LONG-DISTANCE RELATIVIZATION IN VARIETIES OF DUTCH Master Thesis by Eefje Boef, 0332704 Linguistics: The Study of the Language Faculty Utrecht Institute of Linguistics OTS Utrecht University Thesis Supervisor: Prof.dr. Sjef Barbiers Second Reader: Prof.dr. Norbert Corver July 2008 0 TABLE OF CONTENTS General introduction .............................................................................................................................3 Chapter 1 Theoretical background ......................................................................................................5 1.1 Introduction ...................................................................................................................................5 1.2 The syntax of relativization ...........................................................................................................5 1.2.1 The Head External Analysis of relative clauses .....................................................................5 1.2.2 The Head Raising Analysis of relative clauses ......................................................................6 1.2.3 The Matching Analysis of relative clauses .............................................................................8 1.3 Relativization in Standard Dutch .................................................................................................10 1.3.1 The framework .....................................................................................................................10 1.3.2 Properties of Dutch restrictive relative clauses ....................................................................11 1.3.3 An analysis of Dutch relative clauses ...................................................................................12 1.3.3.1 De Vries (2002) .................................................................................................................13 1.3.3.2 Long-distance relativization in Dutch ...............................................................................14 1.3.4 The investigation of the variation in long-distance relativization in Dutch .........................16 1.4 Summary .....................................................................................................................................17 Chapter 2 Presentation of the SAND data on long-distance relativization ....................................18 2.1 Introduction .................................................................................................................................18 2.2 Data collection – the SAND project ............................................................................................18 2.2.1 The Syntactic Atlas of the Dutch Dialects (SAND) .............................................................18 2.2.2 Method of data collection ......................................................................................................18 2.3 The data .......................................................................................................................................19 2.3.1 Data on long-distance relativization – long A’-movement without resumptive pronoun .....23 2.3.2 Data on long-distance relativization – long A’-movement with resumptive pronoun ..........25 2.3.3 Data on long-distance relativization – PP-relatives...............................................................26 2.4 Summary .....................................................................................................................................27 Chapter 3 Long-distance relativization in Dutch dialects – subject/object asymmetries .............29 3.1 Introduction .................................................................................................................................29 3.2 Presentation of the data ...............................................................................................................29 3.3 Theoretical background ...............................................................................................................32 3.3.1 The Empty Category Principle (ECP) ...................................................................................32 3.3.2 Mayr (to appear) ...................................................................................................................34 3.4 The analysis – subject/object asymmetries in varieties of Dutch ................................................38 3.4.1 Some notes on the status of die and dat ...............................................................................38 3.4.2 The analysis ...........................................................................................................................39 3.4.3 Predictions ............................................................................................................................41 3.4.4 Some problems – special properties of long subject relativization ......................................45 3.4.5 Intermediate summary ..........................................................................................................46 3.5 Alternative accounts ....................................................................................................................48 3.5.1 Pesetsky & Torrego (2001), Gallego (2004) ........................................................................48 3.5.2 Rizzi & Shlonsky (2007) ......................................................................................................52 3.5.3 Alternative accounts – summary ..........................................................................................54 3.6 Summary .....................................................................................................................................55 Chapter 4 Long-distance relativization in Dutch dialects – no subject/object asymmetries ........56 4.1 Introduction .................................................................................................................................56 4.2 Presentation of the data ...............................................................................................................56 4.3 The analysis – no subject/object asymmetries in varieties of Dutch ...........................................57 4.3.1 Absence/presence of relative pronouns ................................................................................57 4.3.2 Syntactic doubling ................................................................................................................58 1 4.3.2.1 Phonetic realization of multiple copies – Nunes (2004) ...................................................59 4.3.2.2 Reduced copies – Van Koppen (2007) ..............................................................................60 4.3.2.3 Partial copying – Barbiers, Koeneman & Lekakou (2008) ...............................................64 4.3.2.4 Intermediate summary .......................................................................................................69 4.3.2.5 Predictions ..........................................................................................................................69 4.4 Summary .....................................................................................................................................72 Chapter 5 Conclusion and topics for future research ......................................................................73 5.1 Introduction .................................................................................................................................73 5.2 Conclusion ...................................................................................................................................73 5.3 Topics for future research ............................................................................................................74 5.3.1 PP-relatives ...........................................................................................................................74 5.3.2 Resumptive pronouns ............................................................................................................79 References ............................................................................................................................................82 Appendices ...........................................................................................................................................85 Appendix I The SAND data on relativization ...................................................................................85 Appendix II The SAND data on short relativization – geographic distribution ................................94 2 GENERAL INTRODUCTION In this thesis I will discuss the syntax of long-distance relativization structures in varieties of Dutch. The empirical basis for this study is formed by the SAND data on relativization, which show that there exists a considerable amount of variation regarding long-distance relativization in the Dutch speaking language area. The aim of this thesis is to provide a description and a (unified) analysis of this microvariation. Restrictive relative clauses A restrictive relative clause is a subordinate clause that modifies a noun – the head of the relative clause.1 The relativized constituent in the matrix clause is semantically and syntactically related to the gap inside the relative clause, in other words, it has a pivot function (cf. De Vries 2002). In the following English sentence, the relativized constituent the man is modified by the relative clause who/that I saw yesterday, which is introduced by the relative pronoun who or the complementizer that. Furthermore, the relative head noun (the) man functions as a pivot as it is syntactically the subject in the matrix clause and the object in the relative clause. Notice that semantically, the man has two different roles as well. (1) The man [RC who/that I saw __ yesterday] ran away. In Dutch, the relative clause always follows the noun it modifies (head-initial/ postnominal relative clauses), and, in contrast to English, the restrictive relative clause always needs to be introduced by a relative pronoun. It thus follows that relative pronouns in Dutch are always sentence initial. Given this observation and the observation that cross-linguistically relative pronouns are often homophonous to interrogative pronouns (cf. De Vries 2002, Bhat 2004), since Chomsky (1977), it is assumed that the syntax of wh-clauses and relative clauses is related. More in particular, relative pronouns are traditionally taken to be wh-moved. We can formulate these properties of Dutch restrictive relative clauses in a more abstract manner, as in (2). The co-indexing of the DP and the gap in the relative clause merely functions to indicate that the two are syntactically and semantically related. (2) [MATRIX SENTENCE … DPi [RC relative pronoun … gapi …]] The topic of this thesis Similar to wh-phrases, relative constituents can be extracted from embedded clauses, giving rise to socalled long-distance relativization, as illustrated by (3), and abstractly exemplified in (4) (3) Dat is de man [die ik denk [dat __ het verhaal verteld heeft]] that is the man die I think that the story told has ‘That is the man who I think told the story.’ (4) [MATRIX SENTENCE … DPi [RC relative pronoun … [finite embedded clause complementizer … gapi …]]] Inspection of the SAND data on long-distance relativization shows that there is a considerable amount of variation regarding this construction.2 More specifically, the Dutch dialects show variation along the following four parameters: (i) the form of the element that introduces the relative clause (die/dat), (ii) the form of the element that introduces the most deeply embedded clause (die/dat), (iii) the presence/absence of a complementizer, and (iv) the presence/absence of an overt subject/object – resumptive pronoun – at the extraction site. In addition to this variation, as illustrated in (5), many dialects make use of an alternative strategy in which the relative clause is introduced by a 1 With respect to the semantics of restrictive relative clauses, it is standardly assumed that their meaning is acquired by set intersection of the meaning of the head noun and the RC (cf. Grosu & Landman 1998). 2 This thesis restricts its attention to Dutch headed restrictive relative clauses that require die in Standard Dutch: in all (test) sentences the antecedent is 3rd person non-neuter (neuter singular antecedents require dat, which is identical in form to the Dutch finite complementizer). 3 prepositional phrase – so-called PP-relatives (6). Such constructions differ from the ones in (5) in that a prepositional phrase is relativized in the higher clause and the subject/object in the lower clause is retained. These structures also show variation with respect to the presence of resumptive pronouns, i.e. whereas in Standard Dutch the relativized constituent needs to be related to a resumptive pronoun located at the extraction site, some dialects allow the head noun to be related to a gap. (5) … DP [RC die/dat (dat) … [finite embedded clause die/dat (dat) … gap/resumptive pronoun …]] (6) Dat is de man [waarvan that is the man whereof ik denk [dat hij/die het verhaal verteld heeft]] I think that he/die the story told has This wealth of variation immediately raises the question of how to account for the attested variation in long-distance relativization. It seems natural to treat the data in a uniform way, and therefore, I will pursue the default hypothesis that all the variants of long-distance relativization found in the Dutch speaking language area – abstracting away from PP-relatives – have the same underlying structure. More in particular, I will assume a raising analysis of relative clauses (cf. Kayne 1994, Zwart 2000, De Vries 2002) according to which relative clauses are derived by successive cyclic wh-movement of the relativized constituent consisting of the head noun and the relative pronoun. I will thus argue that there is a common syntactic basis to all attested variants of long-distance relativization, and I will show that four micro-parameters can account for the attested differences. The framework The discussion of the long-distance relativization structures should be situated in the syntactic framework of Minimalism (cf. Chomsky 1995,2000,2001), and Kayne’s (1994) antisymmetry hypothesis. In addition to these assumptions about the derivational component, I will follow Van Koppen (2005) and further assume that this view on Syntax should be combined with Distributed Morphology (cf. Halle & Marantz 1993), according to which Syntax only operates on roots and feature bundles. After the syntactic component finishes the derivation, at the level of Morphology, the abstract feature bundles are replaced by Vocabulary Items. Outline of the thesis In the first chapter of this thesis, I will provide an overview of the literature on the syntax (and semantics) of restrictive relative clauses. Moreover, in this chapter, I will lay out my assumptions regarding the syntactic and morphological component that I will assume throughout this thesis, and I will introduce the analysis of long-distance relative clauses that I take to be the common syntactic basis to all the different variants of relativization structures that are attested in varieties of Dutch. In chapter 2, I will present the SAND data on relativization that form the empirical basis for this thesis. On the basis of three generalizations, I try to get some order in the wealth of variation shown by the SAND data. This thesis primarily focuses on six patterns of long-distance relativization that are not PP-relatives and in which no resumptive pronoun is present. Chapter 3 provides an analysis for two patterns of long-distance relativization that show subject/object asymmetries. I show that the two systems that are discussed in this chapter can be derived by two micro-parameters: the presence/absence of relative pronouns (micro-parameter 1) and the spell out/non-spell out of a φ-agreement relation between the most deeply embedded C0 and (the copy of) the subject relative DP in its specifier position (micro-parameter 2). These two parameters predict the existence of (at least) 4 different systems of long-distance relativization which are all attested in Dutch. Chapter 4 basically deals with patterns of relativization in which no subject/object asymmetries are displayed. Most prominently in this chapter are patterns that show doubling of the relative marker. On the basis of the assumption that dialects differ with respect to whether they allow multiple chain links to be spelled out (micro-parameter 3), and the assumption that dialects differ with respect to which part of the constituent the mechanism copy targets – either the XP or a subpart of this XP – (micro-parameter 4), the final two systems of long-distance relativization are derived. Finally, chapter 5 summarizes the main conclusions of this thesis and suggests some points for further research. 4 CHAPTER 1 Theoretical background 1.1 Introduction This chapter consists of two parts. In the first part of this chapter, I will give a short state of the art overview of the literature on the syntax (and semantics) of restrictive relative clauses. 3 As it is not possible within the scope of this thesis to present a complete overview of the existing literature on the syntax of relativization4, I will only focus on the three most influential analyses of restrictive relative clauses: the head external analysis, the head raising analysis and the matching analysis. I will briefly discuss their main properties and some of the problems they face. For a more in-depth study of the fine-grained differences between the various proposals, see amongst others Bianchi (1999), Alexiadou et al. (2000), De Vries (2002), and Salzmann (2006). The second part of this chapter consists of the presentation of the theoretical framework I will adopt throughout this thesis, i.e. it will lay out my assumptions regarding the computational system and its interaction with Phonology/Morphology. Moreover, this part of the chapter provides a detailed description of the analysis I will assume for restrictive long-distance relative clauses in Standard Dutch. I will pursue the default hypothesis that this analysis forms the common syntactic basis for all the attested variants of long-distance relativization in varieties of Dutch.5 1.2 The syntax of relativization In this section the three main analyses of restrictive relative clauses (henceforth RCs) will be discussed: the head external analysis, the head raising analysis and the matching analysis. It will be shown that these different approaches to the syntax of relativization differ with respect to (i) the origin of the head noun (internal or external to the RC), (ii) the relation between the relative head noun and the RC (adjunction or complementation), and (iii) the way in which the external relative head noun is related to the element in Spec,CP (by means of predication, ellipsis or movement). 1.2.1 The Head External Analysis of relative clauses Within the Government and Binding framework (Chomsky 1981,1982,1986), the Head External Analysis (HEA) of relative clauses is the most widely accepted analysis.6 This approach essentially assumes that the head noun of a RC is base-generated outside that clause – the base-generated head hypothesis – and that RCs are (right) adjoined to the relative head NP – the adjunction hypothesis.7 This NP is selected by an external determiner. Within the RC there is wh-movement of an empty operator or a relative pronoun to Spec,CP. This operator or pronoun is in turn linked to the relative head outside the CP, by means of predication. This is exemplified in (1). (1) … the picturei [CP [whichi/OPi]k he likes tk] The HEA was taken to be the standard analysis of RCs until Kayne’s (1994) antisymmetry hypothesis was widely adopted: the HEA is at odds with the antisymmetry hypothesis because it relies on 3 As this thesis is only concerned with restrictive relative clauses, I will not discuss the syntax (and semantics) of appositive relative clauses, free relatives, circumnominal relative clauses and correlatives; see De Vries (2002; chapter 2) for a comprehensive typological overview. 4 There exists an extensive literature on the syntax of relativization, see e.g. the bibliography compiled by De Vries (2002), and Alexiadou et al. (2000). For an overview and a treatment of the semantics of (different types of) relative clauses, see Grosu & Landman (1998). 5 In this chapter I will not be concerned with (the syntax of) PP-relatives. For a detailed description and analysis of these constructions in German, Dutch (and Zurich German) see Salzmann (2006). In chapter 5, I briefly discuss this analysis of PP-relatives. 6 This analysis is very prominent in the literature and it is argued for by many scholars; see, amongst others, Ross (1967), Chomsky (1977), Chomsky & Lasnik (1977), Maling & Zaenen (1978) and Borsley (1997). For a head external analysis of relative clauses in (dialects of) Dutch, see Bennis (1983), Haegeman (1983), Bennis & Haegeman (1984), Van der Auwera (1984). 7 It has also been argued that restrictive relative clauses are not adjuncts of the head noun, but are rather complements of the head noun (e.g. Platzack 2000). The HEA together with this revision is called the ‘revised standard theory’ of restrictive relativization by De Vries (2002:73). 5 rightward adjunction.8 Moreover, the presence of reconstruction effects is hard to account for under the HEA as there is no direct movement relation between the element in Spec,CP and the external relative head NP. To illustrate this, consider reconstruction for variable binding in a RC as given in (2). The relative head containing the pronoun his needs to be interpreted in the scope of the universal quantifier every man in order to be bound. In order words, the relative head needs to undergo reconstruction into the RC to be c-commanded by the coindexed quantifier. (2) … the [picture of hisi girlfriend] that every mani likes __ best [Salzmann 2006:22] The HEA cannot easily capture such reconstruction effects, i.e. somehow the relative operator needs to act as a mediator for reconstruction. Moreover, given that reconstruction should be seen as the interpretation of a lower copy in a syntactic chain – according to the Copy Theory of Movement (Chomsky 1993) – the reconstruction effects are impossible to account for under the HEA. As will become clear in the following section, the head raising analysis can straightforwardly account for the reconstruction data, as it assumes that the relative head NP is directly represented inside the RC. 1.2.2 The Head Raising Analysis of relative clauses The core property of the Head Raising Analysis (HRA) of relative clauses is the assumption that the relative NP is base-generated inside the RC and raises towards the matrix clause (i.e. to an operator position within the RC) to become the relative head noun. Moreover, it is assumed that an external determiner selects a CP as its complement – the D-complement hypothesis, as originally formulated by Smith (1964) and more recently reinstated by Kayne (1994). The D-complement hypothesis together with the raising analysis of RCs, is called the promotion theory (e.g. De Vries 2002), and is illustrated in (3). (3) … the [CP [picturei which/OP ti]k he likes tk] Although this is the most common implementation of the HRA (cf. Kayne 1994, De Vries 2002), there are several alternatives. As noted by Citko (2001), analyses differ with respect to (i) the assumption about the nature of the relation between the relative head noun and the RC (complementation or adjunction), (ii) the categorical status of the relative element undergoing movement (NP or DP), and (iii) the landing site of the moved relative element (CP internal or CP external). In the remainder of this section, I will not discuss all the details of the different existing implementations of the HRA, instead, in what follows, I will mention some of the most important arguments in favour of a HRA in general. Some of the main advantages of the HRA over the HEA are that the pivot function of the head noun follows without stipulations, and that, as already mentioned in section 1.2.1, the HRA straightforwardly accounts for reconstruction effects. Moreover, it is often mentioned in the literature that collocations/idioms (fixed verb-object pairs) can be split across a RC (cf. De Vries 2002:78 and references cited there), as illustrated by (4). (4) The headway [RC we made __ ], was great. Given that the verb selects for the object, the grammaticality of sentences like (4) follows straightforwardly under the HRA because at one point in the derivation the object and the verb were adjacent, i.e. the object originated RC internally adjacent to the verb. Another argument in favour of the HRA is mentioned by Bianchi (1999) and De Vries (2002). They argue that ‘internally headed relative clauses’ as in (5), in which the relative head appears in the argument position inside the RC, provide strong evidence for the HRA. That is to say, under the HRA such RCs can receive a similar analysis as externally headed RCs – the only assumption that needs to be made is that languages may differ with respect to the ‘level’ at which the relative head moves Some main consequences of Kayne’s (1994) antisymmetry hypothesis are that rightward movement, rightward adjunction and multiple adjunction/specifiers are disallowed. 8 6 (overtly or covertly at LF) – whereas a HEA needs a completely different structure for these particular RCs. (5) [[Mari [owiza wa] kage] ki] he ophewathu. Mari quilt a make the Dem I buy ‘I bought the quilt that Mari made.’ [Lakhota]9 Besides these arguments in favour of the HRA,10 there are also some unsatisfactory aspects of the analysis. First, as argued by Borsley (1997), all versions of the HRA assume that the relative head noun, that is base-generated as the complement of the relative pronoun, moves to either the specifier of the relative pronoun (cf. De Vries 2002), or to the specifier of a higher head (cf. Bianchi 1999,2000, Zwart 2000). The trigger for this particular movement operation is not entirely clear, however, and different analyses have been proposed. Salzmann (2006) argues that although some of these proposals solve the problem, such solutions all have an unwarranted construction-specific flavour; moreover, this problem is not encountered with the matching analysis of relative clauses. A second, rather famous argument against the HRA analysis (cf. Borsley 1997, Alexiadou et al. 2000, Citko 2001, De Vries 2002) is formed by examples like the one in (6). (6) Widzialem tego pana saw-1SG this-ACC man-ACC ‘I saw the man who broke your glass.’ który which-NOM zbil ci szybę. broke you glass [Polish; Borsley 1997:638] This example shows that there is a mismatch in Case between the relative head noun and the relative wh-pronoun. Under a HRA in which the phrase który pana ‘which man’ moves from the subject position to Spec,CP, and pana moves to the specifier position of który, this Case mismatch is unexpected. Several attempts have been made to account for this Case problem, but, according to Salzmann (2006), all these accounts are only descriptive, not explanatory (see section 1.3.3.1 for a solution to the Case mitmatch problem as proposed by De Vries 2002). 9 This example is taken from Bianchi (1999:62), but it was originally documented by Williamson (1987:171). Notice that most of the arguments in favour of the HRA have to do with evidence that shows the need for an internal representation of the external head inside the relative clause (cf. also the argument of subcategorization (cf. Larson 1985, Bhatt 2002), and evidence from amount relatives (cf. Grosu & Landman 1998)). Crucially, these arguments are often unable to distinguish between the HRA and a matching analysis of relative clauses (section 1.2.3). Bhatt (2002) provides an argument in favour of the HRA that is based on the observation that adjectival modifiers on the RC head can be interpreted in RC internal positions, suggesting that the RC head must have originated inside the RC. Consider sentence (i) and its two possible readings below 10 (i) ... the first book that John said that Tolstoy had written High reading (interpreting the highest CP-internal copy) In 1990, John said that Tolstoy had written Anna Karenina; in 1991, John said that Tolstoy had written War and Peace. Hence the NP is Anna Karenina. (i.e. order of saying matters, order of writing is irrelevant) Low reading (interpreting the lowest CP-internal copy) John said that the first book that Tolstoy had written was War and Peace. Hence the NP is War and Peace. (i.e. order of writing matters, order of saying is irrelevant) According to Bhatt, the low reading of the adjectival modifier first can only be derived under a HRA, not under a matching analysis (or HEA). That is to say, Bhatt assumes a semantics for reconstructed phrases that is based on Trace Conversion (Fox 1999), a mechanism that converts copies into definite descriptions. When this mechanism of interpreting reconstructed phrases is applied to the LF provided by the HRA, the low reading is generated. However, when Trace Conversion is applied to the LF provided by the matching analysis, Bhatt shows that the low reading cannot be generated because – given that the external head and the RC internal representation are not related by movement – the external head must always be interpreted in RC external position, and therefore, the structure provided by the matching analysis in combination with Trace Conversion does not give the intended interpretation (for the details of this analysis, see Bhatt 2002). 7 Another, very important argument against the HRA is formed by the ‘lack of Principle C effects’ (cf. Citko 2001, Salzmann 2006). The HRA predicts that configurations like (7) are ungrammatical on a coindexed reading, because the pronoun would c-command the name, yielding a Condition C violation.11 However, this prediction is not borne out, as illustrated by the grammaticality of (8). (7) [DP … namei …]j [CP … pronoun … tj] (8) The [picture of Johni] which hei likes __ is on the front page. [Citko 2001:139] Such examples show that the relative head cannot always reconstruct. A possible solution for these facts (under the HRA), is to assume that reconstruction in RCs is optional. Naturally, stating that reconstruction is optional is not a solution but just a description. That is to say, we need an account of the optionality of reconstruction. Safir (1999) assumes the workings of an additional mechanism called Vehicle Change (cf. Fiengo & May 1994) – a mechanism by which an R-expression can be turned into a pronoun in an ellipsis site – in order to account for the lack of Principle C effects.12 Alternatively, it has been argued, on the basis of these reconstruction effects, that languages may exhibit multiple analyses of RCs. Bhatt (2002) and Sauerland (2003), for example, argue for the HRA whenever there is reconstruction, but when there is no reconstruction, they assume a matching analysis; as we shall see in section 1.2.3. the lack of Principle C effects is easily captured under a matching analysis of RCs. A fourth argument against the HRA that has been put forward by Salzmann (2006:18) is the observation that the HRA violates a certain locality constraint on movement: the Condition on Extraction Domains. This constraint states that once a phrase has moved, no element from this phrase can be extracted. This is illustrated by the ill-formedness of (9). (9) *Whoj do you think [CP tj that [DP pictures of __j]k were painted __k]? However, movement of a larger phrase and subsequent subextraction from this phrase is exactly what is found under the HRA.13 Other arguments against the HRA have to do with Negative Polarity Item Licensing (Citko 2001), and the selectional properties of relative pronouns (cf. Aoun & Li 2003, Salzmann 2006). 1.2.3 The Matching Analysis of relative clauses The Matching Analysis (MA) of relative clauses, or a Deletion Under Identity account (cf. Citko 2001), is a combination of the two other approaches to the syntax of RCs (HEA and HRA). It involves base-generation of the relative head in a CP external position – the external head – and adjunction of the RC to this head (similar to the HEA). In addition, it is assumed that there is also a representation of the relative head inside the RC – the internal head. This internal head moves from the RC internal position to Spec,CP (similar to the HRA). Under identity with the external head, PF deletion of the noun in Spec,CP is triggered (cf. Citko 2001). (10) a. b. [DP the picture [CP [DP which picture]j [TP he likes tj]]] [DP the picture [CP which picture [TP he likes tj]]] wh-movement PF-deletion under identity As mentioned above, this analysis is attractive mostly because it is can straightforwardly account for the lack of Principle C effects in (English) RCs (cf. Bhatt 2002, Sauerland 2003). To illustrate how 11 Notice that reconstruction is the default under the HRA. That is to say, according to the Preference Principle (Chomsky 1995) the restriction is deleted in the higher position and retained in the lower position (cf. Salzmann 2006). 12 For some criticism on Safir’s (1999) account of the lack of Principle C effects, see Citko (2001). 13 As noted by Salzmann (2006), this argument holds for the implementations of the HRA by Bianchi (1999,2000), Zwart (2000) and Bhatt (2002), and only to a lesser extent for the version of the HRA of De Vries (2002). That is to say, the latter analysis assumes that movement of the relative head noun to the specifier of the relative pronoun applies before movement to Spec,CP (cf. section 1.3.3.1). However, the subsequent movement of the features of the head noun to the external determiner still violates the CED. 8 this works exactly, consider again sentence (8), here repeated as (11). The derivation of this sentence is given in (12): (12a) illustrates wh-movement, and (12b) shows PF-deletion under identity. (11) The [picture of Johni] which hei likes __ is on the front page. [Citko 2001:139] (12) a. [TP [DP the picture of Johni [CP [which picture of Johni]k [TP hei likes tk is on the front page]]]] b. [TP [DP the picture of Johni [CP [which picture of Johni]k [TP hei likes tk is on the front page]]]] The copy in Spec,CP reconstructs into the VP at LF, yielding a Condition C violation, as illustrated in (13). (13) [TP [DP the picture of Johni [CP [which picture of Johni]k [TP hei likes picture of Johni is on the front page]]]] In order to account for the lack of Principle C effect in (11), Citko (2001) argues that the offending copy (i.e. the reconstructed copy) in (13) can delete at LF as in (14) – under the assumption that deletion under identity is not restricted to PF – because its content is recoverable from the external head. More specifically, Citko assumes that an LF representation contains two copies of the nominal, one in the external head position and one in the reconstructed position, and that in principle, one of them can delete at LF, since its content is recoverable from the remaining copy. The decision of which copy to delete is determined by independent principles (like scope). (14) [TP [DP the picture of Johni [CP [which picture of Johni]k [TP hei likes picture of Johni is on the front page]]]] An analysis as the one described above can also explain why Principle C effects re-emerge in case of variable binding and the interpretation of collocations, as illustrated by (15). (15) *The headway on Maryi’s project which shei had made __ pleased the boss. [Citko 2001:140] At LF, the reconstructed copy cannot be deleted because of the adjacency requirement on idiom interpretation. LF deletion of the lowest copy is thus blocked for independent reasons and a Principle C violation is attested. As mentioned before, Bhatt (2002) and Sauerland (2003) assume a MA only in case there is no reconstruction (as in example (11) above). In other cases, i.e. in case of reconstruction, they assume a HRA. On the contrary, Citko (2001) and Salzmann (2006) argue in favour of a MA as being a sufficient analysis of RCs, i.e. they argue that there is no need to assume a dichotomy between the HRA and the MA within/between languages. We already saw that Citko shows that a MA can straightforwardly account for the fact that Principle C effects re-emerge in certain contexts. More specifically, in terms of Citko’s account of RCs, Principle C effects can be explained by assuming that the reconstructed copy cannot be deleted at LF. Moreover, recall that a problem for the HRA is formed by the observation that in some languages there is a Case mismatch between the relative head noun and the relative pronoun, and note that the MA actually predicts a mismatch in Case as there is no movement chain between the external head and the RC internal gap. Besides Bhatt’s (2002) earlier mentioned argument against the MA stating that the structures provided by the MA cannot generate the low reading of sentences involving adjectival modification (cf. footnote 10), he mentions two other arguments against a MA of RCs. First, he notes that under a MA it is not explained why the external head is pronounced and not the internal head, whereas this follows naturally under a HRA as the external head is the highest copy of a movement chain and normally, the highest copy is spelled-out. Moreover, given that deletion of the RC internal representation is a kind of ellipsis, it is not explained why this ellipsis is obligatory whereas other elliptical reductions are optional.14 Second, Bhatt notes that in ellipsis constructions one phrase is 14 Bhatt (2002:77) himself notes that this argument is not that strong as it can be argued that the kind of ellipsis used in a MA analysis of relative clauses, is similar to ellipsis found in the domain of comparative deletion: an 9 pronounced, whereas both phrases are interpreted. In RCs, however, there are cases where the head NP as a whole is interpreted RC internally (e.g. idiom interpretation). 1.3 Relativization in Standard Dutch Having briefly discussed the different views on the syntax of RCs, in the remainder of this chapter, the syntax of (long-distance) relativization in Dutch will be discussed. I will follow De Vries (2002) and argue that Standard Dutch RCs should be analyzed by a HRA, more specifically, by a promotion theory of RCs. Before I will give a detailed outline of the analysis of RCs in Dutch that I will assume, it is necessary to lay out my assumptions regarding the computational system and its interaction with Phonology/Morphology. Therefore, I will start out with a short introduction into the framework I adopt throughout this thesis. 1.3.1 The framework I will assume the syntactic framework of Minimalism (cf. Chomsky 1995 and subsequent work), together with Kayne’s (1994) antisymmetry hypothesis. That is to say, I will assume a model of grammar that takes Syntax to be a purely derivational system that builds a hierarchical representation of terminal elements by means of the operations (external) merge and move (i.e. internal merge). The operation merge takes two terminal elements out of the lexicon to combine them into a larger unit, and the syntactic operation move takes a terminal element that is already present in the derivation and remerges it in a higher position. The latter mechanism can only be applied when necessary to satisfy some interface requirement,15 that is to say, movement is licensed iff it allows the elimination of uninterpretable formal features (i.e. the Last Resort condition on movement). In addition to these assumptions about the derivational component (i.e. Minimalism and Antisymmetry), I will follow Van Koppen (2005) and further assume that this view on Syntax should be combined with Distributed Morphology (cf. Halle & Marantz 1993, Halle 1997, Harley & Noyer 1999), according to which phonological/morphological information becomes available only after the syntactic component finished the derivation. In other words, Syntax is assumed to be operating only on roots and feature bundles, and not until Syntax is finished manipulating these elements, at the level of Morphology the abstract feature bundles are replaced by Vocabulary Items. A representation of this model of grammar is given in (16) (cf. Van Koppen 2005:12). (16) the computational system Lexicon [roots, feature bundles] Syntax [merge, move, feature checking] PF Vocabulary Items LF Morphology A feature of Distributed Morphology that is important in the light of this thesis, is the Subset Principle as formulated by Halle (1997:428), and given in (17). This principle ensures that in the case of elliptical process called ‘movement deletion’, which is obligatory and can be non-local. Both properties are also found in the domain of relative clauses: the clause internal chain in relative clauses is obligatorily deleted and locality effects are also present in relative clauses. Thus, once one recognizes the reality of (the properties of) ‘movement deletion’, the argument raised against the MA does not hold. 15 The central assumptions of Minimalism (Chomsky 1995) are (i) the interface levels PF (the perceptual/ articulatory system) and LF (the conceptual/intensional system) are the only relevant linguistic levels, i.e. syntax is purely derivational, (ii) all conditions are interface conditions, and (iii) a linguistic expression is the optimal realization of the interface conditions (i.e. economy of representation). 10 competing Vocabulary Items – both items match (a subset of) the feature bundle – the item with the largest number of features matching the feature bundle, will be inserted. (17) Subset Principle16 The phonological exponent of a Vocabulary item is inserted into a morpheme in the terminal string if the item matches all or a subset of the grammatical features specified in the terminal morpheme. Insertion does not take place if the Vocabulary item contains features not present in the morpheme. Where several Vocabulary items meet the conditions for insertion, the item matching the greatest number of features specified in the terminal morpheme must be chosen. 1.3.2 Properties of Dutch restrictive relative clauses17 Now that the framework I adopt throughout this thesis is clarified, we can move on to a description of the properties of RCs in Dutch. First, RCs in Dutch are postnominal/head-initial, which means that the RC always follows the head it modifies. Second, contrary to what is found in English, the Dutch RC is always introduced by a relative pronoun.18 Third, the relativized constituent leaves a gap at the extraction site (not a resumptive pronoun). And fourth, given that a RC is a subordinate clause, the verbs cluster at the end of the sentence. Thus, Dutch restrictive RCs obey the format in (18). (18) … relative DPi [RC relative pronoun … gapi … V ] Moreover, in Standard Dutch relative clauses a complementizer cannot be present in addition to the relative pronoun, as illustrated by (19). (19) * … de man [RC die dat het gedaan heeft] the man die that it done has As for the reconstruction properties in Dutch RCs, see (20). Dutch restrictive relatives show reconstruction for Principle A, variable binding and idiom interpretation. (20) a. b. c. … het [gerucht over zichzelfi] dat Jani niet __ verdragen kan the rumour about SE-SELF that Jan not bear can ‘… the rumour about himselfi that Jani cannot bear’ … de [foto van zijni geliefde] die iedere mani __ in zijn portefeuille heeft19 the picture of his beloved which every man in his wallet has ‘… the picture of his beloved that every man keeps in his wallet’ De [streek] die hij me leverde, riep om wraak. [De Vries 2002:78] the nasty joke which he me delivered, cried for revenge A famous illustration of the workings of the Subset Principle is provided by Sauerland’s (1996) example of adjectival inflection (cf. Van Koppen 2005:16-17). The paradigm of Dutch strong adjectival inflection is given in the following table, which shows that there are two Vocabulary Items competing for insertion, namely e and ø, a zero morpheme. Only if the environment in which the Vocabulary Item needs to be inserted is [-plural, +neuter], the zero-morpheme will be inserted. In all other cases (i.e. the elsewhere environment) –e will be inserted (as the default). Moreover, –e cannot be inserted in the specific environment [-plural, +neuter], because the zeromorpheme is more specific, i.e. it matches the largest number of features. 16 - plural + plural - neuter -e -e + neuter ø -e 17 This thesis restricts itself to restrictive relative clauses in which a subject or object is relativized (i.e. no locative or possessive relative clauses). Moreover, in this chapter I will not discuss PP-relatives (see chapter 5). 18 The rules for the correct choice of relative pronouns in Standard Dutch relative clauses are rather complex. Since I am only concerned with subject/object restrictive relative clauses that require die in Standard Dutch, I will not discuss these rules, but see Smits (1988:367-386), and Haeseryn et al. (1997:327-345) for an overview. 19 (20a) and (20b) are translated from German examples in Salzmann (2006b:66). 11 In all three sentences above, the part of the relative head that is between brackets gets reconstructed into the underlined position in the RC. However, reconstruction is not always possible in Dutch RCs. There are constructions in which the external head cannot be reconstructed into the RC because of independent principles, e.g. the adjacency requirement on idiom interpretation in (21a) and Principle C in (21b) (cf. Salzmann 2006,2006c for German restrictive relatives). (21) a. b. Hij leverde [een streek] he delivered a nasty joke … het [verhaal over Jani] the story about Jan die om wraak which for revenge dat hiji mij __ that he me riep. cried liever niet verteld had rather not told has Sentences like these seem to force the conclusion that reconstruction is not obligatory. This conclusion is corroborated by the ambiguity of the following sentence (De Vries 2002:82). (22) De kunstenaari vervaardigde de buste van zichzelfi/k die de koningk had besteld.20 the artist made the bust of SE-SELF which the king has ordered While there is a vast literature on (the properties of) RC formation (in Dutch), little is known about long-distance relativization constructions in which the relativized constituent is extracted from a deeply embedded sentence. In the Standard Dutch long-distance relativization structure, the RC is introduced by a relative pronoun (identical to the Dutch short relativization structure) and the finite embedded clause containing the extraction site (a gap) is introduced by the finite subordinate complementizer dat ‘that’ (cf. Haeseryn et al. 1997).21 This is exemplified in (23). (23) … relative DP [RC relative pronoun … V [finite embedded clause dat … gap … V ]] 1.3.3 An analysis of Dutch relative clauses I will follow Bhatt (2002) and Salzmann (2006) amongst others and assume that reconstruction effects as in (20) imply that the external head has its origin within the RC. This means that the HEA of RCs (section 1.2.1) is ruled out as a possible analysis. The reconstruction facts in (20) however, do not make it possible to distinguish between the HRA and the MA. Although Salzmann (2006) argues for a MA of RCs on the basis of reconstruction data in German that are similar to the Dutch data in (21), I will follow Kayne (1994), Zwart (2000), and De Vries (2002) amongst others and argue for a HRA of RCs in Dutch.22 More specifically, I will adopt De Vries’ analysis and extent it to long-distance relativization, i.e. I will assume that local and long-distance relativization require the same analysis.23 20 As noted by De Vries (2002:82), grammaticality judgements regarding sentences as the one in (22) are not very clear, i.e. they vary with the particular example, context, intonation and speaker. 21 According to ANS (Haeseryn et al. 1997:1304), the construction in (23) is not used very easily, and it is mainly found in written language. A similar conclusion has also been reached by Brachin (1973,1974) regarding long subject relatives in Dutch. 22 Assuming a HRA for Dutch relative clauses means that we need a solution for the problems raised in section 1.2.2 and the reconstruction data in (20)-(21). In the main text I will discuss some of the solutions to the problems, but see De Vries (2002) for a more comprehensive overview. It is important to note that essentially, the choice for a MA or a HRA of relative clauses is not really relevant in light of this thesis, as I am concerned with the attested variation in Dutch long-distance relativization, and not with the reconstruction effects found in the construction. The most important assumption that I make is that I assume there to be successive-cyclic whmovement in long-distance relativization, and that this holds for all the attested variants. Whether the final step of the derivation consists of PF-deletion of the restriction in the operator position, or whether it concerns subextraction to a higher position, is not really relevant for the discussion of the variation. 23 For alternative analyses, see Salzmann (2006) for German relativization, and Sportiche (2008), and Koopman & Sportiche (2008) for French relativization. The latter theory will be briefly discussed in chapter 3 (footnote 81). 12 1.3.3.1 De Vries (2002) The analysis of RCs put forward by De Vries (2002) is a mixture between Kayne’s (1994) proposal and Bianchi’s (1999,2000) analysis.24 More specifically, De Vries assumes a promotion theory of RCs (i.e. the D-complement hypothesis in combination with the raising analysis25). The complete derivation of the Dutch postnominal restrictive RC in (24) is given in (25) (cf. De Vries 2002:123). (24) Dat is de man die het verhaal that is the man die the story ‘That is the man who told the story.’ verteld heeft. told has (25)26,27 a. [DP-rel die [NP man]] b. [DP-rel [NP man]i die ti] c. [VP [DP-rel [NP man]i die ti] [V’ het verhaal verteld heeft]] d. [TP [DP-rel [NP man]i die ti]k [T’ T [VP tk het verhaal verteld heeft]]] e. [CP1 [DP-rel [NP man]i die ti]k [C’ C [TP tk T [VP tk het verhaal verteld heeft]]]] f. [DP de [CP [DP-rel [NP man]i die ti]k [C’ C [TP tk T [VP tk het verhaal verteld heeft]]]]] g. [DP FFi + de [CP [DP-rel [NP man]i die ti]k [C’ C [TP tk T [VP tk het verhaal verteld heeft]]]]] h. [CP2 dat is [DP FFi + de [CP [DP-rel [NP man]i die ti]k [C’ C [TP tk T [VP tk het verhaal verteld heeft]]]]]] First, the NP man is merged as the complement of the relative pronoun die (25a), and subsequently the NP moves to the specifier position of DP (25b). This movement operation is triggered by the need to check φ-feature agreement with D0.28 Next, the entire relative DP is selected as the subject for the predicate het verhaal verteld heeft (25c). After merging the VP with the functional inflection/tense head, the relative subject DP moves to Spec,TP to check/value (amongst others) its Case feature (25d). Then the TP is merged with C0 and, given that the relative pronoun has a wh-feature29, the relative DP moves to Spec,CP to check the wh-feature of C0 (25e). In line with the D-complement hypothesis, the whole CP is then selected as the complement of the external determiner (25f). In (25g) the formal Similar to the framework I adopt, De Vries (2002:6) is working in a framework that is ‘inclined to the Minimalist Program, with a flavour of Antisymmetry’. 25 For the precise argumentation in favour of both analyses, I refer the reader to De Vries (2002:74-83). 26 I have used t in (25) – and in (29) and (30) – only as a notational convention. That is to say, in line with basic assumptions of Minimalism, I assume that movement leaves a copy at its extraction site (the Copy Theory of Movement; Chomsky 1993), not a trace. 27 For now, as nothing hangs on this, I will simply follow De Vries (2002) and assume that (relative) pronouns are D-heads. However, it will turn out that this assumption cannot account for some of the SAND data on relativization in varieties of Dutch (i.e. partial doubling of pronouns), and therefore, in chapter 4, I will introduce a phrasal analysis of (relative) pronouns, as is argued for extensively in the literature (cf. amongst others Cardinaletti 1994, Cardinaletti & Starke 1999, Déchaine & Wiltschko 2002, Van Koppen 2005, Barbiers, Koeneman & Lekakou 2008). Naturally, the question arises whether a phrasal analysis of (relative) pronouns is compatible with De Vries’ (2002) promotion theory of relative clauses. I will briefly return to this issue in chapter 4 (footnote 112). 28 De Vries (2002) assumes that feature checking can take place (i) in a spec-head configuration, or (ii) in a headincorporation structure. Normally, the φ-features on D0 can be checked by overt N-to-D movement, movement of the formal features of N to D (i.e. covert movement), or movement of NP to Spec,DP. It is assumed that in languages like English and Dutch feature movement applies (by default) – it is the most economical option – to establish the agreement relation in φ-features between NP and D0. In relative clauses, however, N-to-D movement is ruled out independently (i.e. it requires identity of features, but there is a Case mismatch), and movement of NP to Spec,DP – (25b) – is the only possibility. Notice that this way of accounting for the fact that the relative head noun moves to Spec,DP invokes a construction-specific statement (cf. Salzmann 2006:16). 29 I will simply assume that relative pronouns have a wh-feature by which movement to Spec,CP is triggered. Note that it could have been argued as well that relative pronouns bear some sort of [+relative] feature that is in nature similar to a ‘typical’ wh-feature (as proposed by Gallego 2004), but the exact nature of the feature is not relevant for the present discussion. 24 13 features of the relative head noun man incorporate into the external determiner in order to check D0’s φ- and Case-features.30,31 Finally, the whole DP is inserted into the matrix clause (25h). Although not immediately touching upon Standard Dutch relativization, in light of the upcoming discussion of the attested variation in (long-distance) relativization in Dutch, I briefly want to discuss the issue of different types of RCs. As mentioned earlier, Standard Dutch RCs always need to be introduced by a relative pronoun. However, a subset of languages allows zero-relativization, or relatives introduced by a (relative) complementizer32 (sometimes in addition to the relative pronoun). De Vries (2002:126) assumes that the difference between English wh-relatives and that-relatives as in (26) is just a surface effect, in the sense that such structures are not structurally different, and should be treated uniformly (cf. Chomsky 1977). (26) a. b. That is the man who told the story. That is the man that told the story. It is often argued that in addition to the complementizer in (26b), there is an empty operator that is equivalent to the relative pronoun. However, one would then expect the following sentence to be grammatical, contrary to fact. (27) *That is the man who that told the story. Traditionally, such sentences are ruled out by making appeal to the Doubly Filled COMP filter (cf. Chomsky & Lasnik 1977). This filter puts a ban on the simultaneous occurrence of a wh-phrase or relative pronoun and a complementizer in the COMP-domain.33 Now, let us return to the difference between the sentences in (26). Under the promotion theory of De Vries (2002), the empty operator (i.e. the empty relative pronoun) in (26b) is taken to be the determiner DREL that is phonetically empty but has all the formal features that a relative pronoun has (i.e. a wh-feature, φ- and Case-features). Under this analysis, (26b) is thus analyzed identically to (26a). 1.3.3.2 Long-distance relativization in Dutch It seems easy to extent De Vries’ (2002) promotion analysis of RCs to long-distance relativization. The only difference is the presence of an additional subordinate clause that is introduced by the finite complementizer dat and from which the relativized constituent is extracted. An important question that needs to be answered is whether the relativized constituent moves in one fell swoop to the relative COMP, or whether it makes an intermediate landing in the Spec,CP of the most deeply embedded clause. It has always been the standard assumption in generative linguistics that wh-movement proceeds step-by-step, i.e. in a successive-cyclic fashion, and there exists a large amount of data supporting this assumption (e.g. wh-copying; cf. Felser 2004). Let us therefore assume that the 30 The external D0 and the relative head noun thus eventually end up together and bear the same Case feature. Crucially, their Case may differ from the Case on the relative pronoun, providing a solution to the Case-problem introduced in section 1.2.2, cf. example (6). 31 Movement of (the features of) N from Spec,CP to the higher D-head, is allowed because there is no barrier between D0 and N0 (cf. Bianchi 1999,2000). 32 For a comprehensive overview of the syntax and typology of relative elements, see De Vries (2002, chapter 5). 33 It is a well-known fact that the Doubly Filled COMP Filter is not a universal, but it is not entirely clear whether this filter also holds for Standard Dutch, as sentences like (i) are often attested in (informal) speech. However, as we saw in section 1.3.2, in Standard Dutch relative clauses, the filter holds, i.e. a complementizer cannot be present in addition to the relative pronoun. As we shall see, various Dutch dialects do allow doubly filled COMP in relative clauses. (i) Ik vraag je wie of/dat het gedaan heeft. I ask you who if/that it done has However, in contrast to English (zero-relativization), in Dutch relative clauses, there needs to be (at least) one overt element in the COMP-domain (cf. Dekkers 1999). As we will see later on, this requirement holds for all varieties of Dutch, i.e. zero-relativization is impossible. 14 relativized constituent indeed moves through the most deeply embedded Spec,CP. In a framework that assumes movement is only triggered by the need to check some formal features (cf. section 1.3.1), the question that immediately arises is what triggers the intermediate movement step of the relativized constituent to a [-Q] C0-head.34 I will simply follow Chomsky (2000) and assume that features that trigger movement (‘periphery features’: P-features) may optionally be added to C0 (or v0) – the head of a phase35. The presence of P-features on C0 thus triggers movement of the relative DP to the edge of every intermediate phase, accounting for successive-cyclicity of wh-movement.36 Given the above assumptions and the promotion theory of RCs as proposed by De Vries (2002), the derivation of the long-distance RC in (28) is given in (29). For convenience, the treestructure of (28) is given in (30). (28) Dat is de man die ik denk dat het verhaal heeft verteld. that is the man die I think that the story has told ‘That is the man who I think told the story.’ (29) a. [DP-rel die [NP man]] b. [DP-rel [NP man]i die ti] c. [VP [DP-rel mani die ti] [V’ het verhaal verteld heeft]] d. [TP [DP-rel mani die ti]k [T’ T [VP tk het verhaal verteld heeft]]] e. [CP1 [DP-rel mani die ti]k [C’ C [TP tk T [VP tk het verhaal verteld heeft]]]] f. [CP2 ik denk [CP1 [DP-rel mani die ti]k [C’ dat [TP tk T [VP tk het verhaal verteld heeft]]]]] g. [CP2 [DP-rel mani die ti]k [CP2 ik denk [CP1 tk [C’ dat [TP tk T [VP tk het verhaal verteld heeft]]]]]] h. [DP de [CP2 [DP-rel mani die ti]k [CP2 ik denk [CP1 tk [C’ dat [TP tk T [VP tk het verhaal verteld heeft]]]]]]] i. [DP FFi + de [CP2 [DP-rel mani die ti]k [CP2 ik denk [CP1 tk [C’ dat [TP tk T [VP tk het verhaal verteld heeft]]]]]]] j. [CP3 dat is [DP FFi + de [CP2 mani die ik denk [CP1 dat het verhaal verteld heeft]]]] 34 For an overview of the different theories that have been proposed to account for wh-movement to a noninterrogative head, see Felser (2004) and references cited there. 35 The theory of phases, introduced by Chomsky (2000,2001), basically states that the derivation of a sentence proceeds step-wise, i.e. syntactic operations (merge and move) take place phase by phase. The functional projections vP and CP – defined as “complete propositions” – are taken to constitute phases. Each phase is spelled out separately, and after spell out it is no longer accessible for further computational purposes. More specifically, syntactic operations are incapable of looking into a phase below its head, i.e. only heads and specifiers are visible for further computation (Phase Impenetrability Condition, see Chomsky 2000). 36 As noted by Felser (2004), this is a rather ad hoc solution to the problem at hand: P-features are present only when needed to trigger intermediate movement steps (but their presence does not provide an explanation for successive-cyclic movement). In other words, there is no independent evidence that supports the presence of Pfeatures on intermediate heads in long-distance wh-movement. Moreover, such features do not contribute to interpretation. However, on the basis of the overwhelming empirical evidence – among which doubling phenomena in relative clauses (cf. chapter 4; also see Barbiers, Koeneman & Lekakou 2008) – I assume there is successive-cyclic movement of the relative DP in long-distance relativization, and I am less concerned with the trigger for the intermediate movement step to Spec,CP. Therefore, I leave this issue as a topic for further research. 15 (30) CP3 dat is DP FFi + de CP2 DPk C’ D’ mani C0 die TP ti ik denk CP1 C’ DPk D’ mani die dat TP ti DPk het verhaal verteld heeft Up to movement of the relative DP to the specifier position of the most deeply embedded clause (29e), the derivation of long-distance relativization proceeds along the same lines as the derivation of short RCs, with the difference being that movement of the relative DP to Spec,CP1 is triggered by the presence of P-features on the C1-head. Next, CP1 is merged with the verb denk, after which CP2 is further built; I have not presented every step of the derivation of CP2 in (29), but see (29f). The whfeature on the C2-head triggers movement of the relative DP to Spec,CP2 (29g), after which the external determiner is merged with CP2 (29h) and the formal features of N0 incorporate into D0 (29i) in order to check D0’s φ- and Case-features. Finally, the whole DP is inserted into the matrix clause (29j), giving rise to the structure in (30). 1.3.4 The investigation of the variation in long-distance relativization in Dutch Now that the analysis of Standard Dutch (long-distance) RCs is discussed, we can proceed to the next chapters of this thesis. The SAND data on long-distance relativization will be presented in chapter 2, and an analysis of these data is presented in the subsequent chapters. The analysis of (long-distance) relativization discussed in this chapter will give direction to my investigation of long-distance relativization in varieties of Dutch. More specifically, I will pursue the default hypothesis and assume that all the variants to long-distance relativization in Standard Dutch – abstracting away from PPrelatives – have a common syntactic basis: the structure in (30). 16 1.4 Summary In this chapter I have focused on the theoretical background that is necessary for the analysis of the microvariation attested in long-distance relativization in Dutch. First, I introduced three competing analyses of RCs and discussed their strong points and some of the problems they face. Next, I presented the theoretical framework in which the discussion of (the variation regarding) the longdistance relativization structures should be situated. Finally, I presented the analysis of the syntax of Standard Dutch RCs that I will assume throughout this thesis and of which it is assumed that it is the common syntactic basis of all attested variants of long-distance relativization. 17 CHAPTER 2 Presentation of the SAND data on long-distance relativization 2.1 Introduction This chapter presents the data on (long-distance) relativization that are the subject of this thesis. The source of data that I make use of is the SAND-database37 – an online corpus of elicited speech and text regarding syntactic microvariation in the Dutch speaking language area. The database consists of data from 267 dialects collected through oral and telephonic interviews. As we shall see, the data on (longdistance) relativization as found in the SAND-corpus show a considerable amount of variation, e.g. for both long subject and object relatives, eight different variants are found. I am not in the first place concerned with the amount of possible variants. Rather I will focus on the pattern dialects exhibit regarding the combination of long subject and long object relatives. I show that we can basically distinguish six main patterns, and four somewhat more marginal patterns. Of these ten patterns in total, which are presented in section 2.3, I have made the following subcategorization: (i) patterns that show long A’-movement without resumptive pronoun, (ii) patterns that show long A’-movement with resumptive pronoun, and (iii) patterns that make use of a PP-relative (for subject relativization, or for both subject and object relativization) with or without a resumptive pronoun in the extraction site. Before the data are presented, some notes on the structure, objectives and methodology of the SANDproject are in order. Therefore, the following paragraph will briefly outline the design and methodology of the project. As the outline presented here only gives the highlights, for a comprehensive description of these issues, I refer the reader to Cornips & Jongenburger (2001) and Barbiers et al. (2005). 2.2 Data collection – the SAND project 2.2.1 The Syntactic Atlas of the Dutch Dialects (SAND) The SAND-project (Syntactic Atlas of the Dutch Dialects) aims at providing an inventory of syntactic variation found in varieties of Dutch spoken in the Netherlands, Flanders and a small part of France. An important motivation for this project is the ever increasing loss of dialects: under pressure of the standard language, many dialects are losing their original structures; it is thus important to describe the currently existing variation. Moreover, the SAND-project provides such a quantity of data such that current syntactic theories can be extensively empirically tested. In addition, the project makes it possible to study (potential) correlatives of given linguistic variables. The empirical domain of research in the SAND-project is restricted to the following four domains: (i) the left periphery of the clause, (ii) pronominal reference, (iii) negation and quantification, and (iv) the right periphery of the sentence. 2.2.2 Method of data collection In this paragraph I will briefly touch upon the following methodological issues: (i) the choice of measuring points, (ii) the choice of informants, (iii) the phasing of the research/fieldwork and (iv) the nature of the elicitation techniques. As regards the first issue, when determining the (amount of) measuring points, several factors played a role: an evenly spread of the measuring points over the language area, the expected amount of variation in a given area, the origin and history of a given location/area, and the relative isolation of the Dutch Wadden islands. On the basis of these criteria, the measuring points were chosen, yielding a total amount of 267 locations: 102 in Belgium, 158 in the Netherlands, and 7 in France. The informants were selected on the basis of several social variables like age, education and proficiency in their dialect, to make sure their profile was as invariable as possible in order to ensure the reliability of the data. Most important in the selection of informants, was the criterion that the informant still had to speak his or her dialect frequently in various social contexts. The phasing of the research/fieldwork consisted of four different phases. First of all, an inventory of syntactic variation was made, mainly on the basis of literature research. The second stage of the research consisted of a written questionnaire that was sent to all the informants. The goal of this stage was to get an insight into the geographic distribution of the different syntactic phenomena in 37 The SAND-database is available at http://www.meertens.knaw.nl/sand/zoeken/. 18 order to make a motivated and efficient choice for the questions that were to be asked in the oral interviews. These oral interviews comprised the third stage of research, and were conducted in 267 locations in the Dutch speaking language area. In this interview round, the informants were presented with a test sentence in their dialect and they were asked to indicate whether such a sentence occurs in their dialect and how it should be translated. To ensure that informants were not influenced by (the lexical and phonological properties of) the standard language spoken by the interviewer, at least two informants were present when the interview took place. This design led to two scenarios. In the first, the two informants could discuss the sentences between them (Belgium); in the other scenario, one of the informants was trained to be the interviewer (the Netherlands). The final stage of research consisted of telephonic interviews that were held in order to complete missing or unreliable data. In the presentation and discussion of the data on (long-distance) relativization, I will only make use of the material that was gathered in the oral and, when available, telephonic interviews. It is obvious that the data coming from these interview rounds are the most reliable. It should also be noted that since I reanalyzed the original data as published in Barbiers et al. (2005), the maps presented in this thesis may sometimes differ from those published. 2.3 The data This section presents the SAND data on (long-distance) relativization that form the empirical basis for this thesis. More specifically, it concerns the data on subject and object restrictive relative clauses with a 3rd person masculine relativized constituent. In addition to translating short subject (1) and short object (2) relatives – of which the results of the oral and telephonic interviews are given in Appendix II – informants were presented with the long subject relative in (3), and the long object relative in (5) which they were asked to translate. In (3) and (5), we are dealing with the Standard Dutch variant of long-distance relativization (cf. section 1.3.2) in which a finite clause introduced by the subordinate complementizer dat is embedded in a relative clause introduced by the relative pronoun die. The relativized constituent – de man ‘the man’: 3rd person singular masculine – is extracted from the most deeply embedded clause, giving rise to socalled long-distance relativization. For expository reasons, the attested variation regarding longdistance relativization in varieties of Dutch that was gathered in the oral interviews is given in (4)-(6); in addition, the geographic distribution of these structures can be found on maps 1 and 2. It should be noted that I have abstracted away from any phonological variation, e.g. forms like dee, dei or der are taken to be form variants of die. The total amount of dialects for which there are reliable data is 218 for the long subject relatives and 216 for the long object relatives; in a given location more than one variant can be attested. I have included the complete list of the SAND data regarding relativization in Appendix I. (1) Dat is de man die het verhaal that is the man die the story ‘That is the man who told the story.’ heeft verteld. has told (2) Dat is de man die ze geroepen hebben. that is the man die they called have ‘That is the man who they have called.’ (3) Dat is de man die ik denk that is the man die I think ‘That is the man who I think told the story.’ dat that __ __ (4) a. dat is de man die b. dat is de man die c. dat is de man dat d. dat is de man die e. dat is de man dat f. dat is de man dat Ø hij/die Ø Ø Ø hij/die het verhaal heeft verteld het verhaal heeft verteld het verhaal heeft verteld het verhaal heeft verteld het verhaal heeft verteld het verhaal heeft verteld ik denk dat ik denk dat ik denk die ik denk die ik denk dat ik denk dat 19 het verhaal the story heeft verteld. has told g. dat is de man waarvan h. dat is de man waarvan ik denk dat ik denk dat hij/die het verhaal heeft verteld Ø het verhaal heeft verteld Map 1; long subject relative (5) Dat is de man die ik denk dat that is the man die I think that ‘That is the man who I think they have called.’ ze they __ __ (6) a. dat is de man die b. dat is de man die c. dat is de man dat d. dat is de man die e. dat is de man dat f. dat is de man dat g. dat is de man waarvan h. dat is de man waarvan Ø die/’m Ø Ø Ø die/’m die/’m Ø geroepen hebben geroepen hebben geroepen hebben geroepen hebben geroepen hebben geroepen hebben geroepen hebben geroepen hebben ik denk dat ik denk dat ik denk dat ik denk die ik denk die ik denk dat ik denk dat ik denk dat 20 ze ze ze ze ze ze ze ze geroepen hebben. called have Map 2; long object relative38 Abstracting away from PP-relatives (4g,h) and (6g,h) – to which I will return in section 2.3.3 – these data show that the long-distance relatives vary along the following parameters. the form of the element that introduces the relative clause (die/dat) the form of the element that introduces the most deeply embedded clause (die/dat) the presence or absence of an overt subject/object in the most deeply embedded clause (i.e. whether or not a resumptive pronoun39 is present) the presence or absence of an (additional) complementizer For the sake of simplicity, I have not included the latter parameter in the examples in (4) and (6). The sentences in (7a) and (7b) illustrate that a complementizer can be present in the higher clause (7a), as well as in the lower clause (7b). The presence of a complementizer in short as well as in long relatives is mainly found in Friesland and the north-eastern and western part of Antwerp (cf. Barbiers 2005 et al.).40 (7) a. b. Da s de vent die ak denk die asse dat is de man die ik denk die dat ze ‘That is the man who I think they have called.’ Da s de man dieë dak denk dasse dat is de man die dat ik denk dat ze geroepen emme. [Nieuwmoer Dutch] geroepen hebben geroepen emme. [Zandhoven Dutch] geroepen hebben In sum, the variation that is found in long-distance relativization structures (both subject and object)41 – again abstracting away from PP-relatives – is summarized in a more abstract manner in (8). 38 This map differs from the map on long object relativization presented in Barbiers et al. (2005) in that this map also includes the constructions waarvan-dat-Ø and dat-dat-die/’m. 39 In the literature, the term resumptive pronoun is not used uniformly. However, I will use it without theoretical presupposition, but just to indicate that instead of a gap in the lower clause, a personal or demonstrative pronoun appears (see chapter 5 for some notes on resumptive pronouns in long-distance relativization structures). 40 Dialects that obligatorily use a complementizer in addition to the relative pronoun are mostly found in Friesland, and are not included on the maps (cf. Appendix II). 41 Notice that the difference between long subject and long object relatives is quantitative, not qualitative. 21 (8) … DP [RC die/dat (dat) … [finite embedded clause die/dat (dat) … gap/resumptive pronoun …]] In this thesis, I am not in the first place concerned with all the different possibilities regarding long subject or object relatives dialects makes use of, but rather, I will investigate the patterns that dialects exhibit. That is to say, I am concerned with the ‘system’ – the combination of a subject and object relative – a given dialect has with respect to (long-distance) relativization. In order to get an overview of the main systems Dutch varieties make use, the following table shows the amounts of occurrences of the combination of a long object relative with a long subject relative. This table thus shows whether and how often particular long subject and object relatives occur together. Table 1; combinations of long subject and object relatives 42 die-dat-Ø (99) die-dat-die/’m (9) die-die-Ø (24) dat-die-Ø (11) dat-dat-Ø (50) dat-dat-die/’m (5) waarvan-dat-die/’m (48) waarvan-dat-Ø (8) object → 47 13 19 1 4 20 2 2 2 4 1 2 1 3 1 9 2 1 4 - 2 2 1 7 3 2 - 5 3 4 20 16 2 3 - 1 1 3 1 - 2 4 2 3 1 34 5 1 1 3 2 subject ↓ die-dat-Ø (63) die-dat-hij/die (24) die-die-Ø (32) dat-die-Ø (36) dat-dat-Ø (20) dat-dat-hij/die (10) waarvan-dat-hij/die (69) waarvan-dat-Ø (10) As the table shows, some long subject and object relatives often occur together, whereas a combination of others is (almost) never attested. These results forces one to conclude that there must be some sort of system in the possible combinations of subject and object relative constructions. On the basis of table 1, I distinguish the following 6 main systems of long-distance relativization (I-VI) and 4 somewhat more marginal systems of long-distance relativization (VII-X). Map 3 shows the geographical distribution of the six main systems. (9) Systems of long-distance relativization in Dutch dialects long subject relative long object relative system I die-dat-Ø die-dat-Ø system II waar van-dat-hij/die waarvan-dat-die/’m system III waar van-dat-hij/die die-dat-Ø system IV dat-die-Ø dat-dat-Ø system V die-die-Ø die-dat-Ø system I dat-dat-Ø dat-dat-Ø system VII die-die-Ø die-die-Ø system VIII dat-die-Ø dat-die-Ø 42 amount of dialects 47 34 20 20 19 16 9 7 This table shows the amount of dialects that make use of a combination of a particular long subject and object relative, after a thorough inspection of the dialects at stake. That is to say, given that dialects often make use of more than one variant of long-distance relativization, sometimes it is not clear whether a dialect really makes use of a combination of two relatives, or whether these relatives are actually never used ‘together’ and the dialect in fact makes use of two different systems. Therefore, I took all the numbers in the table higher than 4 and checked whether the given dialects make use of a particular system or not. When I was not entirely sure, I have taken the lowest number, i.e. the number that indicates which dialects make use of the system for sure. 22 system IX system X die-dat-hij/die waarvan-dat-Ø die-dat-Ø waarvan-dat-die/’m 13 5 Map 3; 6 systems of long-distance relativization 2.3.1 Data on long-distance relativization – long A’-movement without resumptive pronoun In this subparagraph, I will briefly investigate the system behind the long-distance relatives that do not have a resumptive pronoun in the most deeply embedded clause and that are not PP-relatives. This omission of several structures leaves us with the six systems in (10), which I will call the long A’movement configurations without resumptive pronoun.43 Table 2 indicates for each system of longdistance relativization, the corresponding ‘system’ of short relativization it makes use of. Map 4 shows the geographic distribution of the long A’-movement configurations. (10) system I system IV system V system VI system VII system VIII die-dat-Ø dat-die-Ø die-die-Ø dat-dat-Ø die-die-Ø dat-die-Ø die-dat-Ø dat-dat-Ø die-dat-Ø dat-dat-Ø die-die-Ø dat-die-Ø Table 2; relation between short and long relativization I IV V VI VII VIII short subject die die die dat die die short object die dat die dat die die 44 20 19 10 9 6 long subject die-dat-Ø dat-die-Ø die-die-Ø dat-dat-Ø die-die-Ø dat-die-Ø 43 long object die-dat-Ø dat-dat-Ø die-dat-Ø dat-dat-Ø die-die-Ø dat-die-Ø 47 20 19 16 9 7 The fact that I choose to make use of this term for these particular long relatives, by no means indicates that the other long relativization structures do not make use of long A’-movement. 23 Map 4; 6 systems – long A’-movement without resumptive pronoun These six systems differ with respect to the following three parameters. the absence/presence of a subject/object asymmetry die/dat variation in the higher clause die/dat variation in the lower clause An important question that needs to be answered now is why, of all the imaginable combinations of long-distance subject and object relatives without resumptive pronoun, only these six combinations are (often) attested. Table 3 shows all the possible combinations of these subject and object relatives and their value for each of the three parameters mentioned above. Table 4 shows which combinations of values for the different parameters leads to grammaticality. Table 3; three parameters for the combination of long-distance A’-movement constructions long subject die-dat-Ø die-dat-Ø die-dat-Ø die-dat-Ø dat-die-Ø dat-die-Ø dat-die-Ø dat-die-Ø die-die-Ø die-die-Ø die-die-Ø die-die-Ø dat-dat-Ø dat-dat-Ø dat-dat-Ø dat-dat-Ø long object die-dat-Ø die-die-Ø dat-dat-Ø dat-die-Ø die-dat-Ø die-die-Ø dat-dat-Ø dat-die-Ø die-dat-Ø die-die-Ø dat-dat-Ø dat-die-Ø die-dat-Ø die-die-Ø dat-dat-Ø dat-die-Ø (47) (3) (5) (2) (-) (2) (20) (7) (19) (9) (4) (1) (1) (-) (16) (3) s/o-asymmetry + + + + + + + + + + + + 24 variation in higher clause + + + + + + + + - variation in lower clause + + + + + + + + Table 4; three parameters for the combination of long-distance A’-movement constructions subject/objectasymmetry + + + variation in higher clause + + variation in lower clause + + grammaticality √ * % * On the basis of table 4, the following three generalizations can be formulated. (11) Generalizations - long A’-movement without resumptive pronoun I long-distance relativization without a subject/object asymmetry shows all the possible variants – die-dat-Ø – die-dat-Ø, dat-die-Ø – dat-die-Ø, die-die-Ø – die-die-Ø, dat-datØ – dat-dat-Ø II a subject/object asymmetry can appear only in the CP containing the extraction site, (i.e. a subject/object asymmetry in the higher clause is (almost) never attested) III in case of a subject/object asymmetry, dat occurs in the most deeply embedded clause when the object is extracted, whereas die occurs in the most deeply embedded clause in case of subject extraction The following two chapters of this thesis discuss these three generalizations, and an analysis of the long-distance relativization structures in (10) will be proposed. More specifically, chapter 3 deals with long-distance relativization structures with a subject/object asymmetry (systems IV and V) and tries to explain generalizations II and III. Chapter 4 discusses long-distance relativization configurations without subject/object asymmetries, and proposes an explanation for generalization I. 2.3.2 Data on long-distance relativization – long A’-movement with resumptive pronoun Whereas resumptive pronouns are found both with long subject and long object relatives, as illustrated in (12), they are found more often with subject relatives (34 vs. 14). This tendency is reflected in system IX, in which the subject relative clause has a resumptive pronoun at the extraction site, whereas the object relative clause has a gap (13). (12) a. Dat is de man die b. Dat is de man dat c. Dat is de man die d. Dat is de man dat (13) system IX ik denk dat ik denk dat ik denk dat ik denk dat hij/die het verhaal heeft verteld. hij/die het verhaal heeft verteld. ze die/’m geroepen hebben. ze die/’m geroepen hebben. (24) (10) (9) (5) die-dat-hij/die – die-dat-Ø Resumptive pronouns are only found in relative constructions where the most deeply embedded clause is introduced by the complementizer dat. The absence of resumptive pronouns in clauses introduced by die is probably due to the general tendency of languages to disallow repetition of adjacent identical morphemes (cf. Neeleman & Van de Koot 2006). In chapter 5, I will briefly discuss possible analyses of resumptive pronouns in relative clauses in Dutch. 25 Map 5; resumptive pronouns – system IX 2.3.3 Data on long-distance relativization – PP-relatives As mentioned above, the informants were presented with the Standard Dutch long-distance relativization structures, and were asked to translate such structures/sentences in their local dialects. Interestingly, many informants spontaneously translated the test sentences with a construction in which the relative clause is introduced by a prepositional phrase (14a). In fact, sentence (14a) is not really a subject relative, as it is the relativized variant of (14b), i.e. the prepositional phrase van die man ‘of that man’ is relativized in the higher clause and the subject in the lower clause is retained. (15) shows that preposition stranding is also possible with this construction. Note that the sentences in (14) and (15) are perfectly fine in Standard Dutch. (14) a. b. (15) Dat is de man waarvan ik denk dat hij/die het verhaal verteld heeft that is the man whereof I think that he/die the story told has Ik denk van die man dat hij het verhaal verteld heeft I think of that man that he the story told has Dat is de man waar ik van denk dat hij/die het verhaal that is the man where I of think that he/die the story verteld heeft. told has As (14a) shows, the preposition van ‘of’ follows its object waar ‘where’, giving rise to the complex element waarvan ‘whereof’. However, it is also possible for the preposition to appear in front of its object, resulting in the structure van wie ‘of whom’, as indicated in (16).44 In this thesis I will not go into these different manifestations of the PP-relative, and I will refer to (14a) as the standard PPrelative. (16) Dat is de man van wie that is the man of whom ik denk I think dat hij/die het verhaal verteld heeft. that he/die the story told has Sentences like (16) are almost never attested in the SAND-material. To distinguish between (14a) and (15) – involving waarvan ‘whereof’ – on the one hand, and (16) – involving van wie ‘of whom’ – on the other hand, sentences like (16) are not presented on the maps. 44 26 In order to explain the spontaneous translation of the Standard Dutch long-distance relativization structure with a PP-relative, Barbiers et al. (2005) suggest that long relativization – especially long relativization of subjects (see also system III) – is impossible in many dialects, and that these dialects therefore make use of an alternative construction that is semantically very similar to long-distance relativization: the PP-relative. Interestingly, these PP-relatives show variation with respect to the presence of a resumptive pronoun. That is to say, contrary to Standard Dutch, some dialects allow the resumptive pronoun to be absent, giving rise to sentences like (17) – the grammaticality of which, according to Salzmann (2006:159), is highly unexpected, i.e. the relativized object should be related to a resumptive pronoun instead of a gap. The geographic distribution of PP-relatives without a resumptive pronoun is limited to the north-eastern area of the Netherlands (cf. maps 1-2). (17) a. b. Dat is de man weervan that is the man whereof Dat is de man weervan that is the man of where-of ik tink dat __ het verhaal vertoald het. I think that __ the story told has ik tink dat ja __ ruppen hewwe. I think that they __ called have [Schiermonnikoog Dutch] The PP-relative construction appears in three out of the ten systems of long-distance relativization; the three systems are given in (18). System II is very common and occurs mostly in the Netherlands (cf. map 3). The same holds more or less for system III, with the difference being that it occurs more often in the south of the Netherlands. System X, finally, is only attested 5 times in the north-eastern part of the Netherlands (i.e. provinces Drenthe and Overijssel). (18) system II system III system X waarvan-dat-hij/die waarvan-dat-hij/die waarvan-dat-Ø waarvan-dat-die/’m die-dat-Ø waarvan-dat-die/’m As the main focus of this thesis is not on PP-relatives, I will only briefly discuss a potential analysis (proposed by Salzmann 2006) of this construction in chapter 5. 2.4 Summary This chapter presented the SAND data on (long-distance) relativization. It was shown that although at first sight, the attested variation seemed overwhelming and unstructured, six main systems and four somewhat more marginal systems of long-distance relativization can be found, which are presented in the following table; for convenience, the corresponding ‘systems’ of short relativization are also given. Table 5; relativization in varieties of Dutch – 10 systems I II 45 III IV V VI VII VIII IX X short subject die die die die die dat die die die die short object die die die dat die dat die die die die 46 26 20 20 19 10 9 7 13 5 long subject die-dat-Ø waarvan-dat-hij/die waarvan-dat-hij/die dat-die-Ø die-die-Ø dat-dat-Ø die-die-Ø dat-die-Ø die-dat-hij/die waarvan-dat- Ø 45 long object die-dat-Ø waarvan-dat-die/’m die-dat-Ø dat-dat-Ø die-dat-Ø dat-dat-Ø die-die-Ø dat-die-Ø die-dat-Ø waarvan-dat-die/’m 47 34 20 20 19 16 9 7 13 5 The fact that the amount of dialects that make use of PP-relatives for both long subject and long object relativization is significantly larger than the amount of dialects that use die for short relativization (34 vs. 26) is mostly due to the fact that dialects that obligatory have a complementizer in addition to a relative pronoun, are not included (cf. footnote 40 and Appendix II). In these dialects, mainly found in Friesland, the relative pronoun die only occurs together with the (enclitic) complementizer t, giving rise to diet (cf. Appendix I). The different amounts in table 5 thus not indicate that something special is going on. 27 Now that the framework and the analysis of Standard Dutch relative clauses that I adopt throughout this thesis (chapter 1) is discussed, and the data on long-distance relativization in varieties of Dutch (this chapter) are presented, we can proceed to the following chapters: the data analysis. My main focus will be on the six systems involving long A’-movement without resumptive pronoun. 28 CHAPTER 3 Long-distance relativization in Dutch dialects – subject/object asymmetries 3.1 Introduction In the previous chapter, it was shown that some varieties of Dutch show subject/object asymmetries (henceforth s/o-asymmetries) in (long-distance) relativization, whereas others do not. This chapter will investigate the system underlying these subject/object asymmetries. In other words, this chapter proposes an explanation for the following two generalizations. (1) II III a subject/object asymmetry can appear only in the CP containing the extraction site (i.e. a subject/object asymmetry in the higher clause is (almost) never attested) in case of a subject/object asymmetry, dat occurs in the most deeply embedded clause when the object is extracted, whereas die occurs in the most deeply embedded clause in case of subject extraction The organization of this chapter is as follows. In paragraph 3.2 I will introduce the relevant data regarding s/o-asymmetries in relativization structures. In section 3.3 I will present a brief overview of the scholarly literature on s/o-asymmetries in general, and in relative clauses in particular. This overview starts with the early work on s/o-asymmetries by Kayne (1976), Pesetsky (1982) and Rizzi (1990) amongst others, and ends with a very recent proposal by Mayr (to appear). It should be noted that this section does not attempt to give a comprehensive overview of the existing literature; rather, it just highlights some important issues at which I will come back in later paragraphs. Section 3.4 proposes an analysis for the data on s/o-asymmetries in (long-distance) relativization structures in Dutch, and discusses some predictions the analysis makes and some problems it faces. It will be argued that the element die, found in the most deeply embedded COMP-domain of structures involving (long-distance) subject extraction in dialects that show the s/o-asymmetry, is an instance of the finite subordinate complementizer. In other words, I will follow traditional analyses, like Kayne (1976) and Rizzi (1990), and propose that die is an agreeing form of the complementizer. The proposal is highly influenced by recent work of Mayr (to appear), who argues that subject extraction must be licensed by complementizer agreement. As we shall see, this analysis is capable of deriving the attested patterns of s/o-asymmetries and it explains the main generalizations in (1). However, a number of problematic cases remain, which this analysis is unable to account for. Therefore, section 3.5 discusses two alternative analyses of the s/o-asymmetry, and shows that each has its own shortcomings. Finally, in section 3.6, I will summarize my findings. 3.2 Presentation of the data In the six main systems of long-distance relativization, we found two systems that exhibit a s/oasymmetry. In the first system (IV) the relative clause is introduced by the complementizer dat, and the embedded finite clause is introduced by die in case of subject extraction and by dat in case of object extraction, as illustrated by (2). This system of long-distance relativization is attested in 20 locations and it is found almost exclusively in West-Vlaanderen, as can be seen on map 1. The second system that shows a s/o-asymmetry is illustrated by (3). The relative clause is introduced by die, and the lower clause is introduced by die in case of subject extraction and by dat in case of object extraction. This system is found 19 times and its geographic distribution is somewhat less clear than that of system IV. It is attested in the main part of Oost-Vlaanderen and several times in the Netherlands; there are some isolated occurrences and a small cluster of attestations in the east of the Netherlands near the city Nijmegen. (2) a. b. (3) a. Da s de vent that is the man Da s de vent that is the man da k peizen that I think da k peizen that I think Da s de man die k peize that is the man die I think die da graptje verteld eet. die that joke told has da-n ze geroepen en. that-3P.PL they called have subject extraction die t verhaal verteld ee. die the story told has subject extraction 29 object extraction [Brugge Dutch] b. Da s de man die k peize that is the man die I think da-n ze geropen en. that-3P.PL they called have object extraction [Gent Dutch] Map 1; two systems involving subject/object asymmetries Until now, I mainly focussed on long-distance relativization structures and paid little attention to short (one clause) relatives. In the present case, however, the data on short relativization are of great importance. The West-Flemish data (system IV) are pretty straightforward: all the dialects that show the s/o-asymmetry also show an asymmetry in short relativization structures, i.e. die shows up with subjects and dat appears with objects. The data of system V, however, are less clear and somewhat puzzling at first sight.46 Table 1 shows that an asymmetry is only attested with system V in case of long-distance relativization. In short relativization configurations, on the other hand, die is used with both subject and object extraction.47 Table 1; two systems involving subject/object-asymmetries48 system IV (20) system V (19) short subject die die short object dat die 46 long subject dat-die-Ø die-die-Ø long object dat-dat-Ø die-dat-Ø As we shall see in section 3.5, the derivation of the relativization pattern of system V is problematic for some theories of subject/object asymmetries. That is to say, it is hard to simultaneously account for the presence of an asymmetry in long-distance relativization and the lack of an asymmetry in short relatives within a particular language. 47 Actually, the data are less clear than given in table 1 (for the complete overview of the data, see Appendix I). Of the 19 dialects that make use of system V, all 19 dialects make use of die in short subject and object relatives. But of these 19 dialects there are also 10 dialects that, in addition to die, can also use dat in short object relatives (and sometimes there are even more alternatives possible for both object and subject relatives). However, in what follows, I will assume that the pattern given in table 1 is in fact the core pattern of system V – that is, a system that shows a s/o-asymmetry with long-distance relativization, but no asymmetry with short relativization. The correctness of this assumption is corroborated by the observation that of the few dialects that only make use of system V, the majority shows the pattern in table 1 (based on the data from the oral interviews). 48 In the dialects under discussion the complementizer can be dat or da. In the main text, I will simply use dat for both form variants. 30 When dealing with (micro)variation, it is always important to investigate whether the variable studied – in this case the presence of a s/o-asymmetry in the CP containing the extraction site – covaries with another variable (cf. Kayne 2000). In the present case, a potential correlating property is complementizer agreement (see also Bennis & Haegeman 1984, Mayr to appear). Complementizer agreement is the phenomenon illustrated in sentences (2b) and (3b), i.e. a complementizer agrees with the subject of the clause it introduces.49 Map 2 shows the geographic distribution of complementizer agreement; more specifically, the map shows for each dialect which members of the paradigm show complementizer agreement.50 Map 2; complementizer agreement Comparing this map with map 1, we see that the geographic distribution of dialects that show s/oasymmetries in long-distance relativization roughly corresponds to the geographic distribution of complementizer agreement. That is to say, the core area that shows complementizer agreement – i.e. the area in which complementizer agreement manifests itself overtly on more than one person – matches the area that shows s/o-asymmetries in long-distance relativization configurations. Therefore, in addition to the two generalizations in (1), a third generalization is formulated. This chapter will propose an account of these three generalizations. (4) IV dialects that show a subject/object asymmetry in (long-distance) relativization structures, often also show complementizer agreement 51,52 49 I will assume an analysis of complementizer agreement as proposed by Van Craenenbroeck & Van Koppen (2002), Carstens (2003) and Van Koppen (2005). C0 is assumed to have uninterpretable/unvalued φ-features that need to be linked to interpretable/valued φ-featured by means of the syntactic operation Agree. C0 is thus a probe for φ-features and Agree searches its c-command domain in order to find a suitable goal; this is found in the embedded subject: the interpretable/valued φ-features of the subject erase/value the unvalued/uninterpretable φfeatures on C0. As a result of this agreement relation, the complementizer overtly agrees with the subject. 50 The data on this map are taken from Barbiers et al. (2005). 51 Of the 39 dialects that show the asymmetry, 32 also exhibit complementizer agreement. The dialects that do not show complementizer agreement are generally the ones that are found outside Flanders. 52 It should be clear that this generalization is a one-way implication, i.e. it is by no means true that dialects that exhibit complementizer agreement, also exhibit s/o-asymmetries in long-distance relativization. 31 3.3 Theoretical background This section provides the theoretical background necessary for an understanding of the more recent proposals that try to account for the s/o-asymmetry – e.g. Taraldsen (2001), Pesetsky & Torrego (2001,2004), Gallego (2004), Rizzi & Shlonsky (2007), Mayr (to appear). In order to understand these particular accounts, we need to know something about (the history of) accounting for the s/oasymmetry in general. Traditionally, subject/non-subject asymmetries are explained by appealing to the Empty Category Principle (e.g. Chomsky 1986, Bennis & Haegeman 1984, Rizzi 1990), and therefore, in the first subparagraph, I will give an outline of the first attempts to account for the asymmetry that is attested in many languages. 3.3.1 The Empty Category Principle (ECP) Perlmutter (1971) was one of the first to study subject/non-subject asymmetries like the that-trace effect – i.e. the obligatory absence of a complementizer in the COMP-domain of the clause from which the subject is extracted – as illustrated in (5)-(6) (5) a. b. … the woman that he said (*that) bought the book … the book that he said (that) she bought subject extraction object extraction (6) a. b. Who did he say (*that) bought the book? What did he say (that) she bought? subject extraction object extraction As noted by Mayr (to appear), the crucial question that needs to be answered when dealing with such s/o-asymmetries is the following: what is the difference between extraction from Spec,TP and the complement position of the verb? Provided that moved constituents leave a coindexed trace at their extraction site, in the Government and Binding framework (Chomsky 1981,1982,1986) an answer to this question was given by making appeal to the Empty Category Principle (henceforth the ECP), which basically states that subject traces need to be locally bound, see (7)-(10). (7) Empty Category Principle traces must be properly governed (8) Proper Government α properly governs β iff (i) α governs β and α theta-marks β (theta government), or (ii) α antecedent-governs β (9) Government α governs β iff (i) α c-commands β, and (ii) no maximal projection (except TP) intervenes between α and β, and (iii) minimality is respected (10) Antecedent Government α antecedent-governs β iff (ii) α governs β, and (iii) α binds β (i.e. α c-commands β and α is coindexed with β) From these definitions it follows that object traces will always be properly governed by the governing verb (theta government), whereas subject traces will only be properly governed (i.e. antecedent governed) if they can be locally bound by the moved constituent or its trace in the COMP-domain. In sentences (5a) and (6a) above, the ECP is violated by the subject trace in Spec,TP when complementizer that is present. The presence of that blocks the necessary government relation between the trace of the subject in Spec,TP and the intermediate trace in Spec,CP because minimality is violated, i.e. that is an intervening potential governor. Moreover, that – being the closest potential governer for the trace in Spec,TP – cannot bind this trace (it is not coindexed with the trace) and 32 therefore, cannot antecedent-govern it. Thus, the subject trace in Spec,TP cannot be properly governed – in violation of the ECP – and sentences (5a) and (6a) with complementizer that in the embedded clause are out. Only when that is absent, the necessary antecedent government relation between the trace in Spec,CP and the trace in Spec,TP can be established and the ECP will be satisfied. Since object traces are always governed by the verb, complementizer that may (optionally) surface in the COMP-domain of clauses from which the object is extracted. In light of the data on s/o-asymmetries in (long-distance) relativization in varieties of Dutch (cf. section 3.2), it is interesting to take a look at the que/qui alternation in French (in relative clauses), which is exemplified in (11)-(12). (11) a. b. (12) a. b. … l’homme the man … l’homme the man *que/qui t viendra that/who t will come que/*qui j’aime t that/whom I love t subject extraction … l’homme the man … l’homme the man que tu pense that you think que tu pense that you think subject extraction *que/qui t viendra that/who t will come que/*qui j’aime t that/whom I love t object extraction object extraction In case of subject extraction, instead of the tensed complementizer que, we find the element qui.53 Given the generalization that no DP is allowed in the COMP-position of restrictive relative clauses in French, Kayne (1976) assumes that the instances of the element qui in the examples above, are not relative pronouns/wh-elements, but rather, they are special instances of the French complementizer que. A special rule changes que into qui when the adjacent subject in Spec,TP is relativized. Rizzi (1990) – basically a reformulation of Pesetsky (1982) – argues that the que/qui alternation in French is the result of agreement. Given the ECP as formulated in (7), he assumes that subject traces are properly governed when there is Spec-head agreement in the COMP-domain – this amounts to saying that agreement in COMP licenses subject extraction (cf. Mayr to appear). In other words, C0 can be turned into a governor by agreement in COMP. Rizzi argues that the surfacing of qui instead of que is a manifestation of Spec-head agreement between the intermediate subject trace in Spec,CP and the complementizer que in C0, in short, qui = que + agreement. Similarly, he proposes that West-Flemish die is the reflex of Spec-head agreement between the trace in Spec,CP and the complementizer dat. For both French and West-Flemish it is the case that the occurrence of the particular morphological change is very limited. It is only attested in the COMP-domain of clauses from which the subject is extracted, i.e. qui cannot occur in the higher clause of long-distance subject extraction (13)-(14), nor can it be found in case of object extraction (12b). The latter fact is explained by assuming that Spec-head agreement turns the specifier position into an A-position – i.e. the most deeply embedded Spec,CP becomes an A-position. The chain formed by object extraction will then cross another A-position – Spec,TP – which violates Relativized Minimality. (13) … l’homme que tu pense the man that you think (14) … de vent the man t que/*qui Jean croit t *que/qui t viendra French t that/who Jean believes t that/who t will come *die/da k peizen die/that I think die/*da da graptje verteld eet die/that that joke told has West-Flemish Within the Government and Binding framework, “traces are grammar-internal constructs with very special requirements that regulate their distribution” (Hornstein et al. 2005:12). In GB, movement was taken to be free – move α – and a large part of linguistic theory consisted of rules and constraints to restrict this movement operation. The main reason for why traces came into existence was exactly to 53 Notice that the French data on (long-distance) relativization are reminiscent of the West-Flemish data (system IV), i.e. die has the same distribution as qui: it appears only in the left periphery of clauses from which the subject is extracted. 33 restrain overgeneration. However, with the introduction of Minimalism, a framework in which movement is no longer taken to be free, but rather, is assumed only to occur when it needs to satisfy some interface requirement, the existence of traces as grammatical formatives is no longer postulated. 54 More specifically, traces are replaced by copies. With the abandonment of traces, and more generally, the abandonment of a theory which main objective is to constrain movement, the conceptual reality of principles like the ECP – which was designed to regulate the distribution of traces – was seriously questioned. Moreover, in a framework that wants to give an answer to the question why language is the way it is, there is no longer place for (principles like) the ECP which stipulates the difference between subjects and non-subjects, but does not explain why subjects, but not objects, are subject to a special binding requirement (cf. Pesetsky & Torrego 2001:358). Thus, the ECP was largely abandoned as an explanation of subject/non-subject asymmetries, and several alternative theories were formulated to capture the contrast. In the following section, I will discuss the approach that functions as the basis for my analysis of the s/o-asymmetries that are attested in Dutch. 3.3.2 Mayr (to appear) Recently, Mayr (to appear) has argued for a treatment of long-distance subject extraction in Bavarian in terms of the relation with complementizer agreement (cf. Rizzi 1990). More specifically, Mayr argues that complementizer agreement licenses extraction of subjects. In contrast to English, that-trace phenomena are absent in Bavarian, i.e. both subjects and objects are free to extract independent of the presence of a complementizer (the same holds for long-distance topicalization); this contrast is illustrated in (15)-(16). (15) a. b. Whoi does Mary believe [ti that John left ti]? Whoi does Mary believe [ti (*that) ti left Anna]? (16) a. Weai hot da Michl gsogt [ti dass ti gestan a Biachl kafft hot]? subject extraction who has the Michael said that yesterday a book bought has ‘Who did Michael say that bought a book yesterday?’ Wosi hot da Michl gsogt [ti dass d’Maria ti kafft hot]? object extraction what has the Michael said that the Mary bought has ‘What did Michael say that Mary bought?’ b. subject extraction object extraction Mayr further observes that Bavarian exhibits complementizer agreement, as exemplified in (17). Complementizer agreement in Bavarian does not appear overtly on all persons: only second person singular and plural show overt agreement morphology on C0. (17) Da Franz fragt [ob-st du morgen in d’Schui geh-st] the Frank askes if-2P.SG you-2P.SG tomorrow in the school go-2P.SG ‘Frank askes if you (sg) will go to school tomorrow.’ As the following sentences suggest, complementizer agreement seems to correlate with extraction of subjects in Bavarian. That is to say, the sentences in (18) show that long-distance extraction of the subject is only possible if this subject agrees with the complementizer, whereas complementizer agreement is not fully obligatory when no subject extraction occurs (19). (18) a. b. [Es Kinda] hot da Hans gfrogt [t ob-s t you children has the Hans asked if-2P.PL ‘Hans asked if you childrem will come home.’ *[Es Kinda] hot da Hans gfrogt [t ob-Ø t you children has the Hand asked if-Ø 54 hamkummts] home come hamkummts] home come Besides this particular conceptual argument, there are other arguments, both theoretical and empirical, that favour the Copy Theory of Movement over a movement theory that assumes movement leaves a trace. See for example Hornstein et al. (2005). 34 (19) ? Da Hans hot gfroht [ob-Ø es Kinda ham kummts] the Hans has asked if you children home come ‘Hans asked if you children will come home.’ Another argument for the claim that complementizer agreement is a prerequisite for extraction of subjects in Bavarian is given in (20)-(21). When two subjects with a different specification for person – 2 singular and 3 singular – are coordinated, two agreement patterns may surface (20a)-(20b), one of which shows overt agreement (20b). When long-distance topicalization applies to the coordinated subjects in (20), only the pattern with overt complementizer agreement can surface (21b), suggesting that complementizer agreement licenses subject extraction.55 55 Actually, the data on Bavarian (coordinated) subject extraction and complementizer agreement are somewhat more complex than Mayr presents in his article. The sentences in (i) show that in Bavarian, in case of a coordinated subject, the complementizer can agree either with the first conjunct of the coordinated object (ia), or with the coordinated subject as a whole (ib); the examples are taken from Van Koppen (2005:105). As we saw, Mayr argues that extraction of the coordinated subject is only possible when there is complementizer agreement. However, sentence (iia) seems to show the opposite, i.e. there is complementizer agreement, namely with the first conjunct of the coordinated subject, but subject extraction is impossible. Van Koppen (2005) focuses on complementizer agreement with coordinated subjects and argues that when there are two agreement relations that can be spelled out on C0 (as an affix), the relation that will result in the most specific agreement affix, will be the one chosen (by Morphology). In Bavarian, two options are possible, as illustrated in (i), because, according to Van Koppen (2005), both agreement affixes are equally specific. Moreover, she argues that in case of extraction of the coordinated subject, the complementizer can never agree with the first conjunct, but must always agree with the coordinated subject as a whole (because the internal structure of a copy is not available for entering agreement relations, cf. section 4.3.2.2). Similarly, Van Koppen (2005:101) shows that in Tegelen Dutch, a dialect spoken in the north of Dutch Limburg, complementizer agreement is impossible when the subject is extracted (iiia) – because the complementizer in (iiia) agrees with the first conjunct of the extracted subject – whereas it is obligatory in case the coordinated subject is not extracted (iv). These examples thus show that the data on Bavarian subject extraction in relation to complementizer agreement are somewhat more complex. Moreover, given that Van Koppen’s (2005) claims are on the right track, these data show that the observation that the coordinated subject in (20) can only be extracted when the complementizer shows overt agreement, is actually a coincidence, i.e. if Bavarian were a language with no overt agreement on the complementizer in case of 2 person plural subjects, it would not show complementizer agreement, but it would still allow the subject to be extracted (this is corroborated by the examples in (iii)-(iv)). In sum, the data in (20)(21) do not provide direct evidence for the claim that subject extraction is licensed by complementizer agreement. (i) (ii) (iii) (iv) … das-sd du und d’Maria an Hauptpreis gwunna hab-ds that-2P.SG [you SG and the Maria]2P.PL the first.prize won have-2P.PL b. … das-ds du und d’Maria an Hauptpreis gwunna hab-ds that-2P.PL [you SG and the Maria]2P.PL the first.prize won have-2P.PL ‘… that Maria and you won the first prize’ Du und d’Maria glaub’e [you SG and the Maria]2P.PL believe.I a. * das-sd an Hauptpreis gwunna hab-ds that-2P.SG the first.prize won have-2P.PL b. das-ds an Hauptpreis gwunna hab-ds that-2P.PL the first.prize won have-2P.PL ʻYou and Maria I think that have won the first price.ʼ Doow en Marie denk ik, [you SG and Maria]2P.PL think I, a. * … de-s het spel zull-e winnen. that-2P.SG the game will-PL win b. ? … det het spel zull-e winnen. that the game will-PL win … de-s / *det doow en ich ôs treff-e that-2P.SG that [you SG and I]1P.PL each.other meet-PL ‘… that you and I will meet each other.’ a. 35 (20) a. b. (21) a. b. Da Hans hot gfrogt the Hans has asked Da Hans hot gfrogt the Hans has asked [ob du und da Franz weggengan] if you-2P.SG and the Frank leave-3P.PL [ob-s du und da Franz weggeh-ts] if-2P.PL you-2P.SG and the Frank leave-2P.PL *[Du und da Franz] you2P.SG and the Frank [Du und da Franz] you-2P.SG and the Frank hot da Hans gfrogt [t has the Hans asked hot da Hans gfrogt [t has the Hans asked ob t weggengan] if leave-3P.PL ob-s t weggeh-ts] if-2P.PL leave-2P.PL In contrast to subjects, objects can freely extract, independent of complementizer agreement (22a). (22b) shows that objects cannot agree with the complementizer. (22) a. b. [Die Bauan] hot da Hans gfrogt [t ob-s/ob-Ø es Kölna endlich t bedients] the farmers has the Hans asked if-2P.PL /if-Ø you waiters finally serve ‘John asked if you waiters will finally serve the farmers.’ *[Eich Bauan] hot da Hans gfrogt [t ob-s die Kölna endlich t bedienan] You farmers has the Hans askes if-2P.PL the waiters finally serve ‘John asked if the waiters will finally serve you farmers.’ Another property of Bavarian that has to do with the COMP-domain is the observation that Bavarian does not respect the Doubly Filled COMP filter (cf. section 1.3.3.1), as illustrated by (23). (23) Da Michael hot gefrogt [wer dass die Biancha kafft hot] the Michael has asked who that the books bought has ‘Michael asked who bought the books.’ Now consider the following sentences. The contrast in (24a) shows that a subject can only be extracted when there is a φ-agreement relation with the wh-adverb (similar to complementizer agreement). Interestingly, (24b) shows that subject extraction is also possible in case of doubly filled COMP. (24) a. b. [Es Lehrer] frogn d’Leit [t *warum-Ø /warum-s t nigs orbeits] you teachers ask the people why-Ø/ why-2P.PL nothing work ‘The people ask why you teachers do not work.’ [Es Lehrer] frogn d’Leit [t warum-Ø /*warum-s dass t nigs orbeits] you teachers ask the people why-Ø/ why-2P.PL that nothing work These data led Mayr to conclude that doubly filled COMP should be analyzed as an instance of complementizer agreement.56 Given that this assumption is on the right track, the ungrammaticality of sentences with subject extraction, doubly filled COMP and φ-agreement on the wh-adverb (24b) follows, because in those cases we are dealing with doubly marked agreement with C0. 56 Note that this argument is rather weak, as it is unclear whether subject extraction in (24b) above is licensed by doubly filled COMP or by agreement on the complementizer. This becomes even more clear when we look at the following sentences (taken from Mayr, pg. 9) in which a second person singular subject is extracted, which, in contrast to a second person plural subject (as in (24)), triggers overt agreement on the complementizer dass. Interestingly, sentence (ia) is possible, even though we are dealing with double agreement marking on C0 (i.e. doubly filled COMP and overt agreement on C0). It is thus unclear whether doubly filled COMP should be analyzed on a par with complementizer agreement, i.e. whether it should be treated as a licensing principle of subject extraction. (i) a. b. ?Du frogn d’Leit [wem dass-st t a Buach kafft host] you.2P.SG ask the people whoDAT that2P.SG a book bought have ‘The people ask, whom you bought a book.’ *Du frogn d’Leit [wem-st dass-st t a Buach kafft host] you.2P.SG ask the people whoDAT-2P.SG that2P.SG a book bought have 36 In sum, subjects in Bavarian can be extracted from embedded CPs only if they overtly agree in φfeatures with the local complementizer or with a wh-element. The extraction of objects does not require these licensing requirements. Moreover, it is assumed that when a subject is extracted, φfeatures are always realized on C0. Having established that complementizer agreement licenses subject extraction – giving rise to s/o-asymmetries – the next question that needs to be answered is what is the difference between extraction from Spec,TP and extraction from the complement position of V. Mayr argues that s/oasymmetries are a consequence of the manner in which Merge proceeds, that is to say, subjects are simply merged later in the derivation than objects, which gives rise to the contrast. He introduces the following definition of Internal Merge, i.e. attraction of a goal by a probe. (25) Internal merge IM at derivational stage Σi applies to nodes on the same projection line as the head H probing under c-command, thus a node formed at stage Σi-1, Σi-2, … Σ1, where i > 1 (26) Projection Line X and Y are on the same projection line, iff the head X selects for YP. If the head Y selects for ZP, then by transitivity X and Z are on the same projection line Given these definitions, it should be clear that objects can always undergo movement because they are on the same projection line as a higher probing head. However, when an object has moved to the specifier position of a designated head, the question is how it can move any further, since a specifier is not on the projection line of a higher probing head (i.e. it is not selected for by that particular head). Mayr assumes that objects can always undergo movement, because they are on the projection line in their external merge position. That is to say, since objects, when they are merged into the derivation, they are merged onto the projection line, they can always undergo movement. Subjects, on the other hand, are never on the same projection line as a given head because they are not selected for by any head (they are specifiers). Therefore, given the definitions in (25)-(26), subjects cannot act as goals, i.e. they can never be probed. In order for subjects to be probed, Mayr proposes (27). Given (27), subjects in Spec,TP can be extracted by virtue of agreeing in φ-features with T0. However, we need to derive that it is agreement with C0 that licenses long-distance subject extraction. Therefore, Mayr proposes (28). (27) The role of φ-features: Agreement in φ-features connects an element to the projection line, of which the agreeing head H is a part of. (28) Checking conditions at Spell-Out for long-distance extractions: a. At Spell-Out, which is induced by a phase head, e.g. C, it is checked, whether the dislocated element is on the relevant projection line due to its external merge position or not. If not, b. then a φ-relation with the phase head reconnects the dislocated element to the relevant projection line. c. Any previous φ-relation is deleted after Spell-Out. The conditions in (28) ensure that only local φ-agreement with C0 influences subject extraction. More specifically, when a subject has moved to the COMP-domain, its φ-agreement with T0 is no longer visible and as a consequence, another φ-agreement relation with C0 is established. Once the subject has entered into a φ-agreement relation with C0 – which equals being selected for by the higher V (which selects for C0) – it is on the relevant projection line and therefore, need not undergo any further φ-relations with higher phase heads.57 For objects, problems like the ones described do not arise, The following principle (Mayr, pg. 18) ensures that once the subject has entered into a φ-agreement relation with the local C0, it need not enter into further agreement relations, i.e. this principle – in interaction with 57 37 because they are always selected for, and selectional requirements need to be visible at all stages of the derivation. For structures in English from which subjects are not able to extract, it needs to be assumed that the C0 in those structures does not bear any φ-features. Moreover, for English structures in which the subject is able to extract, Mayr needs to assume that no C (-projection) is merged (cf. Rizzi & Shlonsky 2007). In those cases, the subject gets extracted from Spec,TP where it is in a φ-agreement relation with T0. The latter option is not available in Bavarian. Mayr’s account of s/o-asymmetries is particularly interesting in light of the Dutch data on s/oasymmetries in (long-distance) relativization, because in varieties that exhibit the asymmetry, a correlation (established on the basis of the geographic distribution of both phenomena) with the presence of complementizer agreement is found (generalization IV). 3.4 The analysis - subject/object asymmetries in varieties of Dutch 3.4.1 Some notes on the status of die and dat In the previous section, I discussed a very recent account of s/o-asymmetries that will form the basis for my analysis of the Dutch asymmetries. Before this analysis of s/o-asymmetries in long-distance relativization in Dutch will be presented, it is important to lay out my assumptions regarding the nature of the elements die and da(t) in the dialects under discussion. In the literature there is no agreement with respect to the status of these elements in languages like French and West-Flemish. In Standard Dutch, on the other hand, there seems to be a consensus, i.e. in restrictive relative clauses in which the head noun is non-neuter, die is taken to be a relative pronoun, and dat is the complementizer (cf. Zwart 2000, De Vries 2002). However, the status of the die/qui-elements (and to a lesser extent this holds also for the dat/que-elements) in the dialects that show s/o-asymmetries – henceforth ‘special die/qui’ – is not clear. In principle, special die/qui can be analyzed either as (i) a (weak) relative pronoun – as is assumed by traditional grammars, by Bennis & Haegeman (1984), and recently argued for by Sportiche (2008) – (ii) an agreeing form of the complementizer dat/que (Kayne 1976, Rizzi 1990), or (iii) as a contracted form (e.g. que+i=qui; Taraldsen 2001, Rizzi & Shlonsky 2007). For the dialects under discussion, I will assume that special die, found in the embedded clause of long-distance subject relativization structures (and in short subject relativization structures in WestFlemish), is not a (weak) relative pronoun as assumed by Sportiche (2008), but rather, I will follow Rizzi (1990) and hypothesize that, given the observation that special die in long-distance relativization checking condition (28c) – accounts for the fact that s/o-asymmetries in long-distance relativization structures, are always found in the clause that contains the extraction site (generalization II). (i) Secondary selection If V selects for a CP with DP in Spec,CP which agrees with the head of this CP in φ-features, then the DP behaves as if it were directly selected by V Mayr argues that if secondary selection is the case, condition (28c) cannot be applied because the subject is selected for by the higher verb, and consequently, the subject is on the projection line (permanently). Therefore, there is no need to license further extraction of the subject, i.e. no other φ-agreement relation with higher C’s is established. An alternative way to account for the fact that s/o-asymmetries are only found in the clause containing the extraction site, is by means of a feature checking/valuing approach to (complementizer) agreement (cf. note 49). C0 has uninterpretable φ-features that need to be valued, and this happens by means of agreement with the subject in Spec,TP (dialects differ with respect to the spell out of this agreement relation). Given that agreement always relates the probe to the most local goal with features that match the features of the probe, we account for the fact that only the C0 in the CP containing the extraction site shows overt φ-agreement with the relative subject DP – which is manifested as die – as the higher C’s will always agree in φ-features with the local subject in Spec,TP. This way of explaining generalization II is superior to Mayr’s account because it does not need any additional assumptions or conditions to capture the observation that s/o-asymmetries are only found in the most deeply embedded CP. 38 only appears in the most deeply embedded COMP,58 and given the observation that dialects that make use of special die also exhibit complementizer agreement (generalization IV), special die is a variant of the complementizer dat;59 more specifically, die = dat + the spell out of φ-feature agreement (cf. Mayr to appear)60. As mentioned in chapter 1, the framework I adopt assumes that Syntax only operates on feature bundles, and that feature bundles are replaced by Vocabulary Items in the Morphological component (i.e. the framework of Distributed Morphology; cf. Halle & Marantz 1993, Harley & Noyer 1999). Halle’s (1997) Subset Principle guarantees that in dialects that exhibit the distinction between complementizers die and dat – more general, in dialects that show overt agreement morphology on C0 (generalization IV) – dat will be inserted in C0, unless there is φ-agreement (die). I will come back at this particular issue in more detail in paragraph 3.4.3. 3.4.2 The analysis Let us start with system IV, the system predominantly found in West-Vlaanderen. For convenience, the West-Flemish data on relativization are repeated in (29). (29) a. b. c. d. Da s de vent that is the man Da s de vent that is the man Da s de vent that is the man Da s de vent that is the man die da graptjen ee verteld. die that joke has told da se geroepen en. that they called have da k peizen die da graptje verteld eet. that I think die that joke told has da k peizen da-n ze geroepen en. that I think that-3P.PL they called have [Brugge Dutch] Given the claim that die is a complementizer, and the observation that the s/o-asymmetry is found in both short relativization structures and in the embedded clause of long relativization configurations, we are forced to conclude that West-Flemish does not make use of relative pronouns. In other words, 58 In light of the analysis of long-distance relative clauses in Dutch, as discussed in chapter 1 (i.e. a promotion theory of relative clauses with successive-cyclic movement of the relative DP), assuming that special die in the most deeply embedded COMP of long-distance relative clauses is a (weak) relative pronoun, forces one to assume that either, for some reason, the relative pronoun is stranded in the lower COMP, or that we are dealing with scattered deletion, i.e. only parts of the highest and intermediate copy are spelled out. Both scenarios need the formulation of a specific PF-deletion theory, whereas when we assume that special die is an agreeing variant of the finite subordinate complementizer dat, we can maintain the standard assumption that (normally) only the highest copy of a chain gets spelled out (cf. Nunes 2004). In other words, given the promotion theory of relative clauses with successive-cyclic movement of the relative DP, and given the observation that die appears in the most deeply embedded clause, we are almost forced to assume that die is a complementizer instead of a relative pronoun. 59 An issue that, at least at first sight, is hard to account for when we assume die and dat are complementizers, is formed by data on relative die and dat in West-Flemish, as provided by Vandekerckhove (2003). She found that in case an object is relativized (in one clause relativization), West-Flemish dialects always make use of dat. When a subject is relativized, dialects make use of die, except when the relativized constituent is 3 neuter singular; then dat is used (in +/- 70% of the cases) – this is comparable to Standard Dutch subject and object relativization. These results indicate that not only the function of the relativized constituent is important for the choice of die/dat, but that its feature specification is also of great importance, i.e. it seems to be the case that being a neuter singular antecedent ‘overrules’ having the grammatical function of subject. Given the assumption that die and dat in West-Flemish are not relative pronouns, the question thus arises why dat is used only in case the relativized subject is 3 neuter singular. Given the analysis as presented in section 3.4.2, the element dat, as used with subjects that are neuter singular, is – similar to the element die – also an instance of complementizer agreement. The difference in form (dat vs. die) can be accounted for by the assumption that the subject needs to enter into a φ-feature agreement relation with C0, but that a different φ-feature specification yields a different output, i.e. only in case the feature specification of the relativized subject is [3 P.SG.N], the agreement relation between the (copy of the) subject and C0 is spelled out as dat instead of die. 60 Clearly, this assumption is not sufficient as the dialects under discussion exhibit complementizer agreement that does not manifest itself overtly as die. I will come back at the issue of the morphological plausibility of die as an instance of dat in section 3.4.2. 39 West-Flemish can, just like English, merge an empty DPREL61 (cf. chapter 1, section 1.3.3.1). That is to say, if West-Flemish did make use of relative pronouns – die for extracted subjects, and dat for extracted objects (based on the data on short relativization) – we would predict, given a promotion theory of long-distance relativization (cf. chapter 1, section 1.3.3.2), that the s/o-asymmetry is (also) visible in the highest clause. For subject extraction, the pattern would be die-die, in which the highest die is a manifestation of the relative pronoun and the lower die an instance of the agreeing complementizer, whereas the pattern of object extraction would be dat-dat, in which the highest dat is a relative pronoun and the lower dat the complementizer.62 This is not the pattern we find, and therefore, we need to assume that in West-Flemish the relative DP is headed by a null D. Following Mayr (to appear), the relative subject DP in Spec,TP moves to Spec,CP and enters into a φ-agreement relation with C0, which is spelled-out as die.63 By virtue of this agreement relation, the relative DP is on the projection line, and it can move further to the next Spec,CP (to check some feature on C 0, say [uWh/uRel, +EPP]) without having to enter into a φ-agreement relation with that higher C0 (see footnote 57). The higher C0 is thus spelled out as dat, and we obtain the pattern dat-die. The derivation of the West-Flemish long object relativization structure proceeds along the same line, with the difference being that since objects are on the projection line from the start, the relative object DP does not have to enter into an agreement relation with the complementizer (in fact, this is impossible), and no instances of die surface. We thus derive the pattern dat-dat. Given this analysis of the West-Flemish data on relativization, we make the following predictions: (i) special die appears only in the CP from which the subject is extracted, and (ii) since special die is an instance of dat + overt φ-agreement with the subject (copy) in Spec,CP, and since complementizers can only agree with the subject, special die is only found in case of subject extraction, never in case of object extraction. These predictions are borne out, that is to say, these claims parallel the two generalizations I started this chapter out with. The following paragraph will discuss these (and other) predictions of the analysis in more detail. Interestingly, as I claimed, West-Flemish dialects show the pattern in (29) because they spell out the φ-agreement relation with C0 and do not make use of relative pronouns. We now predict that if there is a variety that also spells out the φ-agreement relation and that makes use of relative pronouns, it should show the following pattern: no asymmetry in short relativization structures (given that the form of the relative pronoun is identical for subject and object relativization64)65, and a s/o-asymmetry in the most deeply embedded clause of long-distance relativization. This is exactly the pattern of 61 It is hard to find evidence that independently motivates this assumption because Dutch does not allow an empty CP in relative clauses (in contrast to zero-relativization in English), i.e. in Dutch, the complementizer will always be spelled out when there is no overt element in Spec,CP. 62 Although it does not occur very often, this pattern (die-die, dat-dat) is indeed attested in three locations (Strijpen, Moerzeke, Didam). A possible explanation for the small amount of dialects that show this pattern might be found in the fact that there are not many dialects that in addition to showing overt complementizer agreement, make use of only die for subjects and only dat for objects – in fact, this is only found in WestVlaanderen. The remaining dialects that show overt complementizer agreement thus often make use of both die and dat for both subject and object relatives. Such dialects thus have a ‘choice’ as to which system to use, and they more often use other systems. 63 Notice that I assume – given the observation that there is also an asymmetry in short relativization – that the relative DP enters into a φ-agreement relation with C0 independent of whether the relative DP as a whole needs to be further extracted, i.e. the φ-agreement relation with C0 is established in both short and long relativization. 64 We predict there to be dialects that do not spell out the φ-agreement relation with C0 and that make use of the relative pronouns die for subject extraction and dat for object extraction. These dialects should show the following pattern of long-distance relativization: die-dat for long subjects and dat-dat for long objects (and die for short subject relativization and dat for short object relativization). This pattern is indeed attested 3 times (in Gistel, Torhout, and Deinze). 65 Notice that in dialects that make use of the relative pronoun die (for both subject and object extraction), in short subject relativization structures, the COMP-domain is filled with the agreeing form of the complementizer die and the relative pronoun die. In none of these dialects both elements are spelled out, i.e. no strings die-die are attested, and so it is not entirely clear whether die in short subject relatives in system V is an instance of the complementizer or whether it is the relative pronoun. I will assume that it is the relative pronoun, and that some sort of haplology rule (cf. Neeleman & Van de Koot 2006) has ‘deleted’ the complementizer die. 40 system V: die-die for long subject relativization, and die-dat for long object relativization. The data that thus seemed puzzling at first sight, follow straightforwardly from this analysis. In sum, the patterns of (long-distance) relativization of the two types of dialects under discussion, are accounted for by assuming (i) that all these dialects spell out the φ-agreement relation between C0 and (the copy of) the subject in Spec,CP as die, and (ii) that some dialects make use of the relative pronoun die for both subject and object relativization (system V), whereas other dialects do not make use of relative pronouns at all (system IV). This is summarized in the following table. Table 2; properties of systems IV and V + relative pronouns (die,die) - relative pronouns + spell out agreement die-die, die-dat (V) dat-die, dat-dat (IV) An important question that is in desperate need of an answer is of course: why is φ-agreement on the complementizer spelled-out as die in dialects with complementizer-agreement, i.e. why is the ‘normal’ pattern of complementizer agreement not sufficient to license subject extraction, as it is in Bavarian? That is to say, the analysis outlined above faces the problem of morphological plausibility (cf. Taraldsen 2001): given the morphological expression of φ-features found in the dialects under discussion, and given the observation that these dialects show an agreeing form of C 0 that gives rise to a different alternation than dat-die, it is not very plausible to think of die as an agreeing form variant of the complementizer (cf. Rizzi & Shlonksy 2007). I am well aware of the fact that we need an explanation for the observation that die is an instance of complementizer agreement in the dialects under discussion.66 Such an explanation would probably have to invoke a construction (and language) specific statement – namely, one that only deals with the relative COMP-system in (varieties of) Dutch. One could for example state that φ-agreement with C0 is manifested as die only if the element in Spec,CP is (the copy of) a relative DP, i.e. the presence of complementizer die is only triggered in case of a relative clause COMP.67 Naturally, this is not a very attractive solution as it is just a description of the phenomenon, not an explanation. However, some of the predictions of the theory that will be discussed in the following paragraph suggest that this conclusion might be somewhat relaxed. 3.4.3 Predictions The analysis above makes several predictions. First, we predict the analysis of die as an agreeing variant of dat in long-distance relativization structures, to be extended to other constructions involving long wh-dependencies. More specifically, languages that make use of special die in (long) relative clauses are predicted to make use of the construction in (30a) when forming long-distance whquestions. In (30a) the question is introduced by the wh-word wie ‘who’, and the embedded clause is introduced by die, which, given the analysis above, is taken to be a complementizer spelled-out as die by virtue of φ-agreement with the copy of the wh-subject wie. 66 Moreover, we need to account for the observation that the relative pronoun is homophonous to (the agreeing form of) the complementizer. It is often noted that there is a striking parallelism between the DP and the clausal structure. Cardinaletti & Starke (1999) therefore argue that the most straightforward way to capture this parallelism is to assume that both DPs as well as CPs realize the same structure. More specifically, they argue that the DP consists of a CP-like layer and an IP-like layer (i.e. they assume the following projections for strong pronouns: CLP – ΣLP – ILP – LP, with L any lexical category and Σ0 a nominal support morpheme). Given this assumption, the fact that the relative pronoun die is homophonous to the (agreeing form of the) complementizer is not so strange, as the head of the DP-projection is – similar to the head of the clausal structure – taken to be C0. In other words, according to Cardinaletti & Starke (1999) pronouns bear C-features. 67 An important problem we face when assuming that the presence of die is triggered in case there is a relative COMP, is that in fact the most deeply embedded COMP, selected by the verb denken ‘to think’ is [-Q] and we thus predict there to be no agreement relation possible between the [+Q] relative DP and the [-Q] C0. It should be clear that we need some additional mechanism that ensures the embedded CP becomes [+Q]. I will not discuss the various possibilities. 41 (30) a. b. Wie denk je die het verhaal verteld heeft? who think you die the story told has ‘Who do you think told the story?’ Wie denk je dat het verhaal verteld heeft? who think you that the story told has Unfortunately, in the SAND-project sentences like (30a) were only questioned with a wh-object, and not with wh-subjects. Regarding construction (30a) with a wh-object, the SAND data show that its distribution is very limited: it is only attested 8 times (cf. Barbiers et al. 2005).68 As my analysis would predict the construction to be impossible, we could state that the 8 occurrences of the construction are a counterargument to my claim. However, I think it is striking that the construction is only found incidentally in the Netherlands, and never in Belgium, and this becomes even more remarkable when we take into consideration the observation that the construction with a wh-subject as in (30a) actually does occur in Flanders. That is to say, at first sight, the prediction that the construction in (30a) should be used by dialects that exhibit special die in (long-distance) relativization, seems correct given the data on long-distance A’-dependencies in Dutch reported by Schippers (2006). The construction in (30a) is indeed attested in West-Vlaanderen, and several times in Oost-Vlaanderen.69,70 Importantly, however, this construction is certainly not the only possible one found in this area; the construction in (30b) – the Standard Dutch variant with a complementizer introducing the subordinate clause – is also attested several times in Flanders. The latter result might lead one to suggest that long-distance whquestions and long-distance relativization structures cannot be analyzed on a par.71 In short, the observation that dialects that show the asymmetry, use both (30a) and (30b), is not a decisive counterargument to my claim. Rather, it suggests that, without additional machinery, the analysis of die as a form variant of the complementizer cannot be extended to structures involving long-distance wh-movement. The precise distribution of the different long-distance wh-movement structures in relation to long-distance relativization should be the subject of future research. Second, given that die is (an agreeing) instance of the complementizer, we predict that languages that show the s/o-asymmetry – and thus make use of special die – never allow doubly filled COMP in special die contexts. That is to say, we predict the dialects that show the asymmetry to never allow strings of die-dat in the most deeply embedded clause of long subject relativization,72 and in short subject relativization structures. The SAND data show that of the 39 dialects that make use of 68 The test sentence with a wh-object is: Wie denk je die ik in de stad heb gezien? This test sentence was almost exclusively presented to the informants in Belgium. In the Netherlands, there are some attestations of the construction because some informants spontaneously produced this sentence when they were presented with another test sentence (cf. Barbiers et al. 2005). Since the construction is not questioned in the Netherlands, and as it might very well be the case that the construction occurs (more often) in this language area, future research seems necessary to see whether this is indeed the case, i.e. whether the amount of occurrences of the long whobject construction with die in the most deeply embedded Spec,CP is actually larger than 8. 69 Schippers (2006) carried out a small postal survey in which, amongst others, test sentence (30a) was presented to informants in Groningen (7), Drenthe (5), West-Vlaanderen (8), Oost-Vlaanderen (6) and Vlaams-Brabant (7). Construction (30a) was attested in none of the locations in Groningen and Vlaams-Brabant, but is was found in 2 out of 4 locations in Drenthe, in 4 out of 5 locations in West-Vlaanderen (5 out of 8 informants indicated that the sentence occurs in their local dialect), and in 2 out of 4 locations in Oost-Vlaanderen. 70 Moreover, as noted by Sportiche (2008:12), sentences like (30a) are acceptable in Nijmegen Dutch. This is exactly the area in which several attestations of subject/object asymmetries were attested (cf. map 1). 71 A similar conclusion is also suggested by other SAND data on long-distance wh-dependencies, and the data gathered by Schippers (2006). Moreover, structures involving (long-distance) wh-movement and (long-distance) relativization structures have different semantic properties (e.g. reconstruction effects), suggesting that they should receive a different analysis (cf. Salzmann 2006, Koopman & Sportiche 2008). 72 Of course, we also predict that strings like dat-dat never occur in the most deeply embedded clause of longdistance relativization structures. However, this prediction is hard to test as the absence of such strings is probably independently motivated by a mechanism that disallows (accidental) repetition of identical morphemes (cf. Neeleman & Van de Koot 2006). 42 special die in (long-distance) relativization, there is only one dialect that uses a complementizer in addition to die: the dialect spoken in Moerzeke, in Oost-Vlaanderen.73 This is illustrated in (31). (31) a. b. … die at tat verteld eet die that that told has … da k peis die dat ta verteld eet that I think die that that told has [Moerzeke Dutch] Although one should come up with an explanation for the grammaticality of the string die-dat in this particular dialect, it should be clear that the prediction is largely borne out (but see footnote 73). Interestingly, according to Haegeman (1992:57), in West-Flemish argument clauses, the complementizer da is always overt, independent of the presence of an additional constituent in Spec,CP. This is exemplified in (32)-(33). (32) a. b. (33) a. b. Kpeinzen da Valère a weg is. I think that Valère already away is ‘I think that Valère is already gone.’ *Kpeizen Ø Valère a weg is. Kweten niet wannièr da Valère goa werekommen. I know not when that Valère goes return ‘I do not know when Valère is going to return.’ *Kweten niet wannièr Valère goa werekommen. [Haegeman 1992:57] Given this observation, it is particularly telling that in West-Flemish (long-distance) subject relative clauses the complementizer da is not present in the embedded clause. My analysis that die is in fact an agreeing form of the complementizer, fits in nicely with the generalization that in West-Flemish the complementizer cannot be deleted. Note that if we were to assume that die is a (weak) relative pronoun (cf. Sportiche 2008), we would have to come up with an explanation for the observation that in relative clauses the complementizer does not need to be overt.74,75 A third prediction we make involves the extension of the above analysis to other language varieties. Given the claim that in case of subject extraction the complementizer always needs to enter into an agreement relation with the (copy of) the element in its specifier position, we predict languages that do not show an overt s/o-asymmetry, but do allow subjects to be extracted, either have no C0 merged in such structures (as Mayr proposes for English), or the agreement relation is always established in case subjects are extracted, but this relation is only visible in dialects that show complementizer agreement (generalization IV). I will assume the latter, and propose that 73 It is important to note that not all informants were explicitly asked about the potential presence of an additional complementizer (only in Hooglede (West-Vlaanderen), the interviewer explicitly asked whether die da instead of die could be used, and the informant gave a negative response). Therefore, the prediction that a complementizer cannot be present in special die contexts cannot be adequately tested on the basis of the SANDcorpus. Moreover, the data are less clear than presented in the main text. That is to say, it is not entirely clear whether the dialect spoken in Moerzeke is the only dialect that makes use of doubly filled COMP, i.e. at least at first sight, the dialects spoken in Koewacht and Lokeren also seem to allow doubly filled COMP. Needless to say that further research into the issue of doubly filled COMP in relative clauses seems necessary. 74 Haegeman (1983) observes for the West-Flemish dialect of Lapscheure that relatives do not allow doubly filled COMP, whereas interrogatives do. She argues that the absence of doubly filled COMP in relatives is due to Complementizer Contraction (cf. Pesetsky 1982), which results in deletion of the complementizer in long subject relatives and deletion of the relative pronoun in long object relatives. 75 The validity of this argument regarding short relativization is dependent on the analysis of relative clauses one assumes, i.e. the assumptions about the relationship between the moved head and the relative clause (complementation or adjunction) are of great importance (cf. chapter 1). If one takes the relative clause to be an adjunct of the relative head noun (as is assumed in the ‘traditional adjunction analysis’), the above argument does not hold. However, when one assumes, as I do, that the relative clause is a complement, i.e. the so-called Dcomplement hypothesis (cf. Kayne 1994 and De Vries 2002 amongst others), the above argument holds. 43 complementizers in (varieties of) Dutch (that appear without overt agreement) are always specified for φ-features (cf. Haegeman 1992:53). This explains the fact that there are no (overt) s/o-asymmetries in other varieties of Dutch, but raises the question of why the complementizer is realized as dat, and not as die in these varieties, since the agreement relation is still established. The most straightforward way to account for this is to assume that in dialects without the overt s/o-asymmetry, no complementizer die is listed in the lexicon, and dat (as default) will always be inserted. To illustrate this, suppose that the embedded C0 in the relative clauses under discussion has the following feature specification: [3P.SG.M, +Q/rel]76, i.e. the uninterpretable φ-features on C0 are valued by the relative DP in Spec,CP, giving rise to the specification [3P.SG.M], and C0 also has a Q/rel-feature due to agreement with the relative DP in its specifier position. Once the syntactic structure is built and the level of Morphology has been reached, this feature bundle needs to be replaced by a Vocabulary Item. Given that the features of die and dat are as given in (34)77, in dialects that have die specified in the lexicon as a complementizer (i.e. we could assume that die also has some sort of C-feature), both Vocabulary Items meet the conditions for insertion as both items match (a subset of) the features of the grammatical features specified in C0. However, as die has more features that match the features in the terminal morpheme C0, according to the Subset Principle (cf. Halle 1997:428), die instead of dat should be inserted. (34) a. b. die da(t) [3P.SG.M, +Q/rel]78 [3P.SG.M] Notice that from this way of accounting for the difference between dialects that show overt φ-feature agreement between C0 and (the copy of) the DP in Spec,CP and dialects that do not, it follows that this difference is rather superficial, in the sense that in some dialects die is specified as a ʻspecial relative clause complementizerʼ whereas in other dialects it is not and dat – as the default complementizer – will always be inserted. In other words, the microvariation regarding s/o-asymmetries between dialects can be reduced to variation in the lexicon (i.e. the Vocabulary Item lexicon). Finally, given that the presence or absence of relative pronouns is a micro-parameter, and given that the spell out of φ-agreement with C0 is another micro-parameter, we predict the existence of the following six systems of long-distance relativization. This prediction is borne out: all six systems are attested (see also footnotes 62 and 64). Notice that with these two rather simple parameters, we generate, in addition to the patterns of systems IV and V, two other systems (of the six main systems of long-distance relativization) that were presented in chapter 2, namely system I (Standard Dutch) and system VI. The following chapter will focus in more detail on these two patterns of relativization. Table 3; two micro-parameters - relative pronouns + relative pronouns (die, die) + relative pronouns (die, dat) + spell out agreement dat-die, dat-dat (IV) die-die, die-dat (V) die-die, dat-dat (3) 76 - spell out agreement dat-dat, dat-dat (VI) die-dat, die-dat (I) die-dat, dat-dat (3) For the present discussion, the exact nature of the feature +Q/rel is not important. Relevant for the present discussion is the assumption that the embedded C0 has – in addition to φ-features – a specific feature related to (at least) relative clauses (and perhaps also to wh-clauses; see the discussion regarding the first prediction of my analysis in section 3.4.3). 77 For convenience, I have specified the φ-features of die and dat to show that they match the φ-features of the embedded C0. However, it should be clear that in the lexicon, the item dat (and also die) does not have the specific feature specification of (34a); rather, it will be underspecified as it can also be used with other φ-feature combinations, e.g. de mannen3P.PL.M die ik denk dat het gedaan hebben ʻthe men that I think have done itʼ. 78 Due to the presence of the +Q/rel-feature on die, this item will never be used in case of object relativization (generalization III) because the relative object DP will never agree with C0. Rather, the local subject that is not endowed with a +Q/rel feature will agree with C0. Moreover, the assumption of the presence of the +Q/relfeature on die, allows us to account for the fact that ʻnormalʼ complementizer agreement in dialects such as Lapscheure Dutch (cf. Haegeman 1992) is never manifested as die. 44 In addition to making predictions regarding the existence of systems of long-distance relativization, the two formulated micro-parameters – and more generally, the analysis that was made explicit in the previous section – also predict the non-existence of several patterns of long-distance relativization. To see which patterns these are, recall the generalizations I started this chapter out with; for convenience these are repeated as (35). (35) II III a subject/object asymmetry can appear only in the CP containing the extraction site (i.e. a subject/object asymmetry in the higher clause is (almost) never attested) in case of a subject/object asymmetry, dat occurs in the most deeply embedded clause when the object is extracted, whereas die occurs in the most deeply embedded clause in case of subject extraction The analysis and micro-parameters proposed in the previous paragraph account for both generalizations. That is to say, generalization II is explained as follows: once the subject has entered into an φ-agreement relation with C0, it is on the relevant projection line and therefore, need not undergo any further φ-agreement relations with higher heads. Generalization III is accounted for by the assumption that only subjects need to agree with C0 in order to be on the projection line and get extracted; objects, on the other hand, are on the projection line from the start. So, patterns that violate one or two of the generalizations in (35) are excluded by the proposed analysis and micro-parameters. The following table shows whether or not the patterns that are predicted to be excluded/non-existent are attested in the SAND data on long-distance relativization. Notice that although patterns die-dat– dat-dat and die-die–dat-dat violate generalization II (and obey generalization III), they are correctly predicted to exist under my analysis (cf. table 3), and therefore, they are not included in the table. Table 4; patterns of long relativization that are excluded by the proposed analysis long subject long object gen. II gen. III die-dat-Ø dat-dat-Ø die-die-Ø dat-die-Ø dat-dat-Ø dat-die-Ø dat-dat-Ø die-dat-Ø die-die-Ø dat-die-Ø dat-die-Ø die-die-Ø die-die-Ø die-dat-Ø die-dat-Ø dat-die-Ø + + - + - # attestations in SAND corpus79 2 0 0 1 0 0 1 0 As the table shows, the patterns that are predicted to be non-existent are not or only very marginally attested in the SAND data. It can thus be concluded that the predictions regarding the existence and non-existence of particular patterns of long-distance relativization are borne out. 3.4.4 Some problems – special properties of long subject relativization Koopman & Sportiche (2008) argue – on the basis of the observation that the properties of French contexts that allow special qui (‘special contexts’) are different from the properties of (French) structures that traditionally are analyzed as involving successive cyclic wh-movement (‘bridge contexts’) – that French special contexts need to be analyzed on a par with pseudo relative small 79 Notice that the amounts in this column are different from those in the comparable tables 1 and 3 in chapter 2, which present all the possible combinations of long subject and long object relativization and the amount of dialects in which these combinations are attested. Whereas in the tables in chapter 2 the amounts – that were originally lower than 5 (cf. footnote 42) – reflect whether or not a given dialect uses a particular long subject relative and a particular long object relative, the amounts in the table in this chapter show how many dialects make use of a particular pattern. That is to say, this column shows how many dialects make use of a particular pattern for sure; thus, only dialects that do not make use of other subject and object relatives are taken up in this table. The reason for this is that it is often uncertain whether a particular dialect makes use of a given pattern or not, as dialects often use more than one construction for subject and/or object relativization. 45 clauses.80 For French, they present a comprehensive list of the properties of special contexts, among which the following: (i) the predicate of the matrix clause must belong to a particular subset of predicates, and (ii) quantifiers, negative quantifiers and sentential negation intervening between the wh-operator and the instance of special qui in the most deeply embedded clause, causes a severe degradation in acceptability. Interestingly, Koopman & Sportiche show that both properties also seem to hold for West-Flemish (i.e. the dialect of Lapscheure, cf. Haegeman 1992). This is illustrated in the following sentences; the West-Flemish sentences in (36) show that intervention of the negative quantifier niemand ‘nobody’ degrades the acceptability of long subject relatives (36a), but is perfectly fine with long object relatives (36b). (36) a. b. ??* … de studenten dat er niemand zeid die-n do geweest oan the students that there nobody said die-PL there been was ‘… the students, noone said had been there’ … de studenten dat er niemand zeid da-j moet contacteren the students that there nobody said that-2nd must contact ‘… the students that nobody said you must contact’ With respect to the sensitivity to the type of predicate in the matrix clause, (37) shows that WestFlemish excludes special die in the context of a desiderative verb. (37) a. b. ??* … de studenten daj zou willen dien de secretaresse ipbellen the students that.you would want die.PL the secretary upcall.PL ‘… the students that you would want to call the secretary’ … the studenten daj zou willen da de secretaresse ipbelt the students da.you would want that the secretary up-calls The question that arises is how to account for these special properties of the long-distance relativization structure. It should be noted that Koopman & Sportiche (2008) themselves do not have an explanation for most of the properties of special contexts. In fact, they only offer an explanation for the predicate restriction on special contexts (cf. note 81). As regards this property, we could invoke the selection requirements of the verb (that selects for C) in explaining the properties of special contexts, and assume that some verbs simply cannot select a CP introduced by the agreeing complementizer die. That is to say, at some point in the derivation, the matrix predicate needs to be merged with its CPcomplement, but is unable to do so, causing the derivation to crash. Since objects never agree with the complementizer, the predicate restriction does not hold for object relativization. Naturally, this account of the facts is just a description, not an explanation, and the properties of special contexts still need to be adequately accounted for. Whether these are best explained by the analysis proposed by Koopman & Sportiche (2008), or by an analysis such as the one proposed in this thesis, needs to be the subject of future research. 3.4.5 Intermediate summary In the foregoing, I have, following Mayr (to appear), proposed an analysis of the s/o-asymmetries in long-distance relativization structures in varieties Dutch. One of the starting points for this analysis was the observation that Dutch dialects that show the asymmetry, often also show complementizer agreement, i.e. overt agreement morphology on C0 (the two phenomena have more or less the same geographic distribution; generalization IV). The analysis is fairly simple and basically states that subject extraction is licensed by complementizer agreement. The analysis predicts the existence of the two attested patterns of s/o-asymmetries in Dutch, it is capable of accounting for the observation that special die is only found with subject relativization (generalization III), and it can account for the fact that asymmetries are never found in the higher clause, but always in the clause that contains the extraction site (generalization II). Although the analysis is able to account for these main For the details of their analysis, I refer the reader to Koopman & Sportiche’s (2008) paper (and see footnote 81). Relevant for the present discussion is that an alternative to the successive cyclic wh-movement analysis of long-distance relativization has been proposed. 80 46 generalizations regarding the s/o-asymmetries, it also has its shortcomings, most prominently the problem of the morphological plausibility of die as an agreeing variant of the complementizer. Moreover, the analysis is in need of some modifications in order to account for some special properties of long-distance subject relativization structures. Therefore, in the following paragraph, I will discuss some alternative proposals of s/o-asymmetries to the analysis outlined above, and I will show that each of them also has its own shortcomings. The theories I will discuss are the T-to-C movement approach developed by Pesetsky & Torrego (2001), and a theory recently formulated by Rizzi & Shlonsky (2007).81 81 Although the theory proposed by Koopman & Sportiche (2008) is appealing in many respects, I have chosen not to outline their approach in this chapter because it is primarily based on French relative clauses and pseudo relative small clauses (PRSC). Given that it is unclear whether Dutch/Flemish has PRSCs (i.e. PRSCs in Dutch/Flemish are of different types than those in French), it is hard to argue – without some research into the properties of PRSCs in Dutch/Flemish – that special die contexts in Dutch/Flemish should be analyzed as involving PRSCs. In addition, Koopman & Sportiche’s theory makes some wrong predictions with respect to the possible patterns of subject/object asymmetries in Dutch. To illustrate this, I will briefly discuss their proposal in this footnote. Koopman & Sportiche show that long-distance ‘special contexts’ share properties with pseudo relative small clauses (cf. section 3.4.4), of which an example is given in (i) – the pseudo relative is italicized. (i) Julie a rencontré [PRSC Hélène qui se promenait] Julie met Helen who was taking a walk Koopman & Sportiche furthermore argue that extraction of wh-subjects out of tensed CPs is impossible in French. Therefore, ‘special contexts’ cannot be analyzed as involving extraction of the subject from the embedded tensed CP (i.e. the ‘standard’ analysis of the structure); rather, it is assumed that ‘special contexts’ are derived from PRSCs. More specifically, ‘special contexts’ are analyzed as introduced by EECM verbs that take tensed PRSCs, as indicated in (ii) – EECM verbs are ECM verbs of which the exceptional case marking option is limited to contexts in which the ECM-marked DP is wh-moved. (ii) WHK PRED/EECM-verb [PRSC SUBK [CP-REL qui/die … ]] As (ii) shows, the wh-subject, instead of moving out of the embedded CP by wh-movement, has wh-moved from the subject position of the PRSC. Case on WHK comes from the EECM verb, and therefore, it is accusative instead of nominative. Since, according to Koopman & Sportiche, in Dutch – in contrast to French – special die is also found in long object relatives, they argue that all instances of special die in Dutch should be analyzed as a kind of pseudo relative structure, as exemplified in an abstract manner in the following table. subject object de NP de NP Highest COMP [dACC [ [dACC [ Special context …V …V SUB [<dACC>[ [<dACC>[ Predicate [<dNOM> [<dACC> <dNOM> … <dACC> This table shows that the higher d-form is always accusative, independent of the function of the extracted element. While this explains my generalization II, it cannot account for the observation that there are some dialects that make use of special die in the most deeply embedded clause of long subject relatives and that use dat as the accusative d-form (short relativization), while showing the pattern die-die for long subject relativization (see table 3). Koopman & Sportiche would (incorrectly) predict the pattern dat-die. Moreover, the analysis predicts that only languages that make use of the relative pronoun die for both nominative and accusative elements, can show the double die pattern (die-die). Languages that only make use of die with nominative elements, should show the pattern dat-die. However, recall the pattern of system VIII in which die is used for short subject and object relativization, whereas long relativization has the form dat-die. Thus, contrary to what is found, a system like our system VIII is predicted to be impossible by Koopman & Sportiche’s account of special qui/die contexts. It should be noted however, that further research (into the exact patterns that different Dutch dialects display) is needed to check whether the account of Koopman & Sportiche really makes the wrong predictions. That is to say, not all dialects that make use of system VIII for long relativization, only use die for short object relativization. Sometimes, also dat is used, in which case Koopman & Sportiche’s account makes the correct predictions. Similarly, dialects with pattern die-die - dat-dat, not always use only die for short subject relativization and dat for short object relativization. In short, although the proposal by Koopman & Sportiche seems to make the wrong predictions, further research is necessary to check whether this is indeed the case. 47 3.5 Alternative accounts In this section I will discuss two alternative accounts of s/o-asymmetries. The first account I will discuss is the T-to-C movement approach, introduced by Pesetsky & Torrego (2001). This approach focuses on s/o-asymmetries in general, and basically states that such asymmetries are the result of different feature specifications on subjects and objects, namely [±uTense]. Besides Pesetsky & Torrego’s general approach, I will outline a proposal by Gallego (2004) that extends this approach to relative clauses. The second analysis of s/o-asymmetries I will discuss is a rather recent analysis proposed by Rizzi & Shlonsky (2007). They assume that subject extraction is prohibited by ‘criterial’ requirements, and that, following Taraldsen (2001), the French element qui found in the que/qui alternation is a contracted form of the complementizer que and a numberless expletive element i. 3.5.1 Pesetsky & Torrego (2001), Gallego (2004) Pesetsky & Torrego (2001) explain the that-trace effect and other subject/non-subject asymmetries in English as the result of T-to-C movement. Their main claims are given in (38)-(40). (38) motivation for T-to-C movement C bears an uninterpretable T feature (henceforth uT) with the EPP property82 (39) the nature of nominative Case nominative Case is uT on D (40) the nature of that that is not C, but a particular realization of T moved to C Pesetsky & Torrego thus argue that C0 has an [uT, +EPP] feature, which has to be deleted by overt movement. There are three scenarios in which this feature can be deleted: (i) by movement of an auxiliary, (ii) by movement of that, and (iii) by movement of the nominative subject.83 The latter possibility follows from the empirical observation that no T-to-C movement is observed when a nominative subject moves. Consider the sentences in (41). (41) a. b. Who do you think (that) Sue met t? Who do you think (*that) t met Sue? In this framework, the that-trace effect in (41) is explained as follows. In these sentences C not only bears [uT,+EPP], but also [uWh, +EPP]. Through movement of the nominative subject bearing uT and iWh, C’s [uT, +EPP] feature as well as its [uWh, +EPP] feature are deleted, and no further movement is required. The obligatory absence of the complementizer that is thus explained by economy: a head triggers the minimum number of operations necessary to erase its uninterpretable features. In contrast, an object, given that it does not bear uT, is incapable of deleting C’s [uT, +EPP] feature when in moves to the COMP-domain, and therefore, T-to-C movement is required to satisfy the interface requirement stating that uninterpretable features need to be erased by the end of the derivation. The apparent optionality of that in (41a) reflects the choice between T-to-C movement (that surfaces in the COMP-position) and subject movement to Spec,CP (no that).84 82 When a head bears a feature with an EPP property, overt movement is triggered. When a feature on a head does not have the EPP property, overt movement does not have to apply and by economy, it must not. In that case the uninterpretable feature of a probe and a corresponding feature of a goal enter into an Agree relation by which the uninterpretable feature gets erased. The effect of the [±EPP] property on features is thus identical to the strong/weak feature distinction, i.e. strong features trigger overt movement, and weak features do not. 83 Note that Pesetky & Torrego (2001) assume that uninterpretable features are capable of deleting other uninterpretable features, e.g. uT on the nominative subject can delete C’s uT feature. It should be noted that this is not an entirely uncontroversial assumption, i.e. it is commonly assumed that uninterpretable features can only be deleted by interpretable features (uFeatures attract iFeatures). 84 In a more recent article, Pesetsky & Torrego (2004) try to expand and improve their T-to-C movement account, and assume that instead of only nominative Case being an instance of uT on D, all instances of structural Case are actually instances of uT on D. They furthermore argue that uT on a complement of the verb 48 Gallego (2004) puts forward an analysis of relative clauses in Romance languages (and English) based on (i) the raising analysis of relative clauses (cf. chapter 1) and (ii) Pesetsky & Torrego’s (2001) proposal, as briefly outlined above. The main data he wants to account for are presented in (42). They involve the absence of that-deletion (42a) and the absence of relative pronouns in Romance languages (42b), i.e. whereas English allows wh-relatives, Romance must introduce overt relative Ds by a preposition. (42) a. b. El hombre *(que) vi. the man that see-PST.1SG ‘The man (that) I saw.’ El hombre *(con) quien habló. the man (with) who talk- PST.3SG ‘The man who talked.’/ ‘The man to whom (s)he talked.’ [Spanish] Basically, Gallego (2004) explains the data in (42) by making appeal to the notion of phases (cf. Chomsky 2000,2001; see also footnote 35). He assumes, following Pesetsky & Torrego (2004:516), the following ‘principle of cyclic derivational dynamics’. (43) Timing of Deletion of Uninterpretable Features An uninterpretable feature [uF] marked for deletion, (i.e. [uF]) within a completed phase P, is deleted the moment a new head H is merged to P. Moreover, he assumes that C0 is endowed with an uninterpretable relative feature [uRel, +EPP], in addition to the [uT, +EPP] feature. This [uRel] feature is similar in nature to the ‘standard’ [uWh] feature. In order to account for the fact that the relative head appears in front of the relative pronoun, the presence of an additional head between CP and the external D is assumed. 85 This head, c – analogous to v in the VP-system – is endowed with uninterpretable φ-features with the EPP property, and thus attracts the closest constituent bearing [iφ], the head noun. English wh-relatives are then derived as follows. (44) a. b. … (the) man who left [cP mank [c’ c [uφ, EPP] [CP [who tk]i [iRel] [uT] [C’ [uT, EPP][uRel, EPP] [TP ti left ti ]]]]] Consider now the English that/zero-relative clauses in (45). Interestingly, contrary to what one would expect, when English subjects are relativized, they do not show a that-trace effect, i.e. in subject relativization structures that is obligatory. (45) a. b. … the boy *(that) called Mary … the boy (that) Mary called Gallego accounts for this pattern by assuming that (i) that/zero-relatives are analyzed as involving a null relative D (cf. Bianchi 1999), (ii) non-overt operators cannot pied-pipe lexical material (cf. Chomsky 2001) and (iii) relative DPs, when headed by a null D, obligatorily remain in their first merge-position. Given these assumptions, sentence (45a) will be derived as follows. The relative head moves to Spec,c to delete c’s uninterpretable φ-features. C’s uninterpretable features are deleted by T- enters an Agree relation with a second occurrence of T that is structurally lower than the main Tense of the sentence. Apparent distinctions between V, N and A reflect distinctions on this lower T. In short, Pesetsky & Torrego (2004) offer a general theory of complementation patterns, based on the theory of Case and the distribution of clausal complements. Since the main claims of the two papers are identical, for the present discussion, it is sufficient to outline Pesetsky & Torrego (2001). 85 This assumption is comparable to Zwart (2000), who argues that the left periphery of Dutch relative clauses consists of more than one CP. The highest C0 triggers movement of the head noun out of the relative DP to its specifier position (cf. also Bianchi 1999). 49 to-C movement ([uT] is erased), and by Agree between [uRel] and the null relative D.86 This derivation is schematically illustrated in (46). (46) [cP boyk [c’ c [uφ, EPP] [CP thati C[uT, EPP] [uRel] [TP Ti [vP [DP Drel tk][iRel] called Mary]]]]] English object relatives are derived in the same manner, as shown in (47)-(48). (47) a. b. c. ... the car [CP which John sold] ... the car [CP that John sold] ... the car [CP Ø John sold] wh-relative that-relative zero-relative (48) a. b. c. [DP the [cP carj c[uφ, EPP] [CP [which tj]i [iRel] John C[uRel, EPP][uT, EPP] [TP tz sold ti]]] [DP the [cP carj c[uφ, EPP] [CP thati C[uRel] [uT, EPP] [TP John Ti sold [DP Drel tj] [iRel] ]]]]87 [DP the [cP carj c[uφ, EPP] [CP Johni C[uRel][uT, EPP] [TP ti sold [DP Drel tj] [iRel] ]]]] As the derivations in (48b-c) show, the available candidates to delete C’s [uT, +EPP] feature are T itself – giving rise to an instance of that (47b) – and the subject DP (48c). However, in English whrelatives, only the latter possibility, i.e. movement of the nominative subject, yields a correct outcome.88 (49) a. b. *... the car which that John sold ... the car which John sold Given this analysis of English relative clauses in terms of Pesetsky & Torrego’s (2001) T-to-C movement account, the question that now needs to be answered is how to account for the Romance pattern in (42) we started out with: the absence of that-deletion and the absence of relative pronouns without prepositions. Gallego argues that in Romance languages, subject DPs cannot check C’s [uT, +EPP] feature because their own [uT] feature has already been deleted. That is to say, it is assumed that English differs from Romance in that in the latter TP instead of vP is a strong phase. Given the principle in (43), we now see why the subject DP in Romance languages cannot check C’s [uT] feature: its [uT] feature is deleted and has become inert for further computation by the time it reaches the COMP-domain. When it is impossible to attract a subject DP, the second most economical option is to attract a PP, i.e. prepositions are assumed to be species of T (cf. Pesetsky & Torrego 2004). When C0 attracts a PP, all its uninterpretable features can be deleted at once: P deletes C’s [uT] and the relative D deletes C’s [uRel]. So, the locus of variation in relative clause formation in English and Note that claiming that C’s [uRel] is deleted by Agree with the null operator, implies that C’s [uRel] feature is not endowed with the EPP property (cf. footnote 82). What this means is that the [uRel] feature on C0 of that/zero-relatives does not have the EPP property, whereas the [uRel] feature on C0 of wh-relatives does have the EPP property (cf. (44b)). The same seems to hold for object relatives (cf. (48)). This raises the question of why this is the case: why is the EPP property of [uRel] only licensed in case there is a wh-element present? I will not try to attempt to answer this question here. 87 Given that c (only) bears an [uφ, +EPP] feature, and given the Attract Closest Principle (Chomsky 1995) – if a head K attracts X, no constituent Y is closer to K than X – the derivations in (48a) and (48b) actually violate this principle and we predict these sentences to be out. That is to say, given that the subject also bears an [iφ] feature, the subject instead of the object is the closest goal for c’s φ-probe and c therefore, should attract the subject, quod non. To circumvent this unwarranted conclusion, we could assume, following Zwart (2000), that c attracts the head noun for semantic reasons (i.e. it needs an element in its specifier position to function as restrictor). 88 Some varieties of Dutch seem to allow (49a), in which we have both an instance of the relative pronoun die and an instance of the complementizer dat (i.e. doubly filled COMP) as illustrated by (i). 86 (i) Da s de man die dan ze geroepen en. that is the man who that-3P.PL they called have ‘That is the man who they have called.’ 50 [Lokeren Dutch; Oost-Vlaanderen] Romance languages is found in the notion of phase, i.e. phases do not seem to behave alike crosslinguistically. With Pesetsky & Torrego’s (2001,2004) alternative theory to account for s/o-asymmetries (in English) and Gallego’s (2004) expansion of this theory to the domain of relativization, we have arrived at the question whether this theory can be used in explaining the s/o-asymmetries in varieties of Dutch. The theory gives us a tool to attribute the attested variation in (long-distance) relativization structures to differences in the feature composition (of C-heads) in different dialects. Let us assume (i) a raising analysis of relative clauses (see chapter 1) and (ii) Dutch dat is, analogously to the English that, an instance of T-to-C movement, and (iii) vP is a phase in the dialects under discussion.89 Consider again the West-Flemish data on relativization in (50). (50) a. b. c. d. Dat is de man [die __ het verhaal verteld heeft] Dat is de man [dat ze __ geroepen hebben] Dat is de man [dat ik denk [die __ het verhaal verteld heeft]] Dat is de man [dat ik denk [dat ze __ geroepen hebben]] At first sight, it seems to be the case that the West-Flemish data on short relativization (50a)-(50b) can be easily accounted for with the T-to-C movement approach. If we assume that C has the following features: [uT, +EPP], [uRel, +EPP], the pattern in (50a) can be explained by movement of the relative subject DP to delete C’s [uRel] feature; this subject movement at the same time deletes C’s [uT] feature. Since the relative object in (50b) is incapable of deleting C’s [uT] feature, in addition to movement of the relative object DP to delete C’s [uRel], T-to-C movement takes place, yielding an instance of dat. For convenience, this derivation of (50a)-(50b) is schematically illustrated in (51) (51)90 Features of C0 subj. rel. (50a): Deletion of [uT] Deletion of [uRel] obj. rel. (50b): Deletion of [uT] Deletion of [uRel] [uT, +EPP], [uRel, +EPP] by subject movement by movement of the relative DP by T-to-C movement (dat) by movement of the relative DP It needs to be noted that in order for the derivation of (50b) to work, we need to stipulate that in object relatives there is no relative pronoun present, otherwise, we would predict object relatives to surface with both a relative pronoun die as well as an instance of dat in the COMP-domain. What about long-distance relativization in West-Flemish? Let us start with the long object relative and assume that in such relatives no relative pronoun is present. As we discussed, Gallego assumes that in case of relative DPs headed by a null D, the relative head moves up on its own, stranding the null D in its base-generated position. Given that objects can never delete C’s [uT] feature, in long-distance object relativization, T-to-C movement is triggered and we find an instance of dat in the most deeply embedded sentence. The object relative head now moves further to the specifier position of the higher COMP-domain. Again, additional T-to-C movement is triggered in order to delete C’s [uT] feature, yielding another instance of dat. Finally, the relative head moves to Spec,cP to check c’s φ-probe. So, except for the assumption that object relatives do not involve relative pronouns, without additional stipulations, the West-Flemish long-distance object relativization structure can be accounted for under the present theory. How about long-distance subject relativization? The first step of the derivation is as indicated in (51). Given that there is a relative pronoun present, the question is whether for some reason this pronoun is stranded in the most deeply embedded CP, or whether the relative DP as a whole moves to the next Spec,CP. In the latter case we need an additional PF-deletion theory to account for the fact that the relative DP in the lower Spec,CP (and the DP in the higher Spec,CP) get ‘partially deleted’. I will not attempt to give an answer to the question which of these two possibilities would be the best. 89 For now, I will be as naive as possible (i.e. just move past the major differences between languages like English and Dutch), and copy the assumptions that hold for English to the Dutch dialects under discussion. 90 Notice that it is implicitly assumed that die in West-Flemish is a relative pronoun, contrary to what is assumed in my proposal. 51 Consider now the data in (52); there is no asymmetry in short relativization, but there is an asymmetry in long relativization structures (system V). (52) a. b. c. d. Dat is de man [die __ het verhaal verteld heeft] Dat is de man [die ze __ geroepen hebben] Dat is de man [die ik denk [die __ het verhaal verteld heeft]] Dat is de man [die ik denk [dat ze __ geroepen hebben]] The short subject relative is identical to the short object relative in West-Flemish and can be easily derived (cf. (51)), the short object relative, however, raises some questions. Given that object relatives never can delete C’s [uT] feature, in object relatives T-to-C movement is always predicted to occur. This means that we would expect the presence of dat in (52b). The only way to circumvent this, is by assuming that [uT] on C is not endowed with the EPP-feature. This is of course a rather unwarranted conclusion, since it does not explain anything. However, for the sake of argument, let us assume that C in object relatives bears [uT, -EPP]; the pattern in (52a,b) is then derived. Given this assumption, we predict that the most deeply embedded C in long object relatives also bears [uT, -EPP]. However, as (52d) shows, this prediction is not borne out, i.e. there is an instance of dat in the embedded clause. Another way of accounting for the lack of a s/o-asymmetry in (52a,b) is by assuming that instead of vP being a phase, TP is a phase. That way, neither the subject nor the object can delete C’s [uT] feature. If we furthermore assume that this [uT] feature does not have an EPP property, this feature can be erased by mere Agree. Although it might be argued that the difference between dialects is due to differences in what counts as a phase (cf. Gallego 2004), it is extremely unlikely that the question of what constitutes a phase can be answered differently within a particular language dependent on the construction. This is, however, exactly what we would need in order to account for the pattern in (52): no asymmetry with short relativization, but the presence of an asymmetry with long relativization. Although it still might be possible to assume that variation found in the domain of relativization can be explained by appealing to the T-to-C movement approach and the assumption that there is variation regarding the (strength of) the features on C, i.e. the feature specification of C is not the same in every dialect, we need to make some unwarranted stipulations in order to account for s/oasymmetries in different dialects. Therefore, I will not further pursue this line of thinking about s/oasymmetries in varieties of Dutch, and conclude that, at least at first sight, the T-to-C movement approach without additional assumptions is unable to account for the observed variation regarding s/oasymmetries in long-distance relative clauses.91 3.5.2 Rizzi & Shlonsky (2007) A recently formulated alternative account to the ECP is provided by Rizzi & Shlonsky (2007). Their account is based on the following two ideas, presented in (53). (53) a. b. Criterial Freezing An element moved to a position dedicated to some scope-discourse interpretive property, a criterial position, is frozen in place. The Classical EPP – clauses need subjects – is restated as the Subject Criterion. These formulations ensure that when a subject moves to the criterial subject position – the highest head within TP: Subj92 – it is frozen in place. The difficulty of moving subjects is thus explained by appealing to Criterial Freezing. Since there is no object criterion that is equivalent to the Subject Criterion, the distribution of objects is not similarly constrained. The (English) examples in which the 91 It is important to note that this conclusion is dependent on the analysis that is chosen for relative clauses in general (cf. chapter 1) and for s/o-asymmetries is particular. It could very well turn out that with another analysis of relativization and s/o-asymmetries, the T-to-C movement approach makes the right predictions. 92 Rizzi & Shlonsky (2007) assume that there exists a head Subj that attracts a nominal to its specifier. This gives rise to the following configuration: [DP [Subj XP]]. The feature that triggers this movement operation is interpreted as aboutness: “about DP, I am reporting event XP”. 52 subject cannot be extracted, follow now straightforwardly: the subject moves to Spec,Subj and by Criterial Freezing, it is frozen in place and not allowed to move any further. In order to account for the French que/qui alternation, Rizzi & Shlonsky adopt the main insights of Taraldsen’s (2001) analysis. He assumes that qui is a contracted form, composed of que and i; the latter being an expletive-like element (similar to the Standard French expletive il). The basic idea is that in subject relative clauses in French, the expletive element i – merged as the finiteness head (Fin0), just below F0 which hosts que, and just above SubjP, i.e. [FP que [FinP i [SubjP]]] – satisfies the Subject Criterion. Due to its nominal nature (i.e. i is merged with a set of φ-features), the expletive i in Fin0 is assumed to satisfy the nominal requirement of Subj 0, as it is in a head-head relation with Subj0. Therefore, the thematic subject is now free to move to the COMP-domain, i.e. the criterial position Spec,Subj can be skipped. The subject first moves to Spec,Fin – which is not criterial – to value i’s φfeatures, after which it can move further to the COMP-domain. Thus, simply put, when the element i is present – giving rise to the appearance of qui – the subject can freely extract from TP. The impossibility of qui appearing in object relative clauses is explained by the fact that there is simply not enough room for the expletive i and the subject DP to occur in the same structure. For the precise implementation of this analysis, I refer the reader to Rizzi & Shlonsky’s (2007) paper; for the present discussion, the above short outline will suffice. In the remainder of this section, I will show, based on arguments by Koopman & Sportiche (2008) and Mayr (to appear), that Rizzi & Shlonksy’s analysis of s/o-asymmetries is empirically and theoretically inadequate. According to Koopman & Sportiche (2008), Rizzi & Shlonksy’s proposal faces two serious problems. First, given that qui is a contraction of the two heads que (F0) and i (Fin0), and given the observation that the existence of these heads is justified on the basis of the possibility of material intervening between the two, we predict the existence of sentences with the string que X i. However, such sentences are never attested. Second, in line with the criticism above, we predict the following sentence (54b) to be well-formed, contrary to fact.93 (54) a. b. Quel enfant est parti? ‘Which child left?’ [Quel enfant]j [F e] tj [Fin i] [Subj tj … [Koopman & Sportiche 2008:76] To account for the ill-formedness of (54b), Rizzi & Shlonsky assume that in French main clauses, a silent version of [Fin i] is allowed. Naturally, such an assumption faces the problem of overgeneration. A third point of criticism regarding the account of Rizzi & Shlonsky as noted by Mayr (to appear) has to do with the observation that it makes the wrong predictions regarding complementizer agreement constructions. This is due to fact that Rizzi & Shlonsky assume that the position that deals with the subject criterion (Subj0) is different from the position that deals with agreement (Agr 0); Agr0 is below Subj0 in the clausal structure. Mayr’s argument is as follows. He notes that Rizzi & Shlonsky assume that in expletive constructions, the subject moves to Spec,Agr, and the expletive moves to Spec,Subj to satisfy the subject criterion. The English construction in (55), however, shows that this assumption cannot be right, i.e. the construction seems to lack subject movement because the subject follows the low adverb often (which marks the edge of vP). (55) There is often a room available. [Mayr to appear:22] On the basis of the grammaticality of (55), Mayr concludes that the agreement relation between the subject and Agr0 in expletive constructions can be a long-distance relation. This conclusion makes a prediction with respect to complementizer agreement constructions. In Rizzi & Shlonsky’s account, complementizer agreement is analyzed on a par with expletive constructions. That is to say, it is assumed that in case of complementizer agreement, the φ-features on the complementizer satisfy the nominal requirement of Subj0 and – just like in expletive constructions, e.g. (55) – subjects should stay in-situ (and not move to Spec,Agr). German shows VP-fronting, and given the above assumptions, we predict that in cases of complementizer agreement, the VP containing the subject should be able to be Note that these points of criticism regarding Rizzi & Shlonsky’s (2007) approach also hold for Taraldsen’s (2001) analysis of the que/qui-alternation. 93 53 fronted, as the subject may stay in-situ. This prediction is not borne out, as illustrated in (56). This observation leads to the conclusion that φ-features on the complementizer cannot satisfy the subject criterion. That is to say, the subject does not skip Spec,Subj in (56), and Mayr concludes that any account that assumes the φ-features of C0/Fin0 satisfy the subject criterion, is not attainable. (56) a. b. [VP a Buach kafft]i hot da Hauns gfrogt [CP ti ob-s es ti hobts] a book bought has the Hans asked if-2PL you-2PL have ‘John asked whether you(pl) bought a book.’ * [VP es a Buach kafft]i hot da Hauns gfrogt [CP ti ob-s ti hobts] From the above, it should be clear that the theory of Rizzi & Shlonsky (2007) faces some serious conceptual/ theoretical and empirical shortcomings. It should be noted, however, that their analysis does make the right predictions regarding the Dutch systems of relativization that show a s/oasymmetry. It follows from Rizzi & Shlonskyʼs proposal that special die is only attested in the most deeply embedded clause because the relativized constituent does not have to move through the Spec,Subj in the higher clause, but only in the lower clause, i.e. only in the lower clause we need an expletive-like element that satisfies the subject criterion, whereas in the higher clause the local subject satisfies the subject criterion. Moreover, given that special die is assumed to be a contracted form – composed of the complementizer and another element in Spec,Subj94 – instead of a (weak) relative pronoun (cf. Koopman & Sportiche 2008), Rizzi & Shlonsky make more or less the same predictions that my theory does. More specifically, it can account for pattern V by assuming that one of the instances of die – the relative pronoun or contracted die – is not spelled out (by some sort of haplology mechanism, cf. footnote 65). 3.5.3 Alternative accounts - summary The preceding subsections have shown several alternative theories to my analysis of the variation attested in long-distance relativization structures in the Dutch speaking language area. First, it was shown that a feature approach to the attested variation, as proposed by Pesetsky & Torrego (2001) and extended by Gallego (2004), is rather arbitrary, i.e. the presence or absence of a certain feature is often not independently motivated. Besides this theoretical/conceptual shortcoming, this theory was also incapable of accounting for the pattern of system V without making ad hoc stipulations. Second, the theory of s/o-asymmetries introduced by Rizzi & Shlonsky (2007) was discussed. Although this theory makes the right predictions regarding the different Dutch relativization structures, it has been shown that this theory is theoretically and empirically inadequate. The following table gives an overview of the predictions the several theories make with respect to the grammaticality of relativization structures involving s/o-asymmetries; “+” means that a particular approach can account for the grammaticality of a given system of relativization, and “-” means that a given approach cannot account for the grammaticality of a given system of relativization. In the table, I also included the predictions made by the recent approach of Koopman & Sportiche (2008), as briefly discussed in footnote 81. Table 5; predictions regarding the grammaticality of relativization structures involving s/o-asymmetries system of relativization die,dat - dat-die, dat-dat (IV) die,die - die-die, die-dat (V) die,dat - die-die, dat-dat (3) this thesis + + + Pesetsky & Torrego (2001) + +? 94 Rizzi & Shlonksy (2007) + + + Koopman & Sportiche (2008) + + - I will not go into the issue whether it could be plausibly argued that special die is a contracted form consisting of the complementizer together with some element in Spec,Subj. 54 3.6 Summary In this chapter, I have provided an analysis for a set of data consisting of two systems of (longdistance) relativization in Dutch dialects that show a s/o-asymmetry. The first system (IV), predominantly found in West-Vlaanderen, shows an asymmetry in both short and long relativization structures. The second system (V), mainly found in Oost-Vlaanderen, only shows a s/o-asymmetry in long-distance relativization configurations. The two attested patterns of s/o-asymmetries seem to be subject to the following generalizations. First, s/o-asymmetries are only found in the COMP-domain of the clause that contains the extraction site (generalization II). Second, when there is an asymmetry, die is always found in case of subject extraction and dat is found in case of object extraction (generalization III). Third, dialects that show the asymmetry often also show complementizer agreement (generalization IV). These observations are accounted for by appealing to a very recent theory of s/o-asymmetries in Bavarian, proposed by Mayr (to appear). He basically argues that (overt) complementizer agreement licenses subject extraction. The subject (or a copy of the subject) needs to enter into a φ-agreement relation with the complementizer in order to be on the projection line and to be able to be further extracted. Extending Mayr’s analysis to the Dutch data yields the correct predictions. Given that the φ-agreement relation between (a copy of) the subject and C0 is morphologically realized as die (instead of dat) in the dialects under discussion, the existence of two patterns of s/o-asymmetries in relative clauses is derived: a pattern that is found in dialects that make use of the relative pronoun die for both subject and object extraction (system V – in which there is only an asymmetry in the most deeply embedded COMP of long-distance relativization structures), and another pattern that is found in dialects that do not make use of relative pronouns (system IV – which shows asymmetries in both short relative clauses and in the most deeply embedded COMP of longdistance relativization configurations). In other words, the two attested patterns of relativization are accounted for by appealing to two micro-parameters: (i) +/- spell-out of φ-agreement with C0 (in case of subject relativization), and (ii) the presence/absence of relative pronouns. Although my theory faces some problems – most prominently, the problem of morphological plausibility – it also makes some right predictions. Moreover, the evaluation of some other (recent) theories of the s/o-asymmetry – Pesetsky & Torrego (2001), Gallego (2004), Rizzi & Shlonksy (2007), Koopman & Sportiche (2008) – shows that each theory has its own shortcomings, i.e. some theories make the wrong predictions and cannot account for the attested variation in Dutch, whereas others are less appealing from a conceptually/theoretically point of view. 55 CHAPTER 4 Long-distance relativization in Dutch dialects – no subject/object asymmetries 4.1 Introduction In the previous chapter, I discussed two systems of relativization that display subject/object asymmetries. It was argued that the presence of a s/o-asymmetry is related to the presence of overt complementizer agreement in these dialects. In this chapter, I will investigate the system(s) underlying the patterns of (long-distance) relativization without subject/object asymmetries, as given in (1). Although at first sight, it seems to be the case that ‘anything goes’ (chapter 2, generalization I), it turns out that there is a system underlying these four patterns. (1) system I system VI system VII system VIII die-dat-Ø dat-dat-Ø die-die-Ø dat-die-Ø die-dat-Ø dat-dat-Ø die-die-Ø dat-die-Ø Given the hypothesis that all variants to long-distance relativization have the same underlying structure (cf. chapter 1), I will argue that the difference between system I – the Standard Dutch pattern of relativization – and system VI, can be derived by assuming that the latter does not make use of relative pronouns, whereas the dialects that display system I, do make use of relative pronouns (die for both subjects and objects). For the derivation of systems VII and VIII, some additional assumptions need to be made. I will assume, following Van Koppen (2007), that movement of XPs does not leave an identical copy at the extraction site, but rather only leaves the φ-features of the moved element, i.e. copies do not have internal structure. Under the assumption that some dialects can spell out more than one chain link (micro-parameter 3), the pattern of system VII can be easily derived. In order to account for system VIII, I will assume that movement/copying can target a subset of the complete feature bundle/structure of an element (cf. Barbiers, Koemenan, Lekakou 2008). Thus, dialects differ with respect to the size of the structure that movement/copying targets (full/partial): micro-parameter 4. Before I will turn to the specifics of the analysis of systems I, VI, VII and VIII in section 4.3, the following section presents the data that are the subject of this chapter. 4.2 Presentation of the data The patterns of short and long relativization of the systems that will be the subject of this chapter are given in table 1. Regarding short relativization, each system makes use of die as a relative marker, with the exception of system VI, which makes use of dat. This difference is reflected in the long relativization structures: only system VI does not show an instance of die, whereas the other three systems either have die in the lower clause (VIII), die in the higher clause (I), or die in both the lower and the higher clause (VII). The data in table 1 thus show that each possible combination of die and dat is attested; this observation is captured by the generalization in (2) (cf. chapter 2). Table 1; four systems of relativization without subject/object asymmetry system I (47) system VI (16) system VII (9) system VIII (7) (2) I short subject die dat die die short object die dat die die long subject die-dat-Ø dat-dat-Ø die-die-Ø dat-die-Ø long object die-dat-Ø dat-dat-Ø die-die-Ø dat-die-Ø long-distance relativization without a subject/object asymmetry shows all the possible variants – die-dat-Ø – die-dat-Ø, dat-die-Ø – dat-die-Ø, die-die-Ø – die-die-Ø, dat-dat-Ø – dat-dat-Ø Whereas the distribution of the dialects that show a subject/object asymmetry was rather clear – and therefore, based on geographic distribution, a correlation with complementizer agreement was found – the distribution of dialects that do not show such an asymmetry is less clear, as indicated on map 1. 56 Map 1; 4 systems without subject/object asymmetry As map 1 shows, system I is found mainly in the Netherlands and in the Belgian provinces Antwerp and Oost-Vlaanderen. The geographic distribution of the other three systems is more or less restricted to Belgium. That is to say, system VI is predominantly found in the south of Oost-Vlaanderen and Vlaams-Brabant, and system VII is mainly found in Oost-Vlaanderen and shows some isolated occurrences in the Netherlands. The majority of occurrences of system VIII is also found in (the south of) Oost-Vlaanderen, and three isolated occurrences of this system are attested in the Netherlands. 4.3 The analysis – no subject/object asymmetries in varieties of Dutch As mentioned in chapter 1, I will pursue the default hypothesis and argue that these four variants to long-distance relativization are just structural variants in the sense that the variation should be (partially) attributed to differences in spell out. That is to say, I will assume that all structures should be analyzed by a raising account of relative clauses (e.g. Kayne 1994, Zwart 2000, De Vries 2002) in which the relative DP successive-cyclically moves to its final landing site in the matrix clause. In this section I will propose an analysis that straightforwardly captures the four patterns of relativization presented in the previous section. 4.3.1 Absence/presence of relative pronouns In chapter 3, an analysis was proposed for two systems of (long-distance) relativization that show subject/object-asymmetries. To account for the attested variation, the following two micro-parameters were distinguished. (3) a. b. +/- presence of relative pronouns +/- spell-out of φ-agreement with C0 (in case of subject relativization) Given these parameters, the analysis of systems I and VI – for convenience repeated in table 2 – is pretty straightforward. That is to say, the systems seem to differ with respect to parameter (3a), i.e. whether or not a given dialect makes use of relative pronouns. 57 Table 2; systems I and VI system I (47) system VI (16) short subject die dat short object die dat long subject die-dat-Ø dat-dat-Ø long object die-dat-Ø dat-dat-Ø System I is the Standard Dutch variant. I will assume that the dialects exhibiting this pattern make use of the relative pronoun die – this is corroborated by the use of this pronoun in short relativization. In long relativization structures this means that the relative pronoun introduces the higher clause, whereas the lower clause is introduced by a complementizer. This is the exact analysis as proposed by De Vries (2002); see also chapter 1, section 1.3.3.2. In order to account for the pattern of system VI, I assume that dat is a complementizer and dialects that exhibit system VI do not make use of relative pronouns. Recall that in order to explain the pattern of system IV in West-Vlaanderen – in which there is a subject/object asymmetry in short relatives and in the lower clause of long relatives – I analyzed special die as an instance of the complementizer and I furthermore assumed that the dialects that exhibit this system do not make use of relative pronouns. More specifically, I assumed that dialects differ with respect to the lexical specification of die as a complementizer. Given this analysis, it follows that system VI – that also does not make use of relative pronouns – is found in this language area (cf. map 1), i.e. the two systems only differ with respect to one micro-parameter: the specification of die in the lexicon, or, in other words, the spell-out of φ-agreement with C0 (3b). In sum, given that the dialects under discussion do not spell out the φ-agreement relation between C0 and (the copy of) the relative subject DP in Spec,CP, the relevant micro-parameter when accounting for the patterns of systems I and VI is the absence or presence of relative pronouns in a given dialect. This is illustrated in the following table. Table 3; two micro-parameters95 + relative pronouns - relative pronouns + spell out agreement die-die, die-dat (V) dat-die, dat-dat (IV) - spell out agreement die-dat, die-dat (I) dat-dat, dat-dat (VI) Notice that the formulation of this simple parameter – absence/presence of relative pronouns in a given dialect – is in fact insufficient to account for the patterns of the two systems in the sense that we still need some sort of rule that states that COMP can never be empty, because otherwise, we would predict the following sentence to be grammatical, quod non. In other words, we need to make the additional assumption that Dutch dialects do not allow zero-relativization (cf. footnote 33). (4) *Dat is de man __ that is the man ik denk __ I think het gedaan heeft it done has 4.3.2 Syntactic doubling Whereas an account of systems I and VI in terms of the presence or absence of relative pronouns in a given dialect, very easily captures the difference between the two patterns, something more complex seems to be going on in the patterns of systems VII and VIII, here repeated in table 4. Table 4; system VII and VIII system VII (9) system VIII (7) short subject die die short object die die long subject die-die-Ø dat-die-Ø 95 long object die-die-Ø dat-die-Ø The parameter +/- relative pronouns in this table indicates whether or not a dialect makes use of die for both subject and object relativization. If we would further specify this parameter with respect to the form of the relative pronouns that are used for (short) subject and object relativization, we predict the existence of two more systems when the relative pronoun die is used with subject relatives and the relative pronoun dat is used to mark object relatives: (i) die-die, dat-dat, and (ii) die-dat, dat-dat. Both systems are indeed attested (see table 3, chapter 3). 58 In short relativization structures, both systems make use of the relative pronoun die, similar to system I. In long relativization structures on the other hand, system VII shows doubling of the relative pronoun (die-die) and system VIII shows stranding of the relative pronoun in the most deeply embedded COMP (dat-die).96 At least at first sight, these data seem to force the formulation of a specific PF-deletion theory. That is to say, given that the lower instance of die in long-distance relativization in system VII is the spell out of (part of) a lower copy, we need a PF-deletion theory that accounts for the fact that, in addition to the highest copy, a lower copy can be (partially) spelled out. Similarly, when we assume that the instance of die in long-distance relativization in system VIII is the spell out of a lower copy instead of the highest copy, we need to account for the fact that (part of) the higher copy (and part of the lower copy) is not spelled out (i.e. scattered deletion, cf. Nunes 2004). In short, we need to account for the fact that these copies are partially spelled out, i.e. we never attested instances of copy spell out in which the lower copy is identical to the higher copy (5). (5) a. b. *… de [mani die ti] ik denk [mani die ti] (dat) [mani die ti] het verhaal verteld heeft *… de [mani die ti] ik denk [mani die ti] (dat) [mani die ti] het verhaal verteld heeft In other words, the analysis of long-distance relative clauses (cf. chapter 1, section 1.3.3.2) that I assume for (varieties of) Dutch, forces one to assume a specific PF-deletion theory because of the presence of the intermediate copy of the relative DP in the embedded Spec,CP. In the following sections, I will pursue the line of reasoning that the difference between the two systems under discussion is due to the interaction between Syntax and Phonology/Morphology. I will first discuss Nunes’ (2004) influential proposal regarding the realization of multiple chain links, and I will show that, as it turns out, this proposal (without additional mechanisms) is unable to account for the data under discussion, mainly because it makes use of the ‘standard’ copy theory of movement (Chomsky 1993). Therefore, I will adopt the reduced copy theory of movement as proposed by Van Koppen (2007) in order to elegantly account for the observation that generally lower copies are reduced variants of the higher copy. Together, the reduced copy theory of movement and the assumption that multiple chain links can be spelled out, can account for the doubling pattern of system VII. To provide an analysis for system VIII, I will follow Barbiers, Koeneman & Lekakou (2008) and argue that Syntax has the operation partial copying by which only part of a constituent gets copied and remerged higher into the structure. 4.3.2.1 Phonetic realization of multiple copies – Nunes (2004) Nunes (2004) adopts Chomsky’s (1993) notion of movement as consisting of the two operations copy and merge – the Copy Theory of Movement – and argues that heads as well as lower copies of chains are subject to the same principles and mechanisms. More precisely, he claims that every chain link is computed for linearization in accordance with Kayne’s (1994) Linear Correspondence Axiom (LCA) according to which linear order is determined by asymmetric c-command.97 Under the assumption that two copies of one and the same element count as ‘identical’, it follows that it is impossible to linearize structures containing identical copies because an element intervening between two copies should simultaneously go before and after one and the same element. Simply put, deletion of copies follows from the requirement to linearize a structure. To account for the observation that in most languages the highest copy of a chain gets pronounced, Nunes argues that in the standard case, the copy with the most formal features erased gets deleted, i.e. the copy that has entered into the most checking relations will be spelled out. However, this option can be overruled by morphophonological considerations (see Nunes 2004 for details). Interestingly, sometimes it seems possible that more than one chain link gets phonetically realized, as illustrated by the examples in (6)-(8). The obvious question arises why such structures do not cause any problems for linearization. 96 Notice that as there is no asymmetry between the long subject relative and the long object relative, we cannot assume that die is an instance of the complementizer (see the analysis of subject/object-asymmetries as proposed in chapter 3). Rather, we need to assume it is a relative pronoun. 97 Simply put, the LCA states that a terminal node α precedes a terminal node β iff α asymmetrically ccommands β, in which asymmetric c-command is defined as follows: X asymmetrically c-commands Y iff X ccommands Y and Y does not c-command X (for the details of this approach to linearization see Kayne 1994). 59 (6) (7) (8) Wêr tinke jo wêr’t where think you where-that ‘Where do you think that Jan lives?’ Jan wennet? Jan lives [Frisian] Wen glaubst Hans wen Jakob gesehen hat? who thinks Hans whom Jakob seen has ‘Who does Hans think Jakob saw?’ [German] Met wie het jy nou weer gesê met wie het Sarie gedog met wie gaan Jan trou. with who did you now again said with who did Sarie thought with who gaan Jan marry ‘Whom did you say (again) that Sarie thought Jan is going to marry?’ [Afrikaans]98 In attempting to answer that question, Nunes assumes that under specific circumstances, heads can undergo morphological reanalysis. Morphological reanalysis takes two terminal nodes and fuses them together into a single terminal node, by which the number of independent morphemes is reduced. Given that the LCA does not apply word internally and given the assumption that successive cyclic wh-movement (in languages that allow spell out of lower copies) proceeds by adjunction to an intermediate C0, we can now give an explanation for the well-formedness of (6) and (7). The wh-copy in the lower COMP is adjoined to C0 and together they undergo morphological reanalysis by which both heads become a single morpheme, and as a result, the wh-copy is invisible for the LCA. With respect to sentence (8), in which we are dealing with a maximal projection (i.e. a PP) instead of a head, Nunes (2004:169) notes that we need to assume that first the preposition and the wh-word undergo fusion, after which this element fuses with the intermediate C0. Given this analysis of the phonetic realization of multiple copies, dialectal variation can now be stated in terms of “the degree of permissiveness of a given dialect or idiolect with respect to morphological reanalysis” (Nunes 2004:43). However, it is hard to see how morphological reanalysis could account for the patterns of systems VII and VIII, because it targets only heads. Given the basic structure we assumed for Standard Dutch relative clauses, as given somewhat simplified in (9), in terms of Nunes’ account, the only thing we can claim is that somehow die (in the intermediate copy) incorporates into C0, by which it becomes invisible to the LCA and is spelled out, yielding the correct outcome (die-die). (9) … de [mani die ti] ik denk [mani die ti] dat [mani die ti] het verhaal verteld heeft This scenario is not very plausible, however. Even if we were to assume that only the pronoun moves, i.e. if we would adopt a head external analysis of relative clauses (cf. chapter 1, section 1.2.1), morphological reanalysis would still not be able to account for the attested variation if we assume that pronouns are not heads, but actually project a full phrasal structure, as is argued for extensively in the literature (cf. amongst others Cardinaletti 1994, Cardinaletti & Starke 1999, Déchaine & Wiltschko 2002, Van Koppen 2005). Besides this particular problem to Nunes’ (2004) theory, it faces some other problems, both theoretically and empirically. Theoretically, it should be impossible for a headadjoined copy to move further to the higher Spec,CP (cf. Felser 2004:556). Empirically, the analysis – taking the copy theory of movement as a starting point – is hard to reconcile with non-identical spell out of copies, i.e. the spell out of copies that are reduced with respect to the higher copy. Therefore, I will follow Van Koppen (2007) and assume an alternative copy theory of movement. The following subsection briefly outlines her approach. 4.3.2.2 Reduced copies – Van Koppen (2007) An alternative to the ‘standard’ copy theory of movement (Chomsky 1993) is provided by Van Koppen (2007). She argues that the head of a movement chain differs from the lower copies in the sense that these are reduced: they contain only the φ-feature set of the maximal projection of the 98 Examples (6)-(8) are taken from Nunes (2004:38-39). 60 moved item and thus have no internal structure.99 Copies are taken to be place holders of the moved item. More specifically, she argues that the operation copy is more restricted than previously assumed: it targets only heads, which means that when an XP moves, only the feature bundle on the maximal projection gets copied (as this is the only information necessary to link the head of the chain to its lower copies), instead of the whole XP. Thus, reduced copies are the result of XP-movement. This is illustrated in (10)-(11). (10) Standard Copy Theory (Chomsky 1993) (11) Reduced Copy Theory (Van Koppen 2007) XP XP X’ ZP [F] Z’ Z0 X’ ZP [F] YP Z’ Z0 ZP [F] YP [F] Z’ Z0 Arguments in favour of this theory of copying come from structures in which an XP moves and more than one copy is spelled-out. The reduced copy theory predicts that in such cases, the lower copy is reduced, whereas the standard copy theory predicts that the lower copy should be identical to the higher copy. The examples in (12) show that when a complex phrase moves, the spelled out copy cannot be identical to the highest copy, which confirms the prediction made by the reduced copy theory.100,101,102 (12) a. b. * