ELSEVIER Journal of Pragmatics 31 (1999) 1655-1674 www.elsevier.nl/locate/pragma On the compositional and noncompositional nature of idiomatic expressions Debra A. Titone ~,*, Cynthia M. Connine b Department of Psychiatry, Harvard Medical School, McLean Hospital, Psychology Research Laboratory, 115 Mill Street, Belmont, MA 02178, USA h Department of Psychology, State University of New York at Binghamton, NY 13902-6000, USA Abstract The present paper reviews linguistic and psycholinguistic perspectives on idiom representation and models of idiom processing. Two approaches in defining idiom representation and processing characteristics are compared. According to the 'noncompositional approach', idioms are represented and processed similar to long words. In contrast, the 'compositional approach' emphasizes the semantic contribution of an idiom's component word meanings in interpretation. We argue that neither approach alone adequately captures the existing body of data on idiom processing, and propose a model of idiom representation and processing that ascribes noncompositional and compositional characteristics to idiomatic expressions. In this view, idiomatic expressions function simultaneously as semantically arbitrary word sequences and compositional phrases. Consistent with this hybrid model, the results of an eye tracking study are presented in which reading rates differ as a function of the inherent decomposability of idioms. © 1999 Elsevier Science B.V. All rights reserved. Keywords: Idiom processing; Compositionality; Eye movement recording; Semantic priming 1. Introduction The study of idiomaticity in language has been of long-standing interest to linguists, intent on characterizing the internal structure of languages, and psycholinguists, intent on characterizing the internal cognitive structure of language users. Idioms are usually grouped within a larger class of linguistic expressions termed figurative or nonliteral language. Also included in this class of expressions are metaphor, indirect speech acts, sarcasm, irony, metonymy, and many other linguistic * Corresponding author. E-mail: dtitone@mclean.harvard.edu 0378-2166/99/$ - see front matter O 1999 Elsevier Science B.V. All rights reserved. PII: S0378-2166(99)00008-9 1656 D.A. Titone, C.M. Connine / Journal of Pragmatics 31 (1999) 1655-1674 constructions that produce an apparent need for comprehenders to go beyond what is literally stated to apprehend the communicative intent of speakers, and, consequently, the 'meaning' of utterances. The present paper reviews linguistic and psycholinguistic perspectives on idiom representation and models of idiom processing. Two approaches in defining the representation and processing characteristics of idioms will be discussed and compared. In the 'noncompositional approach', idioms are hypothesized to be represented as long words. Models of idiom comprehension that adopt the noncompositional approach differ in terms of when and how idiomatic meanings are holistically retrieved, and assume that the semantic characteristics of an idiom's word components do not influence or direct their comprehension. In contrast, the 'compositional approach' posits that the literal meanings of an idiom's word components are critical in their interpretation. In this view, idiomatic word sequences are semantically and syntactically analyzable, and realization of idiomatic meaning is based upon the products of these literal analyses. We will argue that both approaches, by themselves, are flawed. Each alone cannot account for the full range of idiom types and idiom processing characteristics. In this paper, we support a hybrid view of idiom representation and processing that simultaneously treats idioms as noncompositional and compositional word sequences. That is, idiomatic meanings are both directly retrieved and literally analyzed during comprehension. This type of hybrid model better accounts for differences in idiomatic types, and better accommodates data from psycholinguistic studies of idiom processing. 2. The noncompositional approach In both the domains of linguistics and psycholinguistics, idioms have held a special position within the general class of figurative language, primarily due to their unique status as expressions whose 'nonliteral' meanings are highly associated with very specific, and often arbitrary, configurations of words. Idioms have been described as long words that syntactically and semantically behave as lexical entries. Consider the idiomatic expression, kick the bucket. The literal meaning of the expression, to strike a bucket with one's foot has no obvious semantic overlap with the figurative meaning of the expression to die suddenly. The noncompositional definition of idioms has been espoused by linguists (e.g., Chomsky, 1980; Fraser, 1970; van der Linden, 1992), and psycholinguists (e.g., Bobrow and Bell, 1973; Gibbs, 1980; Swinney and Cutler, 1979) alike. Within linguistics, the view that idioms are equivalent to long words initially resulted in speculation that the syntactic behavior of idioms mapped onto idiomatic meaning in a very direct way (e.g., Newmeyer, 1972). Accordingly, kick the bucket was thought to syntactically behave like the intransitive verb 'die' that it denotes. Unfortunately for this view, an isomorphic relationship between idiomatic meaning and syntactic function is not the case for all idioms (e.g., gave up the ghost; Nunberg, 1978). However, that a given idiomatic expression may take different syntactic D.A. Titone, C.M. Connine / Journal of Pragmatics 31 (1999) 1655-1674 1657 forms (e.g., Geoff spilled the beans, The beans were spilled by Geof]) provides data bearing on the development of grammatical theory. For instance, Nunberg et al. (1994) point out that the syntactic flexibility of idioms has a long linguistic history of bolstering claims about the generativity of grammar. Because transformational grammars consider two different syntactic forms of an idiom as comprising two instances of the same idiomatic expression, they are seen as a more parsimonious account of syntactic flexibility than non-generative grammars in which two different syntactic uses of an idiom constitute instances of two separate idioms. These linguistic arguments for the existence of transformations based on the syntactic behavior of idioms have all relied on the assumption of a semantically arbitrary relationship between idiom meanings and idiom parts (e.g., Chomsky, 1980). The classification of idioms as noncompositional, unitary word strings is also evident in early psycholinguistic studies and models of idiom comprehension. Because traditional models of 'literal' language comprehension, in which phrasal meaning is based upon a strict compositional analysis, cannot address how figurative meanings are realized during comprehension, several specific models of idiom processing have been proposed (e.g., Bobrow and Bell, 1973; Cacciari and Tabossi, 1988; Gibbs, 1980; Swinney and Cutler, 1979). These models all assume that idiom comprehension simply requires memory retrieval of stipulated idiomatic meanings, but they differ in terms of how and when these meanings are accessed. The literal processing model (Bobrow and Bell, 1973) posits that a mental 'list' of idiomatic expressions in memory is accessed through a special idiomatic 'mode' of processing. Similar to the three-stage model of metaphor comprehension (Searle, 1979), the comprehension device first attempts to construct a literal interpretation, and if this process fails, an idiomatic mode is instantiated and the figurative meaning is retrieved from the list. Consistent with this hypothesis, Bobrow and Bell found a decrease in the probability of an idiomatic interpretation for idioms preceded by a series of literal phrases. They argued that presentation of literal phrases circumvented instantiation of the idiomatic mode, and reduced the likelihood of subjects generating idiomatic interpretations. However, as Swinney and Cutler (1979) pointed out, this study relied upon self-report measures of comprehension, and therefore may not be useful in drawing conclusions about immediate processing. In contrast to the literal processing model, Swinney and Cutler (1979) proposed the lexical representation model which held that idiomatic expressions were stored and retrieved similar to long words. They argued that computation of literal and figurative meanings is initiated following presentation of the first word of an idiom. Automatic retrieval of idiomatic meanings should, however, precede a complete specification of literal meaning which involves both the retrieval and compositional analysis of each component word. Their model received support from a study using a visual phrase classification task. Response latencies for deciding whether a word string was a valid English phrase were faster for idioms than for control literal expressions (see also Glass, 1983). Similar results showing processing advantages for idioms in favor of the lexical representation model have been obtained for studies using sentence reading (Schweigert, 1986) and sentence monitoring (Estill and Kemper, 1982) paradigms. 1658 D.A. Titone, C.M. Connine / Journal of Pragmatics 31 (1999) 1655-1674 Similar to the lexical representation model, the direct access model proposed by Gibbs (1980, 1986) states that idiomatic meanings have computational priority over literal meanings. However, the direct access model takes this claim a step further by specifically stating that the literal meanings of an idiom's component words will not be combined to form a literal interpretation of the phrase (Gibbs et al., 1989b). Consistent with this view, reading rates were faster for idioms in figuratively biased sentence contexts than for idioms in literally biased sentence contexts (see also Ortony et al., 1978; Schweigert and Moates, 1988). Moreover, recall was better for idioms in literal contexts than it was for idioms in idiomatic contexts. This recall difference was interpreted as a 'double-take' effect in which literally biased idioms benefited from being processed twice (both idiomatically and literally). Taken together, these results are consistent with the direct access model because they suggest primacy of the idiomatic interpretation. Finally, support for the noncompositional view comes from studies examining the role of idiom familiarity. As applied to words, familiarity has been defined as the frequency with which a comprehender encounters a word in its written or spoken form (Gemsbacher, 1984), and the degree to which the meaning of a word is known to comprehenders (Nusbaum et al., 1984). Word familiarity has been shown to have an important influence on word recognition (Connine et al., 1990; Gemsbacher, 1984) and the comprehension of novel metaphors (Blasko and Connine, 1993). The relationship between familiarity and idiom processing was studied by Schweigert (1986; see also Schweigert and Moates, 1988). Reading rates for sentences containing highly familiar idioms were shorter than those for sentences containing low familiar idioms. Cronk and Schweigert (1992) also found that highly familiar idioms were read more quickly than less familiar idioms. These studies demonstrate that idioms, like words, are processed more quickly depending on the degree of experience a comprehender has with a particular phrase. Although the specifics of each of the previously discussed idiom processing models differ in terms of how and when idiomatic meaning is retrieved, they all share the assumption that idiomatic meanings of particular configurations of words are semantically distinct from the literal meanings of the component words. Therefore, an idiomatic interpretation results from retrieval of a unitary, and semantically arbitrary phrasal meaning, whereas a literal interpretation of an idiomatic phrase results from accessing and combining the individual word meanings of the phrase. 2.1. Problems with the noncompositional approach There are a number of problems with the assumption that idioms are best described as noncompositional word strings. For one, the syntactic flexibility of idioms is not arbitrary in a way that would be predicted by assuming that individual word components have nothing to do with idiomatic meaning (Wasow et al., 1983). Given that language users are never explicitly taught what idioms are amenable to syntactic modification (e.g., which idioms passivize and which do not), and that syntactic flexibility of idioms is not based on the frequency of occurrence of particular syntactic forms, a noncompositional definition cannot easily account for the consid- D.A. Titone, C.M. Connine / Journal of Pragmatics 31 (1999) 1655-167~ 1659 erable agreement among language users as to which idioms are syntactically flexible and which are not (Wasow et al., 1983). Additionally, there is evidence showing that idioms possess a great deal of internal semantic structure. Idioms are modifiable with adjectives or relative clauses (e.g., She did not spill any of those precious beans), and parts of idioms may be quantified (e.g., She didn't spill a single bean), emphasized through topicalization (e.g., Those beans, she would never spill), or omitted altogether (e.g., She didn't spill the beans yesterday, but spilled them today) without disrupting comprehension or awareness of their idiomaticity (Nunberg et al., 1994; Sag, 1976; Williams, 1977). Consistent with this view, Nunberg et al. (1994; see also Lakoff, 1987) have identified semantically-based families of idioms in which individual components combine with other words to form distinct but related phrases. These variations are possible based upon either the semantics of the verb (e.g., lose one's mind or marbles, throw someone to the lions or wolves) or the noun phrase (e.g., keep or lose one's cool, step or tread on someone's toes). Some theorists have argued that common idiomatic (and metaphorical) forms result from conceptually grounded metaphors applicable to the referents of all forms of literal and nonliteral language (e.g., Lakoff, 1987). To this end, Nayak and Gibbs (1990) found that readers' judgments about the plausibility of an idiom in a particular context were influenced by the degree to which the context instantiated a conceptual metaphor that comprised the basis of the idiom. Moreover, a comparison of the animacy of idiomatic referents and idiom components shows a statistical tendency for idiomatic phrases to refer to animate entities, whereas there is a tendency for the component noun phrases and verbs of idioms to refer to inanimate entities (Nunberg et al., 1994). This is in contrast to other forms of nonliteral language in which metaphorical mappings between literal meanings of words and figurative meaning tend to preserve animacy. Nunberg et al. (1994) attribute this difference to the proverbiality function that idioms serve in which abstract messages are packaged in concrete idiomatic forms (e.g., look a gift horse in the mouth, horse meaning something that is freely offered). This kind of internal semantic structure found for idiomatic expressions demonstrates that idiom meanings have much to do with idiom parts. In addition to studies of the semantic structure of idiom representation, studies of idiom processing demonstrate that idiomatic word sequences are literally analyzed during comprehension. In an experiment by Peterson and Burgess (1993) that examined the degree to which a syntactic analysis of idioms is performed, subjects heard sentence fragments that ended with truncated idioms (e.g., He nearly kicked the ...), and saw visual targets consisting of a syntactically appropriate completion of the phrase (i.e., nouns) or a syntactically inappropriate continuation of the phrase (i.e., verbs). Naming latencies were faster for noun targets than for verb targets, suggesting that idioms are syntactically processed similar to non-idiomatic word sequences (see also Peterson et al., 1989). These results imply that idioms are literally processed to some degree because syntactic processing does not terminate after encountering an idiom. Other results have shown that literal word meanings are activated during idiom comprehension. In a series of experiments, Cacciari and Tabossi (1988) found prim- 1660 D.A. Titone, C.M. Connine / Journal of Pragmatics 31 (1999) 1655-1674 ing for idiomatic interpretations at the offset of the final word of the idiom, but no priming for the literal meaning of the idiom-final word. A post-hoc inspection revealed that the idiom-final word was highly predictable (i.e., had high cloze probability) for the majority of their materials. In a second experiment that used idioms rated as low predictable, idiomatic priming was not obtained at offset, but was 300 msec later. In contrast, literal word meaning activation for low predictable idioms was obtained at offset, but not 300 msec later. Based upon these results, Cacciari and Tabossi (1988) proposed an account in which idiomatic meaning is comprised of a distributed representation rather than a lexical entry. According to this view, later termed the configuration hypothesis (Cacciari and Glucksberg, 1991), activation of the idiom meaning takes place only after a sufficient portion of the string is encountered; what determines sufficiency varies across idioms. Cacciari and Tabossi used the term 'idiomatic key' to refer to the point at which an idiomatic configuration emerges. Activation of the literal meanings of an idiom's component words occurs, and once the idiomatic key is encountered, the