (iii) John broke a finger, but not one of his own. The Nature of Context in Utterance Interpretation 243 meaning aspects of expressions and that those context-independent meaning aspects are not sufficient for the determination of the proposition actually expressed by the respective sentence in a particular utterance situation. Approaches assuming that semantics does in fact determine the proposition semantically expressed by some sentence do so on the basis of the assumption that – contrary to its traditional characterisation – semantics in fact is not concerned with context-independent meaning only. Rather, semantic processes of meaning composition take into account particular types of contextual information. However, the approaches differ in what they take the nature of these pieces of contextual information to be (i.e., objectively attainable contextual information for Borg; contextual information partly influenced by assumptions about the speaker’s intentions in making his utterance in Cappelen and Lepore’s approach). In any case, contextual information is assumed to be integrated into the semantic form of an utterance, subject to the presence of some linguistically given element that licenses such an integration. That is, Borg as well as Cappelen and Lepore do not assume that unarticulated constituents should be taken to be part of a sentence’s meaning relative to a context of utterance. Having said this, it should be noted that the information provided by linguistic semantics alone may not be sufficient to capture what a reader or hearer understands when she is presented with a sentence out of context. Thus, even if I do not know who John is, when reading a sentence such as John broke a finger, I will understand whoever the name John refers to to have broken one of his own fingers. It is clear that the relevant IMA is not semantically provided and as such is defeasible. However, it arises without a particular context being necessary. Thus, even if applying the traditional criterion for semantic meaning – that it is the meaning understood when presented a sentence without any contextual information (Katz’s anonymous letter sitatuation) – this is not strict enough to exclude so-called unarticulated constituents. Thus, the IMA ‘his own’, in Levinson’s terms, is a preferred interpretation that is determined without any consideration of the potential speaker’s intentions or a likely context of utterance being necessary. Be that as it may, the important point to note is that if it is propositions that the semantics component of the language faculty delivers, then that component can also be assumed to be involved in what is called the semantic interpretation of those propositions. If, however, the semantics component only delivers sub-propositional semantic forms, it cannot be involved in the semantic interpretation, since these semantic forms do not yet allow for an evaluation concerning their truth or falsity. In a sense, then, the question of the semantics/pragmatics distinction seems to boil down to whether the process of semantic interpretation should be assumed to be part of the semantics component. Recall that both in Borg’s as well as Cappelen and Lepore’s approaches semantic interpretation is taken to be part of the semantics component. However, the truth conditions delivered and semantically interpreted by that component are only liberal truth conditions rather than the actual truth conditions assumed to be involved in the particular utterance situation. However, in both approaches it is assumed that these liberal truth conditions have an important role to play. Thus, Borg assumes that they establish the inferential relations between the expressions of a language, whereas Cappelen and Lepore assume that they are the minimal level of meaning a 244 Varieties of Meaning, Context and the Semantics/Pragmatics Distinction hearer can fall back on in cases where for some reason or other it is not quite clear what the actual truth conditions should be taken to be. Turning back to the approaches that assume of the semantics component that it ‘only’ delivers sub-propositional forms: recall that RT in particular does not even recognise a level of meaning corresponding to Borg or Cappelen and Lepore’s liberal truth conditions as playing any role whatsoever in the interpretation process. Rather, RT assumes that it is explicatures, that is, pragmatically enriched semantic forms that play an important part in the interpretation process, as they can be taken to constitute the level of meaning that is the basis for further pragmatic inferences as well as that level of meaning to the expression of which the speaker is bound. Thus, even if one were to assume that what the semantics component of the language faculty delivers are propositions, rather than propositional radicals, semantic interpretation would still have to be assumed to take place again once the actual truth conditions have been determined. In other words, even if an evaluation as to the truth or falsity of liberal truth conditions might be possible within the semantics component of the language faculty, according to all approaches discussed above, it is not possible for the actual truth conditions, thus, those truth conditions which can be considered as the ones really mattering in actual communication situations. Note that Levinson’s as well as RT’s conceptions of semantics seem like attempts at preserving the intuition that semantic interpretation should be part of some sort of semantics system. Thus, whereas Levinson assumes that the semantics component in the language faculty is intruded upon by a restricted number of pragmatic processes, leading him to differentiate between two semantic sub-systems, RT assumes that there actually are two different semantic systems, namely linguistic vs. real semantics, where the latter actually is the ‘language of thought’ and thus the locus of semantic interpretation. What Levinson’s conception better captures is the fact that the information contributed by the pragmatic processes intruding into semantics is essentially defeasible, non-monotonically determined conceptual information. As such these processes are crucially distinct from the kinds of processes involved in semantic interpretation, which essentially are monotonic. That is, the drawn conclusions as such are not defeasible; only the premises, as it were, on which they are based might turn out to be false (as the premises may involve pragmatically determined IMAs). For example, suppose Mary knows that (196d) holds and in her conversation with Jane concerning how to spend their time Jane answers Mary’s question (196a) as in (196b) and Mary interprets this as expressing (196c). On the basis of this information, she will be able to draw the conclusion (196e). (196) a. Mary: How about playing tennis? b. Jane: It is raining. c. It is raining in Leipzig at the moment. d. If it is raining at location li at time tj , it is not possible to play tennis at location li at time tj e. It is not possible to play tennis in Leipzig at the moment. This conclusion follows monotonically from the two premises (196d) and (196c). However, the determination of the proposition (196c) functioning as one of the The Nature of Context in Utterance Interpretation 245 premises actually involves pragmatically determined meaning aspects. Nevertheless, once the interpretation process has decided on the particular representation of the meaning of the utterance at hand, semantic interpretation will straightforwardly apply to that proposition, following the standard rules. If the thus determined proposition is interpreted as true, it can function as a premise in an act of deductive reasoning as exemplified above. However, it might turn out that the pragmatically determined meaning aspect featuring in the supposed proposition expressed by Jane for some reason or other is inappropriate and consequently has to be changed. In that case the premise (196c) assumed in the reasoning process would change also, leading to a different conclusion. However, it would not affect the nature of the reasoning employed as such. Viewed from the perspective of the nature of processes involved in the determination of the proposition expressed by a sentence and its semantic interpretation, Levinson’s view (repeated below in figure 5.4) of the semantics component as allowing the intrusion of pragmatic processes in the determination of an utterance’s meaning is sensible. Semantics lexical meaning semantic composition sub-propositional SEM 1: sentence meaning pragmatic intrusion fully propositional saturation processes GCIs Pragmatics basis for further pragmatic inferences SEM 2: what is said PCI speech acts ... + narrow context + broad context what is meant Figure 5.4: Levinson’s view However, recall Bierwisch’s suggestion that an evaluation of speech acts with respect to their truth or falsity is possible and useful as well (cf. section 4.2). Thus, consider again example (174), repeated here as (197a), where this is said by A to B and B generally thinks that A has a low opinion of B. (197) a. This is a very difficult tour. b. The Coast to Coast Cycling Path is a very difficult tour to do. Suppose further that what B takes A to have expressed (and what A did in fact intend to express) with his utterance is (197b). On the basis of B’s beliefs concerning A’s opinion of B, B might take A to have intended to convey that B won’t be able to make the relevant tour or that it is too difficult for B, etc. More specifically, B might determine on the following proposition: ‘A believes that the Coast to Coast Cycling Path is too difficult for B’. If B assumes this proposition to be true – and based on how he arrived at it, he trivially will – this will have consequences on his knowledge 246 Varieties of Meaning, Context and the Semantics/Pragmatics Distinction base of social interaction. Thus, on the one hand, it will strengthen his assumptions concerning A’s opinion of him as well as possibly lead him to take active steps to change what he believes is A’s opinion of him, etc. Be that as it may, the crucial point is that what is called semantic interpretation actually has to be assumed to take place also at the level of meaning what is meant. Note that the nature of the process that determines the truth or falsity of a given proposition is the same regardless of whether that proposition corresponds to the level of sentence meaning, utterance meaning or communicative sense/what is meant. This fact might be captured by expanding the semantics component to also include what is meant (possibly as Semantics 3 as indicated in figure 5.5), with pragmatic processes – although arguably of a different nature – also intervening or intruding between what is said and what is meant. Semantics lexical meaning semantic composition SEM 1: sentence meaning sub-propositional pragmatic intrusion fully propositional saturation processes GCIs + narrow context basis for further pragmatic inferences SEM 2: what is said pragmatic intrusion PCIs speech acts ... + broad context SEM 3: what is meant Figure 5.5: Semantics throughout However, this view raises the question whether semantics as conceived here generally should be viewed as belonging in its entirety to the language faculty. It is unlikely that anyone would want to assume that the processes needed to determine what is meant should be viewed as intruding into a linguistic semantics component. Thus, after all, RT’s conception of there being two seperate, as it were, semantics systems seems to better capture the similarities and differences between the processes assumed by Levinson to belong to one or the other of two subsystems of semantics. Thus, whereas linguistic semantics deals with the lexical meanings of linguistic expressions and their composition, it is in real semantics that propositions generally are semantically interpreted. Linguistic semantics and real semantics are related, however, in that the former is assumed to ‘inherit’ its properties from the latter. In fact, an approach is possible that denies that there is something like linguistic semantics separate from real semantics at all. Thus, Burton-Roberts (2005, 2007a) argues that linguistic semantics is in fact empty. This follows from his Representational Hypothesis (RH) (Burton-Roberts 1994, 2000, Burton-Roberts and Poole 2006, Chng 1999), which is the assumption that phonological forms only represent – rather than encode – the concepts or thoughts a speaker-hearer might want to com- The Nature of Context in Utterance Interpretation Linguistic Semantics sub-propositional Pragmatics 247 lexical meaning semantic composition sentence meaning saturation processes free enrichment ad-hoc concept form. + broad context ... Real Semantics explicature basis for further pragmatic inferences conversational implicature speech acts + broad context ... Real Semantics what is meant Figure 5.6: RT’s differentiation of linguistic vs. real semantics municate. In other words, there are no such objects, traditionally called e.g., words, morphemes, where those objects ‘consist’ of phonological and semantic material. Crucially, linguistic expressions are not considered to be sound-meaning pairings. Rather, phonological forms represent particular concepts by convention. That is, there is a set of convention rules that govern the particular range of the representation function the individual phonological forms in a language have. Note that under the RH, the view of what a language is also differs crucially from the view taken by approaches that presuppose the existence of Saussurian signs. Thus, Burton-Roberts (2007b) argues that languages are and differ from one another in their phonologies. In other words, different languages are simply different phonologies.31 In contrast, it is the conceptual structures phonological forms represent that are taken to have semantics and syntax. These conceptual structures are part of the Language of Thought. If one accepts the view of phonological forms as merely representing concepts, then it no longer makes sense to speak of ‘encoded meaning’. Note that such a view is in line with the assumption that the semantic form of an expression constitutes a pointer to the set of possible readings of that expression. As such, the semantic form is empty of meaning; its sole function is to link phonological forms and the concepts they represent. However, if that is how pointers are characterised then there actually is no reason why they should be called semantic forms at all. Rather, hearing a particular string of sounds and identifying that string as an instance/a realisation of a particular phonological form is sufficient to trigger the activation of associated concepts. However, if phonological forms, in a sense, are directly linked to and activate associated meanings, several questions arise. One question is how to explain the differences found in the interpretation of homonymous vs. polysemous expressions. Recall that in the case of homonymous expressions, frequency of occurrence of a 31 It is important to stress, however, that under the RH, phonology has to be taken to be much richer than traditionally assumed. Thus, for instance, it also includes the convention rules governing the representation of concepts by phonological forms. 248 Varieties of Meaning, Context and the Semantics/Pragmatics Distinction Phonology phonological forms point to Language of Thought concepts semantic composition ... ... what is meant Figure 5.7: Only real semantics in Burton-Roberts’ approach given meaning as well as the particular contextual circumstances influence the activation of that meaning, whereas in the case of polysemous expressions, these factors do not seem to play any role. One possibility to explain this difference in a RHframework is to assume that hearing the realisation of a particular phonological form that is used to represent concepts from distinct families of concepts actually leads to an activation of both these distinct concept families. The assumption then, I guess, would have to be that initiating and retaining the activation of two distinct concept families is effortful and cognitively taxing and therefore the interpretation process has to ‘decide’ which of the activation paths, as it were, to retain and which to inhibit, leading to the kind of competition effect noted by Klepousniotou (2002).32 In the case of phonological forms pointing to concepts that are semantically related and traditionally were associated with one expression characterised as polysemous, in contrast, only one general concept family is activated and thus, there is no competition. Another question is what happens to the notion of semantic composition and, relatedly to the idea of compositionality of meaning. Generally, assuming that linguistic semantics is in fact empty has as one consequence that the process of semantic composition can no longer be viewed as essentially linguistic. That is, if pointers only point, then there is in fact no lexical meaning. If pointers do not have logical properties, they cannot be combined by operations taking logical properties into account. Thus, semantic composition cannot be linguistic; it has to be purely conceptual. A hearer interpreting an utterance identifies the phonological forms the speaker realised and those phonological forms in turn activate the concepts they may be used to represent by convention. Pragmatic processes will be necessary to determine on the basis of the particular utterance situation exactly which concepts the speaker intended the phonological forms he used on that occasion to represent. 32 Recall that Klepousniotou (2002) found that subjects were faster in their lexical decisions for polysemous as compared to homonymous target words. She explains this result by assuming that in case of homonymy, two distinct lexical entries are initially activated and compete with one another. The Nature of Context in Utterance Interpretation 249 The concepts themselves do have logical properties and those properties will determine how they are combined, finally terminating in a full proposition. Thus, under such a conception, the main task of pragmatic processes is, in a sense, disambiguation. That is, from the phonological forms the speaker used, the convention rules governing their use as well as the situation in which he used them, the hearer has to infer which concepts the speaker intended those phonological forms to represent. To be sure, in those cases where a phonological form is used to represent a concept where this representation is not governed by a convention rule, pragmatic processes are involved also in either determining which concept is in fact represented or in creating such a concept. Note that – as far as I am aware – at present, there exists no clear exposition of how exactly the interpretation of utterances works under the RH.33 As a consequence, there are quite a number of unresolved issues concerning utterance interpretation in this framework. For instance, it is not quite clear to me at which stage during interpretation considerations of speaker intentions come into play and, thus, whether a differentiation is made between the concepts phonological forms are taken to represent in a given utterance situation and the concepts the speaker is taken to have intended to communicate in that situation. In other words, it is not clear whether a differentiation of what is said/utterance meaning and what is meant/communicative sense is assumed, as argued for in chapter 4. Another worry is whether the assumption behind the RH that phonological forms represent concepts by convention can capture the fact that convention actually is a gradual property, as argued in chapter 2. Finally, it seems that the empirical results supporting the view of underspecification of linguistic meaning reported in chapter 2 are not compatible with the RH hypothesis. Generally, looking at different approaches to the differentiation between semantics and pragmatics, it seems that it is not so much a question of whether context plays a role and if so which for that differentiation. Rather, it is the difference in the nature of the processes employed that is to be captured by the differentiation. However, since, at the same time, the differentiation of semantics from pragmatics goes along with assumptions concerning the relations of these systems to the language faculty in general, the problem arises of how much of the semantic or pragmatic processes to allow to figure in the language faculty. In any case, it actually seems that there is no way of differentiating between two systems involved in the interpretation of utterances, if these two systems have to be assumed to strictly follow one another. Rather, the different types of processes employed seem to interleave. Thus, pending a clearer exposition of the RH position, I will assume the following version of the semantics/pragmatics distinction. Taking into account the principle of economy concerning the size and nature of a semantics component within the language faculty, I will assume, following Bierwisch, Dölling and RT, that it provides underspecified semantic forms as the lexical meaning of expressions as well as the rules for their combination, following the principle of compositionality. Semantic interpretation, then, does not take place within the linguistic semantic component, but rather in what RT calls real semantics (Bierwisch’s conceptual system). However, in the spirit of Dölling’s multi-level model of meaning, there are various levels 33 Thus, the only fragmentary graphic illustration of that approach in figure 5.7. 250 Varieties of Meaning, Context and the Semantics/Pragmatics Distinction Linguistic Semantics lexical meaning semantic composition sub-propositional Real Semantics sentence meaning saturation processes free enrichment ... interprets Pragmatics + narrow context utterance meaning conversational implicature speech acts + broad context ... interprets communicative sense Figure 5.8: My view on the semantics/pragmatics distinction of meaning at which semantic interpretation might take place. And these levels of meaning are attained by pragmatic processes enriching the sub-propositional semantic form determined by the linguistic semantic component in a number of steps. For reasons given above, I assume that a differentiation between something like the level of utterance meaning and the level of communicative sense is useful and neccessary. As for the question of which criterion should be assumed to differentiate between the processes involved in determining the utterance meaning of an utterance and those determining the communicative sense of the speaker making that utterance: I am inclined to follow Bierwisch’s suggestion that it is the consideration of the speaker’s intentions in making the utterance she did. Thus, even if assuming that in addition to the traditionally assumed processes something like free enrichment needs to be assumed to determine the actual truth conditions of an utterance, in section 5.2.1 I suggested a way of modelling that process such that it need not make reference to potential speaker’s intentions in making the utterance she did. Let me adapt a metaphor used by Levinson (2000) in his characterisation of Discourse Representation Theory, were he writes the following. . . . there is a common slate, a level of propositional representation, upon which both semantics and pragmatics can write – the contributions may be distinguished, let’s suppose, by the color of the ink: semantics in black, pragmatics in red. Semantics and pragmatics remain modular “pens” as it were: they are separate devices making distinctively different contributions to a common level of representation. (Levinson 2000, p. 193) When Levinson writes semantics, he of course refers to his conception of the linguistic semantics component, consisting of two subsystems. However, I think the metaphor works equally well for a conception of semantics as actually two separate, though related systems that interact with a pragmatics system. Thus, it is the contributions of linguistic semantics and pragmatics that are interpreted by real semantics at various steps during utterance interpretation. While the linguistic semantic Summary 251 contribution is stable, the pragmatic contributions might change, since they are defeasible. Moreover, different pragmatic processes may be seen as adding successively more material, while at particular points during that process the representation is semantically interpreted by real semantics. As concerns the objection that already for the resolving of reference of (true) demonstratives appeal to speaker intentions might be necessary: I do not have an explicit treatment to offer, however, there are two points that I would like to make. First, it seems to me that a speaker is restricted in his use of a true demonstrative to making reference to entities that the hearer can objectively identify in or accommodate to the discourse context, precisely because she cannot rely on the hearer to make use of reasoning about the speaker’s intentions in making an utterance in fixing those referents. Thus, in a situation were A and B are at the zoo, have been looking at the apes and have been talking about what funny things apes are prone to do, A’s utterance of Look at that, accompanied by a pointing gesture towards a particular ape will probably be interpreted by B as meaning ‘Look at what that ape I’m pointing at is doing’. However, if A and B have been talking about some completely different subject and A suddenly stops, points towards some ape and says Look at that, it will not be so clear what he refers to and I would suppose that B will fix on a referent that he assumes is mutually accessible to A and B (e.g., the ape itself). Second, even if in resolving the reference of true demonstratives the hearer employs reasoning as to what it most likely is that the speaker intended to refer to, I would content that this kind of reasoning still is different from reasoning about the speaker’s intention in making the utterance she did as such. Thus, while in the latter case, the hearer will employ more or less specific assumptions drawn from his knowledge base of social interaction, as, for instance, assumptions concerning the particular social relationship between speaker and hearer, I do not think that this type of information is used when it comes to determining the referent of a true demonstrative. Thus, simply making reference to speaker intentions in the differentiation between the pragmatic processes leading to utterance meaning and those leading to what is meant actually might not be fine-grained enough. Whereas in the determination of the former, reasoning concerning what the speaker intended to refer to by his use of a demonstrative is involved, for the determination of the latter, reasoning concerning why the speaker said what he did is necessary and only the latter kind of reasoning requires access to knowledge concerning social interaction. 5.3 Summary In this chapter, we looked at alternative characterisations of literal meaning and non-literal meaning and (broadly) at the role context might be taken to play in the overall interpretation process as well as in the differentiation of semantics and pragmatics. Let me summarise the points that were made. Recanati offers a typology of meanings that is intended to differentiate between what are traditionally taken to be literal meaning and non-literal meaning as well as further meaning aspects that can be identified but may not be taken to fit one or the other description. Thus, he identifies an expression‘s t-literal meaning as the 252 Varieties of Meaning, Context and the Semantics/Pragmatics Distinction meaning conventionally associated with it and as such the meaning traditionally called literal. He differentiates from t-literal meaning a number of more or less deviating meanings, only some of which are non-literal in the ordinary understanding of the term. As we saw, there are a number of problems with Recanati’s classification of the different types of meaning. Thus, his characterisation of t-literal meaning as both being conventional as well as literal in the traditional sense is debatable, since, as we saw, conventionality does not necessarily have to be seen as going along with literalness of meaning. Also, according to his characterisation of nonliteral meaning, conversational implicatures and indirect speech acts should be best exemplars of that category, as they are both secondary and – according to Recanati – fulfil the transparency condition. However, intuitively it is such figurative meanings as metaphor and metonymy that are regarded good examples of non-literal meaning. Moreover, it seems that a notion such as the transparency condition, which relies on speaker-hearers’ intuitions does not produce the desired results. Ariel’s differentiation of three types of what she calls minimal meanings is guided by the assumption that the notion of literal meaning has been used to refer to these three distinct types of meaning because the notion captures the fact that all three meanings are minimal or basic. However, she points out, they are only minimal or basic with respect to the different domains they belong to. Thus, she differentiates linguistic meaning both from salient meaning and privileged interactional meaning. The main problem with Ariel’s classification concerns her characterisation of linguistic meaning and salient meaning as being totally independent of one another. In fact, from how she characterises the two meanings, it seems that both are taken to be independently accessed during the interpretation of a particular expression. Reasons were given for why this seems highly implausible. In contrast to both Recanati and Ariel who assume of lexical meanings that they are full-fledged readings, I argued for viewing lexical meaning as underspecified. As a consequence – and if one wants to keep the intuition that the notion applies to full readings – literal meaning can no longer be regarded as semantic. Thus, literal meaning, like non-literal meaning, has to be seen as pragmatically determined. To differentiate the one type of meaning from the other, I characterised the former as basic and non-derived and the latter as non-basic, derived or deviating and in some sense replacing the meaning from which it derives or deviates. With such a characterisation, metaphor and metonymy are captured as prototypical examples of non-literal meaning, whereas, contrary to Recanati, neither PCIs nor indirect speech acts are considered non-literal. In fact, with this characterisation, the results of ad-hoc concept formation would be considered non-literal, but not the results of free enrichment. This is because whereas in the former case a basic meaning is changed and replaced, in the latter case a meaning is specified, as it were, and aspects are added to it. Similar points have to be made with respect to idioms and irony. Thus, while the former may be characterised as non-literal, this is not so clear for the latter. Note again that this characterisation of literal meaning and non-literal meaning does not distinguish between types of meanings that play distinctive roles in the interpretation process and it does not allow the two notions to be used in differentiating semantics from pragmatics. However, this seems to be simply due to the Summary 253 fact that these notions are rather intuitive and vague to begin with. Another notion traditionally used in the differentiation between semantics and pragmatics is that of context-(in)dependence. In addition, particular types of pragmatic processes are differentiated with respect to the nature of the contextual information they take into account and more specifically, whether they make reference to the speaker’s intention in making a particular utterance. For instance, the process free enrichment is usually assumed to take recourse to potential speaker intentions. However, there are experimental results which indicate that such implicit meaning aspects (IMAs) as taken to be contributed by free enrichment actually seem to arise whether the speaker intended them or not. This effect follows if one assumes that during the interpretation of an utterance, information stored in frames evoked by the expressions used may be made use of. This allows for a contribution of IMAs to the interpretation of an utterance without the need to take recourse to speaker intentions. Moreover, although the process of free enrichment is taken to operate independently of linguistically mandated slots in the semantic form of an utterance, the process is not entirely free. Thus, it does not add any meaning aspects whatsoever, but rather its contribution may be taken to be restricted to meaning aspects that are the values of variables contributed by frames. As concerns the role of the notion of context-(in)dependence in the differentiation of semantics and pragmatics, I argued that, actually, the dichotomy contextindependent vs. context-dependent may only be used to differentiate what might be called linguistic semantics from pragmatics. In contrast, the process of semantic interpretation actually only applies to meaning representations that have already been pragmatically enriched, as the output of the context-independent linguistic semantics component is only sub-propositional. Thus, the notion of context-(in)dependence cannot be used to differentiate between the meaning aspects pragmatics as compared to real semantics deals with. Rather, these two systems are differentiated by the nature of the processes that constitute them: monotonic reasoning with non-defeasible output in the case of real semantics vs. non-monotonic reasoning with defeasible output in the case of pragmatics. Turning, lastly, to the differentiation of types of pragmatic processes, we saw throughout this thesis that amongst others, one important aspect in which pragmatic processes are assumed to differ is whether they take into account the speaker’s potential intentions in making her utterance. For the processes that contribute to the level of utterance meaning, I argued that the information they use is restricted to narrow context34 and conceptual knowledge including common sense knowledge, possibly in the form of frames evoked by the different expressions used, but also in the form of meaning postulates which capture stereotypical behaviour and practices. In contrast, the processes leading to what is meant by the speaker of a particular utterance are assumed to make use of information resulting from reasoning about the speaker’s potential intentions and generally information from the knowledge base of social interaction. 34 possibly with the exception of demonstratives, whose resolution may require reasoning about what the speaker intended to demonstrate 254 Varieties of Meaning, Context and the Semantics/Pragmatics Distinction Chapter 6 Summary 256 Summary List of Figures 2.1 2.2 2.3 2.4 2.5 2.6 2.7 2.8 2.9 2.10 2.11 2.12 2.13 2.14 2.15 2.16 2.17 2.18 2.19 2.20 2.21 2.22 2.23 2.24 2.25 2.26 3.1 3.2 3.3 3.4 3.5 3.6 3.7 3.8 Traditional differentiation of types of meaning . . . . . . . . . . . . . Grice’s distinction of what is said and what is meant . . . . . . . . . . The typical cat-on-the-mat configuration . . . . . . . . . . . . . . . . A rather unusual cat-on-the-mat configuration . . . . . . . . . . . . . Revised differentiation of types of meaning . . . . . . . . . . . . . . . (Non-)Literal meaning and Grice’s levels of what is said/what is meant Busse’s view of degrees of conventionality . . . . . . . . . . . . . . . . Idioms in Busse’s framework . . . . . . . . . . . . . . . . . . . . . . . Salient meanings according to Busse’s characterisation . . . . . . . . Giora’s characterisation within Busse’s framework . . . . . . . . . . . Lexical entry for spülen according to Engelberg (2000) . . . . . . . . Lexical entry for to bake according to Pustejovsky (1995) . . . . . . . Lexical entry for cake according to Pustejovsky (1995) . . . . . . . . . Semantic representation for the VP bake a cake according to Pustejovsky (1995) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Literal and non-literal meaning in Relevance Theory . . . . . . . . . . Lexical entry according to Giora . . . . . . . . . . . . . . . . . . . . . Salient and non-salient meaning in Giora’s approach . . . . . . . . . . A lexical entry adapted from Relevance Theory . . . . . . . . . . . . Partial syntactic structure for Mittie feeds the dog . . . . . . . . . . . e.g. the destruction of the city, the city’s destruction . . . . . . . . . e.g. (John)destroyed the city . . . . . . . . . . . . . . . . . . . . . . . e.g. (John’s) destroying the city . . . . . . . . . . . . . . . . . . . . . The semantic form of Max began to read every novel. . . . . . . . . . The semantic form of Max began every novel. . . . . . . . . . . . . . . Reaction times from Beretta et al. (2005) . . . . . . . . . . . . . . . . M350 latencies Beretta et al. (2005) . . . . . . . . . . . . . . . . . . . Grice’s distinction of what is said and what is meant more detailed . Grice’s types of meaning and their relation to what is said/what is meant . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Bierwisch’s three levels of meaning . . . . . . . . . . . . . . . . . . Minimal proposition vs. proposition expressed . . . . . . . . . . . . The Relevance-theoretic View . . . . . . . . . . . . . . . . . . . . . Bach’s view . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Stanley’s view . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Borg’s view . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 13 15 16 19 20 30 32 33 34 42 45 46 47 57 58 59 60 63 64 64 65 65 65 69 69 . 85 . . . . . . . 87 89 99 100 104 106 121 258 List of Figures 3.9 Cappelen and Lepore’s view . . . . . . . . . . . . . . . . . . . . . . . 122 4.1 4.2 4.3 Levinson’s view . . Recanati’s view . . Bierwisch’s view on nicative sense . . . 5.1 5.2 5.3 5.4 5.5 5.6 5.7 5.8 Non-literal meaning in Recanati’s classification . . The Killing frame . . . . . . . . . . . . . . . . . The Ingestion frame . . . . . . . . . . . . . . . . Levinson’s view . . . . . . . . . . . . . . . . . . . Semantics throughout . . . . . . . . . . . . . . . . RT’s differentiation of linguistic vs. real semantics Only real semantics in Burton-Roberts’ approach My view on the semantics/pragmatics distinction . . . . . . . . . . . . . . the relation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . of utterance meaning and . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167 . . . . . . 191 commu. . . . . . 195 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209 233 237 245 246 247 248 250 Bibliography Ariel, M.: 2002, The demise of a unique concept of literal meaning., Journal of Pragmatics 34, 361–402. Asher, N. and Lascarides, A.: 2003, Logics of Conversation., Cambridge: Cambridge University Press. Assimakopoulos, S.: 2008, Intention, common ground and the availability of semantic content: a relevance-theoretic perspective., in I. Kecskes and J. Mey (eds), Intention, common ground and the egocentric speaker-hearer., Mouton de Gruyter. Bach, K.: 1994a, Conversational Impliciture., Mind and Language 9, 124–162. Bach, K.: 1994b, Semantic slack: what is said and more., in S. L. Tsohatzidis (ed.), Foundations of Speech Act Theory. Philosophical and linguistic perspectives., Routledge, pp. 267–291. Bach, K.: 1997, The Semantics-Pragmatics Distinction: What It Is and Why It Matters., Linguistische Berichte, Special Issue on Pragmatics 8, 33–50. Bach, K.: 2000, Quantification, Qualification and Context. A Reply to Stanley and Szabó., Mind and Language 15, 262–283. Bach, K.: 2001a, Speaking Loosely: Sentence Nonliterality., Midwest Studies in Philosophy 25, 249–263. Figurative Language. Bach, K.: 2001b, You don’t say?, Synthese 128, 15–44. Bach, K.: 2004, Minding the gap., in C. Bianchi (ed.), The Semantics/Pragmatics Distinction., Stanford: CSLI. Bach, K.: 2005, Context ex Machina., in G. S. Zoltán (ed.), Semantics vs. Pragmatics., Oxford: Clarendon, pp. 15–44. Bach, K.: 2006a, Impliciture vs. Explicature: What’s the difference?, userwww.sfsu.edu/~kbach\Bach.ImplExpl.pdf. Paper given at the ‘Workshop on Explicit Communication’, University of Granada, Spain. Bach, K.: 2006b, The Excluded Middle: Semantic Minimalism without Minimal Propositions., Philosophy and Phenomenological Research 73, 435–442. Bach, K.: n.d.a, From the strange to the bizarre: Another Reply to Cappelen and Lepore., userwww.sfsu.edu/~kbach\reply2toC&L.pdf. 260 Bibliography Bach, K.: n.d.b, Minimalism for Dummies: A Reply to Cappelen and Lepore., userwww.sfsu.edu/~kbach\replytoC&L.pdf. Baker, C., Fillmore, C. and Cronin, B.: 2003, The structure of the FrameNet database., International Journal of Lexicography 16, 281–296. Baker, C., Fillmore, C. and Lowe, J.: 1998, The Berkeley FrameNet project., Proceedings of the 36th ACL and 17th COLING, San Francisco, pp. 86–90. Baron-Cohen, S., Leslie, A. M. and Frith, U.: 1985, Does the autistic child have a ”theory of mind”?, Cognition 21, 37–46. Beretta, A., Fiorentino, R. and Poeppel, D.: 2005, The effects of homonymy and polysemy on lexical access: an MEG study., Cognitive Brain Research 24, 57–65. Bezuidenhout, A. L.: 2006, The coherence of contextualism., Mind and Language 21, 1–10. Bezuidenhout, A. L. and Cutting, J. C.: 2002, Literal meaning, minimal propositions, and pragmatic processing., Journal of Pragmatics 34, 433–456. Bezuidenhout, A. L. and Morris, R. K.: 2004, Implicature, Relevance, and Default Pragmatic Inference., in D. Sperber and I. Noveck (eds), Experimental Pragmatics., Palgrave Macmillan, Basingstoke, pp. 257–282. Bianchi, C. (ed.): 2004, The Semantics/Pragmatics Distinction., CSLI Publications. Bierwisch, M.: 1979, Wörtliche Bedeutung - eine pragmatische Gretchenfrage., in G. Grewendorf (ed.), Sprechakttheorie und Semantik., Frankfurt am Main, Suhrkamp Verlag, pp. 119–148. Bierwisch, M.: 1980, Semantic Structure and Illocutionary Force., Speech act theory and pragmatics., D. Reidel, Dordrecht. Bierwisch, M.: 1983, Semantische und konzeptuelle Repräsentation lexikalischer Einheiten., in R. Růžička and W. Motsch (eds), Untersuchungen zur Semantik., Akademie Verlag, Berlin, pp. 61–99. Bierwisch, M.: 1997, Lexical Information from a Minimalist Point of View., in C. Wilder, H. M. Gerlach and M. Bierwisch (eds), The Role of Economy Principles in Linguistic Theory., Akademie Verlag, Berlin. Bierwisch, M. and Schreuder, R.: 1992, From concepts to lexical items., Cognition 42, 23–60. Blasko, D. G. and Connine, C. M.: 1993, Effects of familiarity and aptness in metaphor processing., Journal of Experimental Psychology: Learning, Memory and Cognition 19, 295–308. Borer, H.: 2004, Structuring Sense., Vol. 1 and 2, Oxford: Oxford University Press. Bibliography 261 Borg, E.: 2001, An Expedition Abroad: Metaphor, Thought and Reporting., in P. French and H. Wettstein (eds), Midwest Studies in Philosophy XXV, Oxford: Blackwell, pp. 227–248. Borg, E.: 2004a, Formal semantics and intentional states., Analysis 64, 215–23. Borg, E.: 2004b, Minimal Semantics., Oxford: Oxford University Press. Borg, E.: 2006, Review of Literal Meaning., Mind 115, 461–5. Borg, E.: 2007, Minimalism vs. Contextualism in Semantics., in G. Preyer and G. Peter (eds), Context Sensitivity and Semantic Minimalism: Essays on Semantics and Pragmatics., Oxford University Press, pp. 546–571. Borg, E.: 2009, On three theories of implicature: Default Theory, Relevance Theory and Minimalism., International Review of Pragmatics 1, 63–83. Braun, D.: 2010, Indexicals., in E. N. Zalta (ed.), The Stanford Encyclopedia of Philosophy., summer 2010 edn, Stanford University. Breheny, R., Katsos, N. and Williams, J.: 2006, Are generalised scalar implicatures generated by default? an on-line investigation into the role of context in generating pragmatic inferences., Cognition 100. Brennan, J. and Pylkkänen, L.: 2008, Processing events: Behavioral and neuromagnetic correlates of Aspectual Coercion, Brain and Language 106. Burton-Roberts, N.: 1994, Ambiguity, sentence and utterance., Transactions of the Philological Society 92, 179–213. Burton-Roberts, N.: 2000, Where and what is phonology? A representational perspective., in N. Burton-Roberts, P. Carr and G. Docherty (eds), Phonological knowledge., Oxford University Press, Oxford, pp. 39–66. Burton-Roberts, N.: 2005, Robyn Carston on semantics, pragmatics and ‘encoding’., Journal of Linguistics 41, 389–407. Burton-Roberts, N.: 2006, Cancellation and Intention. Paper given at the ‘Workshop on Explicit Communication’, University of Granada, Spain. Burton-Roberts, N.: 2007a, Varieties of semantics and encoding: negation, narrowing/loosening and numerals., in N. Burton-Roberts (ed.), Pragmatics., Palgrave, Basingstoke. Burton-Roberts, N. (ed.): 2007b, Pragmatics., Palgrave, Basingstoke. Burton-Roberts, N. and Poole, G.: 2006, ‘Virtual conceptual necessity’, featuredissociation and the Saussurian legacy in generative grammar., Journal of Linguistics 42, 575–628. Busse, D.: 1991, Diachrone Semantik und Pragmatik. Untersuchungen zur Erklärung und Beschreibung des Sprachwandels., Tübingen: Niemeyer. 262 Bibliography Cacciari, C. and Tabossi, P.: 1988, The Comprehension of Idioms., Journal of Memory and Language 27, 668–683. Capone, A.: 2004, Speech acts: http://semanticsarchive.net. literal and non-literal. internet resource: Capone, A.: 2006, On Grice’s Circle (a theory-internal problem in linguistic theories of the Gricean type.), Journal of Pragmatics 38, 645–669. Capone, A.: 2009, Are explicatures cancellable? Toward a theory of the speaker’s intentionality., Intercultural Pragmatics 6–1, 55–83. Cappelen, H. and Lepore, E.: 2005, Insensitive Semantics: a defense of semantic minimalism and speech act pluralism., Blackwell. Cappelen, H. and Lepore, E.: n.d.a, Kent Bach on Minimalism for Dummies, userwww.folk.uio.no/hermanc/docs/KBM.pdf. Cappelen, H. and Lepore, E.: n.d.b, userwww.folk.uio.no/hermanc/docs/BachPPR.pdf. Reply to Bach, Carston, R.: 1997, Enrichment and loosening: complementary processes in deriving the proposition expressed?, Linguistische Berichte, Special Issue on Pragmatics 8, 103–27. Carston, R.: 1999, The semantics/pragmatics distinction: A view from Relevance Theory., in K. Turner (ed.), The Semantics/Pragmatics Interface from Different Points of View., Current research in the semantics - pragmatics interface 1, Elsevier Science, Oxford. Carston, R.: 2002a, Linguistic meaning, communicated meaning and cognitive pragmatics., Mind and Language 17(1/2), 127–48. Special Issue on Pragmatics and Cognitive Science. Carston, R.: 2002b, Metaphor, ad hoc concepts and word meaning - more questions than answers., UCL Working Papers in Linguistics 14, University College London. Carston, R.: 2002c, Thoughts and Utterances. The Pragmatics of Explicit Communication., Blackwell Publishing, Oxford. Carston, R.: 2004a, Explicature and semantics., in S. Davis and B. S. Gillon (eds), Semantics: A Reader., Oxford University Press, pp. 817–845. Carston, R.: 2004b, Relevance Theory and the saying/implicating distinction., in L. Horn and G. Ward (eds), Handbook of Pragmatics., Oxford: Blackwell. Carston, R.: 2004c, Truth-conditional content and conversational implicature., in C. Bianchi (ed.), The Semantics/Pragmatics Distinction., CSLI Publications. Carston, R.: 2005, Relevance Theory, Grice and the neo-Griceans: a response to Laurence Horn’s ‘Current issues in neo-Gricean pragmatics’., Intercultural Pragmatics 2(3), 303–319. Bibliography 263 Carston, R.: 2007, How many pragmatic systems are there?, in M.-J. Frapolli (ed.), Saying, Meaning, Referring: Essays on the Philosophy of François Recanati., Palgrave, pp. 18–48. Chierchia, G.: 2004, Scalar Implicatures, Polarity Phenomena, and the Syntax/Pragmatics Interace., in A. Belletti (ed.), Structures and Beyond., Oxford University Press, pp. 39–103. Chierchia, G., Fox, D. and Spector, B.: 2010, The Grammatical View of Scalar Implicatures and the Relationship between Semantics and Pragmatics., www. semanticsarchive.net/Archive/WMzY2ZmY/CFS_EmbeddedSIs.pdf. Chierchia, G. and McConnell-Ginet, S.: 2000, Meaning and grammar: an introduction to semantics, 2. edn, Cambridge, Mass.: MIT. Chng, S. W.: 1999, Language, thought and literal meaning., Phd thesis, University of Newcastle upon Tyne. Clark, H. H.: 1973, Space, time, semantics and the child., in T. E. Moore (ed.), Cognitive development and the acquisition of language., New York: Academic Press, pp. 27–63. Cohen, A.: 2006, Anaphora resolution and minimal models., in J. Bos and A. Koller (eds), ICoS-5, Buxton, England. Cole, P. (ed.): 1981, Radical pragmatics, Academic, New York. Colston, H. L.: 2000, On necessary conditions for verbal irony comprehension., Pragmatics and Cognition 8, 277–324. Colston, H. L.: 2002, Contrast and assimilation in verbal irony., Journal of Pragmatics 34, 111–142. Colston, H. L. and Gibbs, R. W.: 2002, Are Irony and Metaphor Understood Differently?, Metaphor and Symbol 17, 57–80. Curcó, C.: 2000, Irony: Negation, echo and metarepresentation., Lingua 110, 257– 280. Dascal, M.: 1987, Defending literal meaning., Cognitive Science 11, 259–281. Dascal, M.: 1989, On the roles of context and literal meaning in understanding., Cognitive Science 13, 253–257. de Villiers, J., Myers, B. and Stainton, R.: 2010, Differential Pragmatic Abilities and Autism Spectrum Disorders: The Case of Pragmatic Determinants of Literal Content., in M. Macaulay and P. Garces-Blitvich (eds), Pragmatics and Context., Toronto: Antares. de Villiers, J., Stainton, R. and Szatmari, P.: 2007, Pragmatic abilities in autism spectrum disorder: a case study in philosophy and the empirical., Midwest Studies in Philosophy 31, 292–317. 264 Bibliography Dennett, D. C.: 1969, Content and Consciousness., Routledge and Kegan Paul, London. Dews, S. and Winner, E.: 1999, Obligatory processing of literal and nonliteral meanings in verbal irony., Journal of Pragmatics 31, 1579–1599. Dölling, J.: 1997, Semantic Form and Abductive Fixation of Parameters., in J. Dölling (ed.), Systematische Bedeutungsvariationen: Semantische Form und kontextuelle Interpretation, University of Leipzig. Dölling, J.: 2000, Formale Analyse von Metonymie und Metapher., in R. Eckhardt and K. von Heusinger (eds), Meaning Change – Meaning Variation., University of Konstanz, pp. 31–53. Dölling, J.: 2001, Systematische Bedeutungsvariationen: Semantische Form und kontextuelle Interpretation., Linguistische Arbeitsberichte 78, University of Leipzig, Germany. Dölling, J.: 2005, Semantische Form und pragmatische Anreicherung: Situationsausdrücke in der Äußerungsinterpretation., Zeitschrift für Sprachwissenschaft 24, 159–225. Dölling, J. and Zybatow, T.: 2007, Verb Meaning: How much Semantics is in the Lexicon?, in A. Späth (ed.), Interfaces and Interface Conditions., Mouton – de Gruyter, pp. 33–75. Egg, M.: 1994, Zur Repräsentation extrem polysemer Lexeme., in M. Schwarz (ed.), Kognitive Semantik., Narr, Tübingen. Egg, M.: 2002, Semantic construction for reinterpretation phenomena, Linguistics 40, 579–609. Egg, M.: 2003, Beginning Novels and Finishing Hamburgers: Remarks on the Semantics of to begin, Journal of Semantics 20, 163–191. Egg, M.: 2005, Flexible Semantics for Reinterpretation Phenomena., CSLI, Stanford. Eviatar, Z. and Just, M. A.: 2006, Brain correlates of discourse processing: An fMRI investigation of irony and conventional metaphor comprehension., Neuropsychologia 44, 2348–2359. Fillmore, C., Johnson, C. and Petruck, M.: 2003, Background to FrameNet., International Journal of Lexicography 16, 235–250. Fodor, J. A.: 1983, Modularity of Mind: An Essay on Faculty Psychology., Cambridge, Mass.: MIT Press. Fodor, J. A.: 1998, Concepts. Where cognitive science went wrong., Oxford: Clarendon. Bibliography 265 Fodor, J. A. and Lepore, E.: 1998, The Emptiness of the Lexicon: Reflections on James Pustejovsky’s The Generative Lexicon., Linguistic Inquiry 29, 269–288. Frisson, S.: 2009, Semantic Underspecification in Language Processing., Language and Linguistics Compass 3(1), 111–27. Frisson, S. and Pickering, M. J.: 1999, The Processing of Metonymy: Evidence from Eye Movements., Journal of Experimental Psychology 25(6), 1366–1383. Frisson, S. and Pickering, M. J.: 2007, The processing of familiar and novel senses of a word: why reading Dickens is easy but reading Needham can be hard., Language and Cognitive Processes 22, 595–613. Garcia-Carpintero, M.: 2006, Recanati on the Semantics/Pragmatics distinction., Critica 38(112), 35–68. Garrett, M. and Harnish, R. M.: 2007, Experimental Pragmatics: Testing for impliciture., Pragmatics and Cognition 16(1), 65–90. Garrett, M. and Harnish, R. M.: 2008, Skating along the syntactic verge., in W. D. Lewis, S. Karimi, H. Harley and S. O. Farrar (eds), Time and Again. Theoretical Perspectives on Formal Linguistics., Benjamins, Amsterdam. Garrett, M. and Harnish, R. M.: 2009, Q-Phenomena, I-Phenomena and Impliciture: Some Experimental Pragmatics., International Review of Pragmatics 1, 84–117. Gazdar, G.: 1979, Pragmatics: implicature, presupposition and logical form., Academic Press. Gernsbacher, M. A., Keysar, B., Robertson, R. R. W. and Werner, N. K.: 2001, The Role of Suppression and Enhancement in Understanding Metaphors., Journal of Memory and Language 45, 433–450. Gernsbacher, M. A. and Robertson, R. R. W.: 1999, The role of suppression in figurative language comprehension., Journal of Pragmatics 31, 1619–1630. Gibbs, R. W.: 1994, The poetics of mind: Figurative thought, language, and understanding., Cambridge University Press, Cambridge. Gibbs, R. W.: 2002, A new look at literal meaning in understanding what is said and implicated., Journal of Pragmatics 34, 457–486. Gibbs, R. W. and Moise, J. F.: 1997, Pragmatics in understanding what is said., Cognition 62, 51–74. Giora, R.: 1995, On irony and negation., Discourse Processes 19, 239–264. Giora, R.: 1997, Understanding figurative and literal language: The graded salience hypothesis., Cognitive Linguistics 8, 183–206. Giora, R.: 1999, On the priority of salient meanings: Studies of literal and figurative language., Journal of Pragmatics 31, 919–929. 266 Bibliography Giora, R.: 2002, Literal vs. figurative language: Different or equal?, Journal of Pragmatics 34, 487–506. Giora, R. and Fein, O.: 1999, On understanding familiar and less-familiar figurative language., Journal of Pragmatics 31, 1601–1618. Giora, R., Fein, O., Kaufman, R., Eisenberg, D. and Erez, S.: 2009, Does an ”ironic situation” favour an ironic interpretation?, in G. Brone and J. Vandaele (eds), Cognitive poetics. Goals, gains and gaps., Berlin/New York: Mouton de Gruyter, pp. 383–399. Giora, R., Fein, O., Laadan, D., Wolfson, J., Zeituny, M., Kidron, R., Kaufman, R. and Shaham, R.: 2007, Expecting Irony: Context Versus Salience-Based Effects., Metaphor and Symbol 22, 119–146. Giora, R., Fein, O. and Schwartz, T.: 1997, On the time course of understanding metaphor: The graded salience hypothesis., In preparation . Glucksberg, S. and McGlone, M. S.: 1999, When love is not a journey: What metaphors mean., Journal of Pragmatics 31, 1541–1558. Glucksberg, S., P., G. and Bookin, H.: 1982, On understanding non-literal speech: Can people ignore metaphors?, Journal of Verbal Learning and Verbal Behavior 1, 85–96. Gold, R., Faust, M. and Goldstein, A.: 2010, Semantic integration during metaphor comprehension in asperger syndrome., Brain and Language . doi:10.1016/j.bandl.2010.03.002. Grice, H. P.: 1969, Utterer’s Meaning and Intentions., Philosophical Review pp. 147– 177. Reprinted in Grice 1989, pp. 86–116. Grice, H. P.: 1975, Logic and Conversation., in P. Cole and J. L. Morgan (eds), Syntax and Semantics 3: Speech Acts, Academic Press, New York. Reprinted in Grice 1989, pp. 22–40. Grice, H. P.: 1989, Studies in the ways of words., Harvard University Press, Cambridge and London. Groefsema, M.: 2007, Concepts and word meaning in relevance theory., in N. Burton-Roberts (ed.), Pragmatics., Palgrave, Basingstoke, pp. 136–157. Gross, S.: 2006, Can One Sincerely Say What One Doesn’t Believe?, Mind and Language 21, 11–20. Happé, F. G. E.: 1993, Communicative compentence and theory of mind in autism: A test of relevance theory., Cognition 48, 101–119. Hintzman, D. L.: 1986, ”Schema Abstraction” in a Multiple-Trace Memory Model., Psychological Review 93, 411–428. Bibliography 267 Hirst, W., LeDoux, J. and Stein, S.: 1984, Constraints on the Processing of Indirect Speech Acts: Evidence from Aphasiology., Brain and Language 23, 26–33. Hobbs, J. R., Stickel, M., Appelt, D. and Martin, P.: 1993, Interpretation as Abduction., Artificial Intelligence 63, 69–142. Huang, Y.: 2007, Pragmatics, Oxford: Oxford University Press. Huang, Y. T. and Snedeker, J.: 2009, Online interpretation of scalar quantifiers: Insight into the semantic-pragmatic interface., Cognitive Psychology 58, 376–415. Huang, Y. T. and Snedeker, J.: in press, Logic and Conversation revisited: Evidence for a division between semantic and pragmatic content in real time language comprehension., Language and Cognitive Processes . Irmer, M.: 2009, Bridging Inferences in Discourse Interpretation. Ms., doctoral thesis. Israel, M.: 2002, Literally Speaking., Journal of Pragmatics 34. Kadmon, N.: 2001, Formal Semantics. Semantics, pragmatics, presupposition, and focus., Blackwell. Kaplan, D.: 1989a, Afterthoughts., in J. Almog, J. Perry and H. Wettstein (eds), Themes from Kaplan, Oxford. Kaplan, D.: 1989b, Demonstratives., in J. Almog, J. Perry and H. Wettstein (eds), Themes from Kaplan, Oxford. Katz, J.: 1977, Propositional Structure and Illocutionary Force., New York: Crowell. Keysar, B.: 1989, On the Functional Equivalence of Literal and Metaphorical Interpretations in Discourse., Journal of Memory and Language 28, 375–385. Klepousniotou, E.: 2002, The Processing of Lexical Ambiguity: Homonymy and Polysemy in the Mental Lexicon., Brain and Language 81, 205–223. König, E. u. E. C. T.: 1988, Pragmatic Strengthening and Semantic Change: The Conventionalizing of Conversational Implicature., in Hüllen and Schulze (eds), Understanding the Lexicon., Tübingen: Niemeyer, pp. 110–124. Korta, K.: 1997, Implicitures: Cancelability and Non-detachability., Technical Report ILCLI-97-LIC-6, Donostia: ILCLI. Korta, K. and Perry, J.: 2008, Pragmatics., in E. N. Zalta (ed.), The Stanford Encyclopedia of Philosophy., fall 2008 edn, Stanford University. Kratzer, A.: 1996, Severing the external argument from its verb., in J. Rooryck and L. Zaring (eds), Phrase Structure and the Lexicon., Dordrecht: Kluwer, pp. 109– 137. 268 Bibliography Lakoff, G. (ed.): 1987, Women, Fire, and Dangerous Things: What Categories Reveal About the Mind., University of Chicago Press. Lascarides, A. and Asher, N.: 2007, Sdrt: Dynamic semantics with discourse structure., in H. Bunt and R. Muskens (eds), Computing Meaning: Volume 3, Tilburg: Kluwer. Laurent, J.-P., Denhiéres, G., Passerieux, C., Iakimova, G. and Hardy-Baylé, M.-C.: 2006, On understanding idiomatic language: The salience hypothesis assessed by ERPs., Brain Research 1068, 151–160. Leslie, A. M.: 1987, Pretense and representation: The origins of theory of mind., Psychological Review 94, 412–426. Levinson, S. C.: 1983, Pragmatics., Cambridge University Press, Cambridge. Levinson, S. C.: 2000, Presumptive Meanings. The Theory of Generalised Conversational Implicature., MIT Press, Cambridge, Mass. Lewis, D. S.: 1969, Convention., Harvard University Press, Cambridge, Mass. Lyons, J.: 1987, Semantics., in J. Lyons (ed.), New Horizons in linguistics., Vol. 2, London: Penguin, pp. 152–178. MacKay, G. and Shaw, A.: 2004, A comparative study of figurative language in children with autistic spectrum disorders., Child Language Teaching and Therapy 20, 13–32. Marantz, A.: 1997, No escape from syntax. don’t try morphological analysis in the privacy of your own lexicon., in A. Dimitriadis (ed.), Proceedings of the 21th Annual Penn Linguistics Colloquium., Penn Linguistics Club, pp. 201–225. Mashal, N., Faust, M., Hendler, T. and Jung-Beeman, M.: 2008, Hemispheric differences in processing the literal interpretation of idioms., Cortex 44, 848–860. McElree, B. and Frisson, Steven Pickering, M. J.: 2006, Deferred interpretations: Why starting dickens is taxing but reading dickens isn’t., Cognitive Science 30, 181–192. McElree, B., Traxler, M. J., Pickering, M. J., Seely, R. E. and Jackendoff, R.: 2001, Reading time evidence for enriched semantic composition., Cognition 78, B17– B25. McGlone, M.: 1996, Conceptual Metaphor and Figurative Language Interpretation: Food for Thought?, Journal of Memory and Language 35, 544–565. Morgan, J. L.: 1978, Two Types of Convention in Indirect Speech Acts., in P. Cole (ed.), Syntax and Semantics. Vol. 9., New York/ San Francisco/ London, pp. 261– 280. Nicolle, S. and Clark, B.: 1999, Experimental pragmatics and what is said: a response to gibbs and moise., Cognition 69, 337–354. Bibliography 269 Nissim, M., Sansó, A. and Soria, C.: 1999, Towards a compositional frame semantics., Proceedings of the European Conference on Cognitive Science . Siena, Italy. Noveck, I. A.: 2001, When children are more logical than adults: Experimental investigations of scalar implicature., Cognition 78, 165–188. Noveck, I. and Posada, A.: 2003, Characterising the time course of an implicature: An evoked potentials study., Brain and Language 85, 203–210. Nunberg, G.: 1978, The Pragmatics of Reference., Indiana University Linguistics Club 1978, Bloomington, IN. Nunberg, G.: 1979, The non-uniqueness of semantic solutions: polysemy., Linguistics and Philosophy 3, 143–184. Nunberg, G.: 1995, Transfers of Meaning., Journal of Semantics 12, 109–132. Nunberg, G.: 2002, The Pragmatics of Deferred Interpretation., in L. Horn and G. Ward (eds), Blackwell Encyclopedia of Pragmatics., Blackwell. Nunberg, G., Sag, I. A. and Wasow, T.: 1994, Idioms., Language 70(3), 491–538. Ortony, A., Schallert, D. L., Reynolds, R. E. and Antos, S. J.: 1978, Interpreting Metaphors and Idioms: Some Effects of Context on Comprehension., Journal of Verbal Learning and Verbal Behavior. 17, 465–477. Pagin, P.: 2005, Compositionality and context., in G. Preyer (ed.), Contextualism in Philosophy., Oxford University Press, Oxford, pp. 303–48. Pagin, P. and Pelletier, J.: 2007, Content, context and composition., in G. Peter and G. Preyer (eds), Content and Context. Essays on Semantics and Pragmatics., Oxford University Press, Oxford, pp. 25–62. Papafragou, A.: 1996, Figurative language and the semantics/pragmatics distinction., Language and Literature 5, 179–93. Perry, J.: 1998, Indexicals, Context and Unarticulated Constituents., Proceedings of the 1995 CSLI-Amsterdam Logic, Language and Computation Conference., Stanford: CSLI Publications. Pexman, P. M. and Glenwright, M.: 2007, How do typically developing children grasp the meaning of verbal irony?, Journal of Neurolinguistics 20, 178–196. Pickering, M. J., McElree, B., Frisson, S., Chen, L. and Traxler, M. J.: 2006, Underspecification and aspectual coercion., Discourse Processes 42, 131–155. Pickering, M. J., McElree, B. and Traxler, M. J.: 2005, The difficulty of coercion: A response to de almeida., Brain and Language 93, 1–9. Piñango, M. M., Winnick, A., Ullah, R. and Zurif, E.: 2006, Time-course of semantic composition: The case of aspectual coercion., Journal of Psycholinguistic Research 35, 233–244. 270 Bibliography Piñango, M. M., Zurif, E. B. and Jackendoff, R.: 1999, Real-time processing implications of enriched composition and the syntax-semantics interface., Journal of Psycholinguistic Research 28, 395–414. Pustejovsky, J.: 1991, The generative lexicon., Computational Linguistics 17(4). Pustejovsky, J.: 1995, The Generative Lexicon., MIT Press, Cambridge, Mass. Putnam, H.: 1975, The Meaning of ”Meaning”., in K. Gunderson (ed.), Language, mind, and knowledge., University of Minneapolis Press, Minneapolis, pp. 131–193. Pylkkänen, L., Llinás, R. and McElree, B.: 2004, Distinct effects of semantic plausibility and semantic composition in meg, in E. Halgren, S. Ahlfors, M. Hämäläinen and D. Cohen (eds), Biomag 2004: Proceedings of the 14th International Conference on Biomagnetism, Biomag: Boston, USA. Pylkkänen, L., Llinás, R. and Murphy, G. L.: 2006, The Representation of Polysemy: MEG Evidence., Journal of Cognitive Neuroscience 18, 1–13. Pylkkänen, L., Martin, A. E., McElree, B. and Smart, A.: 2009, The Anterior Midline Field: Coercion or decision making?, Brain and Language 108, 184–90. Pylkkänen, L. and McElree, B.: 2006, The syntax-semantic interface: On-line composition of sentence meaning., in M. Traxler and M. A. Gernsbacher (eds), Handbook of Psycholoinguistics., Elsevier, pp. 537–77. Pylkkänen, L., Oliveri, B. and Smart, A. J.: 2009, Semantic vs. world knowledge in prefrontal cortex., Language and cognitive processes 24, 1313–1334. Reboul, A.: 2004, Conversational Implicatures: Nonce or Generalised?, in I. Noveck and D. Sperber (eds), Experimental Pragmatics., Palgrave Macmillan, Basingstoke, pp. 322–333. Recanati, F.: 1989, The pragmatics of what is said., Mind and Behaviour 4, 295–329. Recanati, F.: 1995, The Alleged Priority of Literal Interpretation., Cognitive Science 19, 207–232. Recanati, F.: 2001, Literal/Nonliteral., Midwest Studies in Philosophy 15, 264–274. Recanati, F.: 25, 299345. 2002, Unarticulated constituents., Linguistics and Philosophy Recanati, F.: 2004, Literal Meaning., Cambridge University Press, Cambridge. Recanati, F.: n.d., Crazy minimalism. internet resource. Rodd, J., Gaskell, G. and Marslen-Wilson, W.: 2002, Making sense of semantic ambiguity: semantic competition in lexical access., Journal of Memory and Language 46, 245–66. Bibliography 271 Rolf, E.: 1995, Zur Grammatikalisierung konversationeller Implikaturen., in F. Liedtke (ed.), Implikaturen. Grammatische und pragmatische Analysen., Tübingen: Niemeyer, pp. 87–102. Rubio Fernández, P.: 2007, Suppression in Metaphor Interpretation: Differences between Meaning Selection and Meaning Construction., Journal of Semantics 24, 345–71. Rubio Fernández, P., Breheny, R. and Lee, M. W.: 2003, Context-independent information in concepts: An investigation of the notion of ”core features”., in F. Schmalhofer, R. M. Young and G. Katz (eds), Proceedings of EuroCogSci 03: The European Cognitive Science Conference 2003., London: Lawrence Erlbaum. Ruhl, C.: 1989, On Monosemy. A Study in Linguistic Semantics., State University of New York Press, Albany. Rundblad, G. and Annaz, D.: 2010, The atypical development of metaphor and metonymy comprehension in children with autism., Autism 14, 29–46. Rundblad, G. and Annaz, D.: in press, Metaphor and Metonymy Comprehension: Receptive Vocabulary and Conceptual Knowledge., British Journal of Developmental Psychology . Ruppenhofer, J., Ellsworth, M., Petruck, M. R. L., Johnson, C. R. and Scheffczyk, J.: 2010, FrameNet II: Extended Theory and Practice., http://framenet.icsi. berkeley.edu/book/book.pdf. Sag, I.: 1981, Formal Semantics and Extralinguistic Context., in P. Cole (ed.), Radical Pragmatics., Academic Press, New York, pp. 273–294. Saussure, F. d.: 1931, Grundfragen der allgemeinen Sprachwissenschaft, de Gruyter, Berlin. edited by Charles Bally and Albert Sechehaye. Schmidt, G. L. and Seger, C. A.: 2009, Neural correlates of metaphor processing: The roles of figurativeness, familiarity and difficulty., Brain and Cognition 71, 375–386. Schulz von Thun, F.: 1982, Miteinander reden. Störungen und Klärungen: Psychologie der zwischenmenschlichen Kommunikation., Vol. 1, Rororo. Schwoebel, J., Dews, S., Winner, E. and Srinivas, K.: 2000, Obligatory Processing of the Literal Meaning of Ironic Utterances: Further Evidence., Metaphor and Symbol 15, 47–61. Searle, J. R.: 1975, Indirect Speech Acts., in P. Cole and J. L. Morgan (eds), Syntax and Semantics 3: Speech Acts., Academic Press, London. Searle, J. R.: 1978, Literal Meaning., Erkenntnis 13, 207–224. Searle, J. R.: 1979, Metaphor., in A. Ortony (ed.), Metaphor and Thought., Cambridge University Press, Cambridge, Mass. 272 Bibliography Searle, J. R.: 1980, The background of meaning., in J. R. Searle, F. Kiefer and M. Bierwisch (eds), Speech act theory and pragmatics., Dordrecht. Shapiro, A. M. and Murphy, G. L.: 1993, Can You Answer a Question for Me? Processing Indirect Speech Acts., Journal of Memory and Language 32, 211–229. Sperber, D. and Wilson, D.: 1995, Relevance: Communication and Cognition., Blackwell, Oxford. second edition. Sperber, D. and Wilson, D.: 2002, Pragmatics, Modularity and Mind-reading., Mind and Language 17, 3–23. Sperber, D. and Wilson, D.: 2008, A Deflationary Account of Metaphor., in R. Gibbs (ed.), Handbook of Metaphor and Thought., Cambridge University Press, Cambridge. Stainton, R.: 2005, In Defense of Non-Sentential Assertion., in Z. G. Szabo (ed.), Semantics vs. Pragmatics., Oxford: OUP, pp. 383–457. Stanley, J.: 2000, Context and Logical Form., Linguistics and Philosophy 23, 391– 434. Stanley, J. and Szabo, Z. G.: 2000, On Quantifier Domain Restriction., Mind and Language 15, 219–261. Stern, J.: 2000, Metaphor in Context., MIT Press, Cambridge, Mass. Stern, J.: 2006, Metaphor, Literal, Literalism., Mind and Language 21, 243–279. Stern, J.: 2009, Metaphor doi:10.1007/s11098-009-9486-3. and Minimalism., Philosophical Studies . Storto, G. and Tanenhaus, M.: 2005, Are scalar implicatures computed online?, in E. M. et al. (ed.), Proceedings of Sinn and Bedeutung 9, Nijmegen: Nijmegen Center of Semantics. Strawson, P. F.: 1950, On referring., Mind 59, 320–44. Stringaris, A. K., Medford, N. C., Giampietro, V., Brammer, M. J. and David, A. S.: 2007, Deriving meaning: Distinct neural mechanisms for metaphoric, literal, and non-meaningful sentences., Brain and Language 100, 150–162. Taylor, J. R.: 1994, The Two-Level Approach to Meaning., Linguistische Berichte 149, 3–26. Taylor, J. R.: 1995, Approaches to Word Meaning: The Network Model (Langacker) and the Two-Level-Model (Bierwisch) in comparison., in R. Dirven and J. Vanparys (eds), Current Approaches to the Lexicon., Peter Lang Verlag, Frankfurt. Todorova, M., Straub, K., Badecker, W. and Frank, R.: 2000, Aspectual coercion and the online computation of sentential aspect., Journal of Memory and Language 53, 1–25. Bibliography 273 Traxler, M. J., Pickering, M. J. and McElree, B.: 2002, Coercion in sentence processing: Evidence from eye-movements and self-paced reading., Journal of Memory and Language 47, 530–547. Wearing, C.: 2010, Autism, Metaphor and Relevance Theory., Mind and Language 25, 196–216. Wharton, T.: 2002, Paul Grice, saying and meaning., UCL Working Papers in Linguistics 14. Wilson, D. and Sperber, D.: 1992, On verbal irony., Lingua 87, 53–76. Wilson, D. and Sperber, D.: 2000, Truthfulness and Relevance., UCL Working Papers in Linguistics 12, 215–254. Wunderlich, D.: 1977, On Problems of Speech Act Theory., in R. E. Butts and J. Hintikka (eds), Basic Problems in Methodology and Linguistics., Dordrecht, Holland, pp. 243–258.