idiomatic configuration emerges and the meaning is retrieved. One potential concern with this model as described, however, is that retrieval of the idiomatic meaning, although initiated by the access of word meanings, is a unitary step that may or may not be qualitatively tied to what those word meanings actually are. That is, the configuration may emerge on the basis of knowledge of the frequency of co-occurrence of idiom parts rather than on the basis of knowledge about the semantics of those parts. The model implies, therefore, that subsequent literal processing may be terminated once the key is accessed. Indeed, the pattern of results for high predictable idioms obtained by Cacciari and Tabossi (1988) showed no activation at offset of the literal meaning of the idiom-final word. However, this is not to say that literal word meanings were never activated. It is possible that the idiom-final word may have been activated early, but inhibited by the end of the phrase (i.e., the point at which the visual target was presented). This possibility might be especially attractive, given that the idioms used by Cacciari and Tabossi did not have plausible literal phrase-level interpretations. For these types of noncompositional idioms, sustained activation of the literal meaning of the idiom-final word would not be necessary for any potential higher-level representation that might be constructed from the sequence. Thus, the configuration model predicts that activation of the idiomatic meaning subsequent to access of the idiomatic key may terminate literal processing. In this way, the model is reminiscent of the noncompositional approach. In another cross-modal priming experiment addressing this issue, Titone and Connine (1994a) explicitly investigated the effects of predictability and literal plausibility on idiomatic and literal activation. Idiomatic activation was observed as early as the penultimate word of the idiom for high predictable idioms but not for low predictable idioms, and idiomatic activation was observed at offset of the final word for both phrase types. However, in the Titone and Connine (1994a) study, literal activation varied as a function of predictability and literal plausibility of the phrase. Literal activation of word meanings was obtained regardless of literal plausibility for low predictable idioms. In contrast, literal activation was obtained only for high pre- D.A. Titone, C.M. Connine / Journal of Pragmatics 31 (1999) 1655-1674 1661 dictable idioms with plausible literal interpretations. High predictable idioms without plausible literal interpretations did not show literal word meaning activation. These results suggested that early idiomatic activation for high predictable idioms also involves early access of the idiom-final word and a subsequent evaluation of the phrase's literal plausibility in a way that is not possible for low predictable idioms. Literal meanings remain activated if they are potentially relevant to a higher-level representation of the phrase's meaning (i.e., when a literal interpretation of the phrase is possible), but are suppressed if they are not relevant (i.e., the phrase is not literally plausible). The important point to be abstracted from this work is that literal word meanings should always be activated at some point during comprehension. Therefore, accessing the idiomatic key does not terminate word meaning activation for later words in the idiom. 3. The compositional approach The previous linguistic analyses and processing studies suggest that traditional noncompositional definitions of idiomaticity, and processing models based upon these definitions are inadequate by themselves. Idiomatic expressions exhibit a high degree of internal semantic structure, and literal processing does not stop when an idiom is encountered during comprehension. Much the same way that literal and other figurative aspects of language (e.g., metaphor) are comprehended, the compositional approach to idiom representation and processing is based on the notion that idiomatic meanings are built simultaneously out of literal word meanings and the specific interpretation of these word meanings within a particular context. Consequently, much of the work adopting the compositional approach has sought to understand the ways in which idioms differ in terms of their internal semantic structure and of the processing consequences these differences entail. Semantic taxonomies have been posited to describe how idioms differ in their compositionality, and how these differences may have implications for process models of idiom comprehension. Nunberg (1978) proposed a typology for characterizing how literal word meanings of idiom components contribute (or do not contribute) to the overall interpretation of idiomatic phrases (see also Gibbs, 1992; Gibbs and Nayak, 1989; Gibbs et al., 1989a,b). According to this system, idiomatic expressions may be arranged into three different classes. Normally decomposable idioms are expressions in which a part of the idiom is used literally (e.g., the question in pop the question, save in save your skin). Abnormally decomposable idioms are expressions where the referents of an idiom's parts can be identified metaphorically (e.g., maker in meet your maker, buck in the idiom pass the buck). Finally, semantically nondecomposable idioms fit the traditional definition because the idiom meaning is less likely to be compositionally derived from the words that comprise the string (e.g., chew the fat). Using this typology, Gibbs and colleagues (e.g., Gibbs and Nayak, 1989; Gibbs et al., 1989a,b) demonstrated that subjects are capable of reliably classifying idioms into these three categories. In terms of idiom processing, Gibbs et al. (1989b) found 1662 D.A. Titone, C.M. Connine / Journal of Pragmatics 31 (1999) 1655-1674 that people read sentences containing decomposable idioms faster than sentences containing nondecomposable idioms. These results suggest that the literal meanings of words facilitate comprehension of idioms to the extent that they semantically overlap with the idiomatic meaning. The degree of decomposability has also been found to be related to the syntactic (Gibbs and O'Brien, 1989) and lexical (Gibbs et al., 1989a) flexibility of idioms. In these studies, idioms rated as decomposable are judged to be more syntactically and lexically flexible than idioms rated as nondecomposable (as they should be if semantic characteristics of word meaning drive idiomatic meaning). Here, word order and selection are less important than the meaning of the parts put together. Semantically nondecomposable idioms should be less flexible because the idiomatic meaning depends more on their being recognized in a unitary form rather than combining individual word meanings. Although characterizing different classes of idiomatic expressions into the normally, abnormally, and nondecomposable typology has been a useful way of moving beyond the early noncompositional definition, additional research suggests that there may be some pitfalls with this exact classification scheme. For instance, Titone and Cormine (1994b) found that although subjects could easily classify idioms as decomposable or nondecomposable, the reliability of classifying idioms rated as decomposable overall into the normal and abnormal subtypes was of questionable reliability. Second, although this classification scheme provides a description of decomposable idioms in terms of their internal semantics and processing characteristics, it does not easily address the potential differences in semantic flexibility among nondecomposable idioms, and implies that the processes involved in comprehension and production of nondecomposable and decomposable idioms may differ markedly. Although literal word meanings for nondecomposable idioms are likely to play less of a role in structuring idiomatic meaning, literal meanings may partially constrain an interpretation. For example, although the words kick, the, and bucket do not readily map on to the idiomatic meaning to die, John kicked the bucket in the car accident is a more plausible statement than is John lay kicking the bucket due to his chronic illness (Glucksberg, 1991; see also Cacciari and Glucksberg, 1991). This would be so because the verb kick denotes a sudden action, therefore, its participation in a phrase that overall refers to a sudden event is more plausible than a phrase that overall refers to an event that unfolds over time. Somewhat related to this point, nondecomposable idioms have been shown to be semantically productive. McGlone et al. (1994; see also Glucksberg, 1993), for example, have found that semantic characteristics of the component words of idioms play a large role in innovative uses of even semantically nondecomposable idioms (termed opaque in their classification scheme; e.g., burning the candle at three ends). They found that novel idiomatic uses are understood more slowly than familiar idioms, but as quickly as literal paraphrases, suggesting that integrating literal meanings into novel idiomatic forms does not require additional processing time compared to what is necessary for processing 'literal' language. Given the semantic flexibility of even nondecomposable idioms, it is likely that there is substantial overlap in the processing characteristics of nondecomposable and decomposable idioms. D.A. Titone, C.M. Connine / Journal of Pragmatics 31 (1999) 1655-1674 1663 Finally, idioms often convey their figurative meanings by a metaphorical mapping between a literal interpretation of the phrase and an idiomatic referent (e.g., counting chickens before they're hatched; Glucksberg, 1993). In this way, a compositional analysis of idioms based upon a literal analysis of the entire phrase is required, rather than an analysis of how an idiom's parts map onto the figurative meaning of the phrase. In another taxonomy of idiom compositionality, Cacciari and Glucksberg (1991; see also Glucksberg, 1993) classify idioms as opaque, transparent, or quasimetaphorical. Opaque idioms are phrases such as kick the bucket in which there is some degree of semantic constraint on interpretation of the idiom given the meanings of the component words. In contrast, transparent idioms are phrases in which there is a direct mapping of literal word meanings to idiomatic meaning. For example, spill in the idiom spill the beans which literal translates to divulge a secret directly maps onto the verb divulge, and the beans directly maps a secret. Finally, idioms classified as quasi-metaphorical are phrases in which the overall literal meaning of the phrase metaphorically maps onto the idiomatic meaning. Similar to metaphorical expressions, these idioms convey their meaning via allusional content (e.g., Glucksberg, 1991), that is, they simultaneously refer to an ideal exemplar of a concept and to the contextually determined referent in a particular communicative situation. In contrast with the previous two taxonomies that describe differences in one semantic dimension of idiomatic expressions (i.e., compositionality), Nunberg et al. (1994) argue that idioms may differ along three orthogonal semantic dimensions, compositionality, conventionality, and transparency. Further, they argue that many of the difficulties in studying idioms arise from a confusion among these three dimensions. According to the Nunberg et al. analysis, compositionality refers to the degree to which a known idiomatic meaning can be analyzed into parts in a post hoc way. That is, compositionality refers to the ease with which literal word meanings of idiomatic expressions can be mapped onto components of idiomatic meaning once the idiomatic meaning has been apprehended and is clearly known. For some idioms such as grease the wheels, this mapping may be readily apparent with a first encounter with an idiom. For other idioms such as get your goat, this mapping may only become apparent with accumulated experience with an idiom. The notion of acquired compositionality is similar to the Phrase Induced Polysemy model (PIP) described by Glucksberg (1993). According to the PIP model, the individual word components of idiomatic expressions become polysemous because of their frequent use in idiomatic expressions (i.e., new senses for words are created by extrapolating from their compositional meaning within the idiomatic context). This backward engineering of semantics enables idioms to be utilized in novel ways without producing disruptions in their comprehension. In contrast to compositionality, idiom conventionality refers to the degree to which idiomatic meanings are not predictable based upon knowledge of the word components in isolation, and knowledge of the conventions of a particular language environment (Nunberg et al., 1994). This is a much more restrictive definition of idiomaticity in that it reflects the degree to which a speaker would be able to 1664 D.A. Titone, C.M. Connine / Journal of Pragmatics 31 (1999) 1655-1674 recover an idiomatic meaning in the absence of all contextual information (i.e., understand an idiom for the first time) in a particular communicative environment, irrespective of how idiom parts map onto idiom meanings. To illustrate the difference between compositionality and conventionality, Nunberg et al. discuss the difference between two literal terms for referring to the structure in the middle of a highway that separates the two directions of traffic, center divider and middle separator. Both terms are compositional in that a post hoc analysis of their literal meanings are semantically consistent with the actual referent. However, the term center divider is used far more often in the language in referring to this particular entity than is middle separator and is, therefore, likely to be comprehended more efficiently than the latter unconventional term. Applied to idiomatic sequences, although the idiomatic meaning to die would not be predictable given the word sequence kick the bucket for a naive language user, individuals who are members of a particular linguistic community would have no trouble understanding the sequence given its conventionality within that linguistic community. In this way, conventionality of an expression is likely to be related to the frequency with which an expression is encountered. However, this is not to say that conventionality directly maps onto idiom frequency. Words have highly conventionalized relationships with their associated meanings, yet they may differ in their frequency of occurrence in a language. Similarly, although idioms have highly conventionalized relationships with their figurative meanings, they may also differ in frequency of occurrence in a language. In addition to conventionality and compositionality, the transparency of idiomatic expressions refers to the degree to which the original motivation of these phrases is immediately accessible. To illustrate transparency, consider the idiom jump the gun. This idiom would be classified as transparent insofar as its idiomatic meaning, start ahead of time, is directly related to a literal interpretation of the phrase in which jumping the gun at the start of a race refers to beginning a race before one is supposed to. In this way, jump the gun is both transparent and compositional because both the motivation for the expression and identification of how the individual parts map onto the expression are easily discernible. However, transparency is orthogonal to compositionality such that idioms like saw logs are transparent (i.e., it is clear how this expression motivates the idiomatic meaning to sleep) but not compositional (i.e., it is not clear how the individual components of the phrase map on the idiomatic meaning to sleep; Nunberg et al., 1994). Nunberg et al. (1994) argue that these three dimensions, compositionality, conventionality, and transparency, have largely been misconstrued as compositionality in previous idiom research. That is, compositionality has been used as a catch-all to describe any potential semantic relationship existing between idiom meanings and idiom parts. To eliminate this confusion, they propose a terminology labeling idioms as idiomatic combinations or idiomatic phrases that is based upon differences in compositionality as they define it. Idiomatic combinations are word sequences in which the idiomatic meaning is distributed across the word components of the phrase (either in a post hoc or conventional way), whereas idiomatic phrases are word sequences in which the idiomatic meaning is not distributed across the word compo- D.A. Titone, C.M. Connine / Journal of Pragmatics 31 (1999) 1655-1674 1665 nents. Moreover, both idiomatic combinations and idiomatic phrases may be more or less conventional and transparent. 3.1. Problems with the compositional approach To summarize the compositional approach, idiom representation and processing are very much like literal language. Idiomatic parts usually map onto idiomatic meanings either in a way that is clear upon the first encounter with an idiom, or in a way that evolves over experience with an idiom. Therefore, when a phrase is encountered, the idiomatic meaning is built (to some degree) out of the literal meanings of the parts of the phrase. This model adequately captures linguistic data showing the semantic and syntactic flexibility of idioms, and accounts for studies of idiom processing showing that idioms are literally processed during comprehension. Clearly, idioms are not simply nondecomposable word sequences that function as individual words that are not subject to literal processing. However, a compositional analysis of idiomatic expressions is not all that is involved in their comprehension. Consistent with the noncompositional approach, idiomatic expressions are highly overlearned word sequences that comprehenders have experience with as holistic units. That the component words of idioms influence their interpretation does not discount the possibility that there exists a prepackaged meaning associated with very particular configurations of words. It stands to reason, therefore, that idiomatic sequences are processed more quickly and efficiently than nonidiomatic sequences, and familiarity plays a large role in how quickly idiomatic meanings are apprehended during comprehension (e.g., Cronk and Schweigert, 1992; Gibbs, 1980; Swinney and Cutler, 1979). Additionally, evidence from cross-modal priming studies shows that, like words, idioms have recognition points (i.e., idiomatic keys) that also vary as a function of experience (Cacciari and TabossL 1988; Titone and Connine, 1994a,b). These results showing differences in retrieval of idiomatic meanings indicate that some kind of direct look-up occurs that may be influenced by the degree of experience a comprehender has with an idiomatic string. Given that location of the idiomatic key does not correlate with degree of compositionality, it is likely that accessing an idiomatic key simply provides enough information to language users to enable them to extract the idiomatic pattern and retrieve a meaning. In addition to apprehension of idiomatic meanings, literal meanings of the idioms are simultaneously constructed from a compositional analyses of the component words in the phrase. In the idiom processing studies described thus far, we have presented evidence showing that literal processing of idioms is carried out to some degree; however, it is less clear to which degree a complete literal analysis of idiomatic expressions is performed. In evaluating the different studies of idiom processing, it is important to take into consideration the kinds of evidence used to bolster claims about the literal processing of idiomatic expressions. For instance, in the cross-modal priming studies previously described, it is important to distinguish between literal meaning as applied to activation of individual words, and literal meaning as applied to a constructed literal interpretation of the entire phrase (Titone 1666 D.A. Titone. C.M. Connine / Journal of Pragmatics 31 (1999) 1655-1674 and Connine, 1994a). One potential problem with the use of cross-modal priming for investigating figurative phrases such as idioms (e.g., Cacciari and Tabossi, 1988; Titone and Connine, 1994a) is that although idiomatic meanings can usually be captured by a single word target, literal meanings of phrases cannot. Kick the bucket idiomatically means die (i.e., one word), but literally means strike a pail with one's foot (i.e., multiple words). Literal targets in cross-modal priming experiments, consequently, are related to the literal meaning of one word in the idiom. Activation of individual words alone does not guarantee that the literal meaning of the phrase has been processed to completion (although no activation of individual words is sufficient to conclude the converse). In the Titone and Connine (1994a) study, literal processing at the phrasal level was concluded based upon the difference in literal word meaning activation for low and high predictable idioms with and without plausible literal phrasal interpretations. Given that literal word meanings were activated for all idiom types except high predictable implausible idioms, it is likely that early determination of phrasal implausibility for high predictable idioms was responsible for the suppression of literal word meanings. Additional evidence that literal phrasal meanings are constructed comes from another cross modal priming study by Colombo (1993) in which literal activation was obtained regardless of contextual bias for nondecomposable idioms. Indeed, Colombo (1993) also reported results of a word-by-word reading study which suggested that there was a preference for readers to integrate the literal phrasal meaning over the idiomatic meaning in a neutral context. 4. A synthesis of noncompositional and compositional approaches The findings from psycholinguistic studies of idiom processing provide evidence for both noncompositional and compositional approaches to idiom processing. On the one hand, we know that idiomatic expressions, especially highly familiar ones, enjoy a processing advantage over nonidiomatic phrases, and we know that idioms produced in idiomatic contexts tend to be processed more quickly than idioms produced in nonidiomatic, literal contexts. This evidence supports the noncompositional approach. On the other hand, we also know that idiomatic meanings are constrained to a large degree by the component meanings of their parts (i.e., idioms can be used innovatively, there exist families of semantically related idioms sharing component word parts), and we know that literal processing (i.e., syntactic and semantic analysis) does not terminate when an idiomatic word sequence is encountered. Rather than argue for one approach over another, we wish to argue for a hybrid model of idiom comprehension that characterizes idiomatic expressions both as unitary word configurations and compositional word sequences (see also Cacciari, 1993; Glucksberg, 1993), thereby incorporating aspects of the noncompositional and compositional approaches. According to this hybrid model, activation of idiomatic meanings, and the activation and use of literal meanings during comprehension, will be a function of the degree to which idioms are conventional and compositional (Nunberg et al., 1994). This model is similar in approach to previously articulated D.A. Titone, C.M. Connine / Journal of Pragmatics 31 (1999) 1655-1674 1667 models of idiom processing (e.g., Cacciari and Glucksberg, 1991; McGlone et al., 1993). Minimally, it differs by adopting the idiomatic classification scheme of Nunberg et al. (1994) in which idioms may be sorted as a function of their compositionality, transparency, and conventionality. Consider first compositionality (i.e., the degree to which idiom components can be readily mapped onto the overall meaning of the phrase). We know that idioms are processed literally to some degree, and that literal word meanings are activated during idiom comprehension. It is highly likely, then, that the products of these literal analyses of idiomatic expressions will contribute to the apprehension and interpretation of idiomatic meanings, especially for decomposable idioms (i.e., phrases in which the semantics of idiom parts are related to the semantics of the entire phrase). Consider next conventionality, which refers to the degree to which a particular configuration of words is likely to be idiomatically meaningful within a particular linguistic environment. Highly conventionalized idioms are phrases that all language users have a great deal of accumulated experience with as holistic entities. Consequently, regardless of compositionality, the association between the particular configuration of words and a specific idiomatic meaning is highly overleamed. This property of idioms is especially important for the comprehension of noncompositional, idiomatic phrases. First. it makes possible the rapid comprehension of idiomatic phrases (for which the products of literal analyses are of little help) by allowing for direct retrieval of a stipulated idiomatic meaning. It is likely that location of the idiomatic key within the confines of the configuration model (e.g., Cacciari and Tabossi, 1988; Cacciari and Glucksberg, 1991) varies as a function of idiom conventionality. Support for this idea comes from a normative study by Titone and Connine (1994b) showing that idiom predictability significantly correlates with idiom frequency. As previously mentioned, activation of the idiomatic key and retrieval of the idiomatic meaning may not be a semantic effect. Direct retrieval of idiomatic meanings may be triggered simply by the co-occurrence of words embedded in conventionalized idiomatic combinations and phrases. Direct retrieval as a function of encountering the idiomatic key would be true for both idiomatic phrases (i.e., nondecomposable idioms) and idiomatic combinations (i.e., decomposable idioms). However, although direct retrieval of idiomatic meanings should be similar for nondecomposable idiomatic phrases and decomposable idiomatic combinations, literal analysis of the phrase, ongoing during idiom processing, should contribute differentially to the idiomatic interpretation for these phrase types. Specifically, literal processing should augment the idiomatic interpretation for idiomatic combinations, but not for idiomatic phrases. This would be so because of the close semantic relationship between idiomatic meanings and the literal meanings of the idiom's component parts, and the distant semantic relationship between idiomatic meanings and the literal meanings of the idiom's component parts. We propose that the disparity between literal and idiomatic meanings of nondecomposable idioms should constitute a hindrance in processing that does not exist for decomposable idioms, whose literal and idiomatic meanings are semantically related. If one presumes that the product of idiom comprehension is an integration of 1668 D.A. Titone. C.M. Connine / Journal of Pragmatics 31 (1999) 1655-1674 the meanings of its literal constituents and idiomatic sense into a discourse representation, then nondecomposable idioms should incur a processing cost that decomposable idioms should not. 4.1. A test of a hybrid model of idiom processing: An experiment using eye movement recordings To investigate the potential differences in processing of nondecomposable and decomposable idioms we conducted an eye tracking study. In designing this experiment, we were interested in determining the degree to which idiomatic and literal meanings are initially computed during idiom processing, and in determining the degree to which idiomatic and literal meanings, once available, are integrated into a sentence level representation for these idiom types. A set of 16 nondecomposable and 16 decomposable idioms (rated as highly frequent, highly meaningful, and of high literal plausibility according to a set of norms collected by Titone and Connine, 1994b) were embedded in sentences. Similar to experiments investigating the role of context in lexical ambiguity resolution (e.g., Frazier and Rayner, 1990), these sentences had biased contexts that preceded or followed the idiom, and they were biased towards either the idiomatic or literal meaning. An example sentence set for nondecomposable idioms are the following: She finally kicked the bucket after being ill for months (idiom context following), After being ill for months, she finally kicked the bucket (idiom context preceding), She finally kicked the bucket, forgetting to move it from the path (literal context following), and Forgetting to move it from the path, she finally kicked the bucket (literal context preceding). An example sentence set for decomposable idioms are the following: He tried to save his skin by getting his work done on time (idiom context following), By getting his work done on time, he tried to save his skin (idiom context preceding), He tried to save his skin by avoiding the tanning salons (literal context following), and By avoiding the tanning salons, he tried to save his skin (literal context preceding). The logic of this experiment was to use differences in reading time in the idiom and disambiguating regions of the sentences to determine the nature of the representation that was computed for the idiomatic sequences. A conflict between the computed representation and the subsequent bias of the sentence should produce a disruption in reading rate. We were specifically interested in investigating the pattern of reading time for the idiom regions of these sentences as a function of contextual bias and location of the disambiguating information. Analysis of reading rates for the ambiguous idiom region as a function of context position (preceding and following) and bias (idiomatic and literal) provides information regarding computation of idiomatic and literal meanings. If reading rates of idioms preceded by biased contexts (literal and figurative) are slower than reading rates of idioms not preceded by biased contexts, we may conclude that there is a processing cost associated with having to select between two active meanings of the phrase. In contrast, if an increase in reading rates of the idiom region occur only when a literal context precedes the idiom, we may conclude that a literal analysis of idiomatic sequences either did not occur, or the processing cost associated with selecting between two active phrasal D.A. Titone, C.M. Connine / Journal of Pragmatics 31 (1999) 1655-1674 1669 meanings is reduced (as it might be for decomposable idioms whose idiomatic and literal phrasal meanings are semantically related). In contrast to the analysis of reading rates of the idiom region, an analysis of reading rates of the disambiguating regions that follow or precede the idioms provides information regarding what meaning was integrated into a sentence-level representation during comprehension. If reading rates of literally biased disambiguating regions that follow an idiom are slower than literally biased disambiguating regions that precede an idiom, but the same difference does not hold for idiomatically biased disambiguating regions, we may conclude that the idiomatic meaning is available to be integrated during comprehension. Alternatively, if reading rates of idiomatically biased disambiguating regions that follow an idiom are slower than those that precede an idiom, but this difference is not found for literally biased disambiguating regions, we may conclude that the literal meaning of the phrase was available during comprehension. These conclusions rest on the assumption that reading rates of subsequent disambiguating regions that are inconsistent with the meaning integrated in the preceding phrase reflect the time required for a revised interpretation of the preceding phrase. To the extent that semantic commitments are delayed during idiom processing (i.e., activation of both meanings is maintained and integration of a particular meaning is delayed), reading rates of disambiguating regions following idiomatic expressions should not differ as a function of whether they are idiomatically or literally biased. All experimental sentences were counterbalanced across conditions and presented to 24 undergraduates at the State University of New York at Binghamton, using a microcomputer. Eye movements were recorded using a SRI Dual Purkinje Eyetracker, and the dependent variable was reading rate per character based on gaze durations (i.e., first fixation durations and intra-word regressions) for the ambiguous idiom region and the disambiguating region. A 2 (idiom type: nondecomposable and decomposable) x 2 (context position: preceding and following) x 2 (contextual bias: idiomatic and literal) within-subject design ANOVA was conducted for both reading rate of the idiom region and reading rate of the disambiguating region. The results of each of these separate analyses will be discussed in turn. The results for the idiom region showed that reading rates of nondecomposable idioms were significantly slower when context (either of an idiomatic or literal bias) preceded the idiom than when context followed the idiom (65 vs. 50 msec per character, respectively). In contrast, reading rates of decomposable idioms showed no difference in the idiom region of the sentence as a function the location of the biasing context (52 vs. 52 msec per character, respectively; F(1,23)=15.9, M S e = I 6 7 , p<.01). These results suggest that activation/generation of idiomatic and literal meanings of idiomatic phrases is mandatory for idiomatic expressions regardless of contextual bias. Because nondecomposable idioms have semantically distinct idiomatic and literal phrasal meanings, it takes longer for readers to integrate a contextually appropriate alternative. In contrast, because decomposable idioms have semantically related idiomatic and literal meanings (i.e., the products of a literal analysis contribute to both interpretations), there is no processing cost associated with integrating a contextually appropriate meaning. 1670 D.A. Titone, C.M. Connine / Journal of Pragmatics 31 (1999) 1655-1674 The results for the disambiguating region showed a nonsignificant trend for idiomatically biased disambiguating regions, collapsed across idiom type and context position, to be read more quickly than literally biased disambiguating regions (44 vs. 47 msec per character, respectively; F( 1,23)=4, MSe=I 11, p=.07). Inspection of the data suggests that this effect is mostly due to the difference between disambiguating regions following the idiom (44 vs. 49 msec per character). Although these results hint that for both nondecomposable and decomposable idioms, only the more frequent and conventional idiomatic interpretation gets integrated into a higher level representation of the sentence, the fact that this effect was not statistically robust suggests that literal phrasal meanings may be integrated instead of idiomatic meanings on some occasions (consequently reducing the difference in reading rates between literally and idiomatically biased disambiguating regions). Taken together, the results from reading rates of the idiom region and the disambiguating region are consistent with a model of idiom processing in which idiomatic meanings are directly retrieved, and literal analysis of the phrase is carried out. Idiomatic and literal meanings of idioms are computed during comprehension regardless of whether these phrases are preceded by contextual information biased toward a particular interpretation. This is most clearly shown for nondecomposable idioms in which there is an elevation of reading rate when these idiom types are preceded by a context that biases either the idiomatic or literal meaning. For decomposable idioms, although there is no elevation in reading rate for the preceding context conditions, we may infer that both meanings are activated given that there is no difference in reading rate of the disambiguating region for a particular interpretation. That is, if only one meaning of the idiom had been accessed during comprehension, there would have been an elevation in reading rates for disambiguating regions favoring the unpreferred alternative. This was not the case. We may also conclude from these results that there is a cost associated with integrating one meaning over another into a biased context for nondecomposable idioms for which idiomatic and literal interpretations are semantically distinct, but not for decomposable idioms for which alternative interpretations are semantically related. This result suggests that direct retrieval of idiomatic meaning does not solely drive apprehension of the idiomatic meaning during comprehension. When literal word meanings simultaneously contribute to idiomatic and literal interpretations of the phrase, selection between idiomatic and literal meanings is faster than when literal word meanings only contribute to a literal interpretation of the phrase. 5. Conclusion In this paper, we have argued for a view of idiomaticity in which idioms are processed simultaneously as noncompositional and compositional word sequences. To this end, we adopt the idiom classification scheme put forth by Nunberg et al. (1994) in which idioms are defined with reference to three semantic dimensions: conventionality, compositionality, and transparency. Based on this view, idioms behave noncompositionally to the extent that their figurative meanings are conven- D.A. Titone. C.M. Connine / Journal of Pragmatics 31 (1999) 1655-1674 1671 tional. Regardless of whether an idiom's component words contribute to idiomatic meaning in an explicit way or not, an idiomatic expression is a highly overlearned sequence of words that is consistently associated with a specific phrasal meaning. This highly learned semantic relationship between word configurations and figurative meanings permits direct recovery of idiomatic meanings (relative to construction of literal phrasal meanings) as a function of the frequency with which an idiom is encountered in the language, often prior to the offset of the phrase (Titone and Connine, 1994a). In addition to their conventionality-based noncompositional nature, idioms behave compositionally (i.e., are built out of word components like nonidiomatic phrases) to the degree that they are decomposable (i.e., possess individual word meanings that directly map to components of the idiomatic meaning) and transparent (Nunberg et al., 1994). Given that word meanings are always activated during idiom processing (e.g., Cacciari and Tabossi, 1988; Titone and Connine, 1994a), component words of idiomatic sequences may contribute substantially to the construction of idiomatic meanings (as they would for inherently compositional or transparent idiomatic combinations) or a minimal semantic contribution to the construction of idiomatic meanings (as they would for less compositional or transparent idiomatic phrases). Therefore, during idiom processing, the idiomatic meaning is directly retrieved when a sufficient portion of the idiom is encountered (i.e., at the idiomatic key), and a literal analysis of the phrase is carded out (e.g., word meanings are activated and compositionally combined). Based upon our eye tracking results, if the products of a literal analysis overlap with the idiomatic meaning (i.e., as they do for decomposable idioms), interpretation is facilitated, whereas if the products of a literal analysis are distinct from the idiomatic meaning (i.e., as they do for nondecomposable idioms) interpretation is more difficult. A hybrid view of idiomaticity is consistent with investigations of idiom processing using other linguistically interesting subject populations. Although it has been routinely thought that the left cerebral hemisphere is responsible for language ability, a good deal of research has indicated that the right cerebral hemisphere also contributes to language ability (Beeman, 1993; Chiarello, 1990). Specifically, right hemisphere (RH) damaged patients demonstrate impairments in many areas of semantic/pragmatic processing that are relatively spared with left hemisphere (LH) damage. RH, but not LH damaged patients are impaired in processing connotative aspects of word meaning (Brownell, 1988; Brownell et al., 1984), and elements of nonliteral language such as metaphor (e.g., Winner and Gardner, 1977), indirect requests (Foldi, 1987; Hirst et a1.,1984), and sarcasm (Kaplan et al., 1990). In the domain of idiom processing, Van Lancker and Kempler (1987; see also Myers and Linebaugh, 1988) found that RH damaged individuals were impaired relative to LH damaged patients at selecting one of four line drawings that corresponded to an idiomatic interpretation for a set of idioms (LH damaged patients, however, performed worse than RH damaged patients on selecting line drawings on the basis of novel literal phrases). Interestingly enough, although RH damaged patients were unable to correctly interpret idioms in the picture selection task, a word monitoring experiment by 1672 D.A. Titone, C.M. Connine / Journal of Pragmatics 31 (1999) 1655-1674 Tompkins et al. (1992) found that R H and LH damaged patients did not differ from control subjects in showing priming for idiomatic expressions. This is not to say that the idiomatic meaning was fully retrieved for the RH damaged patients, however, because facilitation in the word monitoring task m a y be explained by an intact sensitivity to co-occurrence frequencies of words in the language. It is possible that if a task had been used in which facilitating effects required apprehension o f the idiomatic meaning (e.g., cross modal priming using lexical decision responses of semantically related targets), priming for RH damaged patients m a y not have been obtained. Therefore, the results from this experiment suggest, minimally, that some sensitivity to idiomatic configurations remains intact with RH damage despite the fact that idiomatic meanings are not integrated into a phrase-level interpretation. To conclude, we have presented a view o f idiomaticity that supports noncompositional and compositional approaches to idiom representation and processing. In addition to supporting the existing data on idiom processing, a model that incorporates the word-like aspects as well as the metaphorical and compositionally derived aspects of idioms is likely to be more generalizable across m a n y classes of idiomatic and other nonliteral sequences. The advantage of such a model is that it accounts for the similarities and differences in representation and processing of many classes of idiomatic sequences as well as other nonliteral sequences. References Beeman, Mark, 1993. Semantic processing in the right hemisphere may contribute to drawing inferences from discourse. Brain and Language 44: 80-120. Blasko, Dawn G. and Cynthia M. Connine, 1993. Effects of familiarity and aptness on the processing of metaphor. Journal of Experimental Psychology: Learning, Memory, and Cognition 19: 295-308. Bobrow, Samuel A. and Susan M. Bell, 1973. On catching on to idiomatic expressions. Memory and Cognition 1: 343-346. Brownell, Hiram H., 1988. Appreciation of metaphoric and connotative word meaning in brain damaged patients. In: C. Chiarello, ed., Right hemisphere contributions to lexical semantics. New York: Springer. Brownell, Hiram H., Heather H. Potter, Diane Michelow and Howard Gardner, 1984. Sensitivity to lexical denotation and connotation in brain damaged patients: A double dissociation? Brain and Language 22: 253-265. Cacciari, Cristina, 1993. The place of idioms in a literal and metaphorical world. In: C. Cacciari and P. Tabossi, eds., Idioms: Processing, structure, and interpretation, 27-52. Hillsdale, NJ: Erlbaum. Cacciari, Cristina and Sam Glucksberg, 1991. Understanding idiomatic expressions. Memory and Cognition 1: 343-346. Cacciari, Cristina and Patrizia Tabossi, 1988. The comprehension of idioms. Journal of Memory and Language 27: 668-683. Chiarello, Christine, 1990. Interpretation of word meanings by the cerebral hemispheres: One is not enough. In: Paula J. Schwanenflugel, ed., The psychology of word meanings, 251-273. Hillsdale, NJ: Erlbaum. Chomsky, Noam, 1980. Rules and representations. New York: Columbia University Press. Colombo, Lucia, 1993. The comprehension of ambiguous idioms in context. In: C. Cacciari and P, Tabossi, eds., Idioms: Processing, structure and interpretation. Hillsdale, NJ: Erlbaum. Connine, Cynthia M., John Mullinnex, Eve Sheruoff and Jennifer Yelen, 1990. Word familiarity and frequency in auditory and visual word recognition. Journal of Experimental Psychology: Learning, Memory and Cognition 16: 1084-1096. D.A. Titone, C.M. Connine / Journal of Pragmatics 31 (1999) 1655-1674 1673 Cronk, Brian C. and Wendy A. Schweigert, 1992. The comprehension of idioms: The effects of familiarity, literalness and usage. Applied Psycholinguistics 13: 131-146. Estill, Robert B. and Susan Kemper, 1982. Interpreting idioms. Journal of Psycholinguistic Research 11 : 559-568. Foldi, Nancy S., 1987. Appreciation of pragmatic interpretations of indirect commands: Comparison of right and left-hemisphere brain-damaged patients. Brain and Language 31, 88-108. Fraser, Bruce, 1970. Idioms within a transformational grammar. Foundations of Language 6: 22-42. Frazier, Lynn and Keith Rayner, 1990. Taking on semantic commitments: Processing multiple meanings vs. multiple senses. Journal of Memory and Language 29: 181-200. Gernsbacher, Morton A., 1984. Resolving 20 years of inconsistent interactions between lexical familiarity and orthography, concreteness and polysemy. Journal of Experimental Psychology: General 113: 256--281. Gibbs Jr., Raymond W., 1980. Spilling the beans on understanding and memory for idioms in conversation. Memory and Cognition 8: 149-156. Gibbs Jr., Raymond W., 1986. Skating on thin ice: Literal meaning and understanding idioms in conversation. Discourse Processes 9: 17-30. Gibbs Jr., Raymond W., 1992. What do idioms really mean? Journal of Memory and Language 31: 485-506. Gibbs Jr., Raymond W. and Nadini Nayak, 1989. Psycholinguistic studies on the syntactic behavior of idioms. Cognitive Psychology 21: 100-138. Gibbs Jr., Raymond W., Nadini Nayak, John Bolton and Melissa Keppel, 1989a. Speakers' assumptions about the lexical flexibility of idioms. Memory and Cognition 17: 58-68. Gibbs Jr,, Raymond W., Nadini Nayak and Cooper Cutting, 1989b. How to kick the bucket and not decompose: Analyzability and idiom processing. Journal of Memory and Language 28: 576-593. Gibbs, Jr., Raymond W., and Jennifer O'Brien, 1989. Idioms and mental imagery: The metaphorical motivation for idiomatic meaning. Cognition 36: 35-68. Glass, Arnold L., 1983. The comprehension of idioms. Journal of Psycholinguistic Research 12: 429--442. Glucksberg, Sam, 1991. Beyond literal meanings: The psychology of allusion. Psychological Science 2: 146-152. Glucksberg, Sam, 1993. Idiom meanings and allusional context. In: C. Cacciari and P. Tabossi, eds., Idioms: Processing, structure, and interpretation, 3-27. Hillsdale, NJ: Edbaum. Glucksberg, Sam, Mary Brown and Matthew McGlone, 1993. Conceptual metaphors are not automatically accessed during idiom comprehension. Memory and Cognition 21:711-719. Hirst, William, Joseph LeDoux and Susanna Stein, 1984. Constraints on the processing of indirect speech acts: Evidence from aphasiology. Brain and Language 23, 26-33. Kaplan, Joan A., Hiram H. Brownell, Janet Jacobs and Howard Gardner, 1990. The effects of right hemisphere damage on conversational remarks. Brain and Language 38, 315-333 Lakoff, George, 1987. Women, fire and dangerous things. Chicago, IL: University of Chicago press. McGlone, Matthew, Sam Glucksberg and Cristina Cacciari, 1994. Semantic productivity and idiom comprehension. Discourse Processes 17: 167-190. Myers, P.S. and Craig W. Linebaugh, 1981. Comprehension of idiomatic expressions by fight-hemisphere-damaged adults. In: R.H. Brookshire, ed., Clinical aphasiology: Conference proceedings, 254-261. Minneapolis: BRK. Nayak, Nadini and Raymond Gibbs Jr.. 1990. Conceptual knowledge in the interpretation of idioms. Journal of Experimental Psychology: General 3: 315-330. Newmeyer, F., 1972. The insertion of idioms. In: Papers from the eighth regional meeting of the Chicago Linguistic Society. Chicago, IL: Chicago Linguistic society. Nunberg, Geoffrey, 1978. The pragmatics of reference. Bloomington, IN: Indiana University Linguistics. Nunberg, Geoffrey, Ivan A. Sag and Thomas Wasow, 1994. Idioms. Language 70: 491-534. Nusbaum, Howard, David B. Pisoni and C.K. Davis, 1984. Sizing up the Hoosier mental lexicon: Measuring the familiarity of 20,000 words. Research on speech perception, progress report 10. Bloomington, IN: Indiana University Speech Research Laboratory. 1674 D.A. Titone, C.M. Connine / Journal of Pragmatics 31 (1999) 1655-1674 Ortony, Andrew, Diane L. Schallert, Ralph E. Reynolds and Stephen J. Antos, 1978. Interpreting metaphors and idioms: Some effects of context on comprehension. Journal of Verbal Learning and Verbal Behavior 17: 465-477. Peterson, Robert R. and Curt Burgess, 1993. Syntactic and semantic processing during idiom comprehension: Neurolinguistic and psycholinguistic dissociations. In: C. Cacciari and P. Tabossi, eds., Idioms: Processing, structure, and interpretation, 201-225. Hillsdale, NJ: Erlbaum. Peterson, Robert R., Curt Burgess, Gary S. Dell and Kathleen Eberhard, 1989. Dissociation of syntactic and semantic analyses during idiom processing. Paper presented at the second annual CUNY conference on human sentence processing, New York. Sag, Ivan A., 1976. Deletion and logical form. Cambridge, MA: M1T dissertation. (Published, New York: Garland, 1980.) Schweigert, Wendy A., 1986. The comprehension of familiar and less familiar idioms. Journal of Psycholinguistic Research 15: 33-45. Schweigert, Wendy A. and Danny R. Moates, 1988. Familiar idiom comprehension. Journal of Psycholinguistic Research 17: 281-296. Searle, John R., 1979. Metaphor. In: A. Ortony, ed., Metaphor and thought, 92-123. Cambridge: Cambridge University Press. Swinney, David A. and Anne Cutler, 1979. The access and processing of idiomatic expressions. Journal of Verbal Learning and Verbal Behavior 18: 522-534. Titone, Debra A. and Cynthia M. Connine, 1994a. The comprehension of idiomatic expressions: Effects of predictability and literality. Journal of Experimental Psychology: Learning, Memory, and Cognition 20:1126--1138 Titone, Debra A. and Cynthia Connine, 1994b. Descriptive norms for 171 idiomatic expressions: Familiarity, compositionality, predictability, and literality. Metaphor and Symbolic Activity 9: 247-270. Titone, Debra A. and Cynthia M. Connine, 1994c. Semantic commitments in the processing of idiomatic expressions: Semantically decomposable and nondecomposable idioms. Paper presented to the CUNY Conference on Human Sentence Processing, New York. Tompkins, Connie A., Richard Boada and Katherine McGarry, 1992. The access and processing of familiar idioms by brain-damaged and normally aging adults. Journal of Speech and Hearing Research 35: 6264537. van der Linden, E., 1992. Incremental processing and the hierarchical lexicon. Computational Linguistics 18: 219-238. Van Lancker, Diana R. and Daniel Kempler, 1987. Comprehension of familiar phrases by left- but not right-hemisphere damaged patients. Brain and Language 32, 265-277. Wasow, Thomas, Ivan Sag and Geoffrey Nunberg, 1983. Idioms: An interim report. In: S. Hattori and K. Inoue, eds., Proceedings of the XIIIth International Congress of Linguistics, Tokyo. Williams, E., 1977. Discourse and logical form. Linguistic Inquiry 8: 101-140. Winner, Ellen and Howard Gardner, 1977. The comprehension of metaphor in brain-damaged patients. Brain 100, 717-729. Cynthia Connine is an Associate Professor of Psychology and Linguistics at the State University of New York at Binghamton. She received her Ph.D. in Cognitive Psychology from the University of Massachusetts at Amherst. Her current research interests include models of spoken word recognition and figurative language processing. Debra Titone received a Ph.D. in Experimental Psychology at the State University of New York at Binghamton, completed a postdoctoral fellowship at Brandeis University, and is currently a research fellow in the Department of Psychiatry at Harvard Medical School. Her research interests include semantic, pragmatic, and biological approaches to figurative and contextualized language understanding. She is presently investigating language-processing deficits in schizophrenia at the Psychology Research Laboratory at McLean Hospital.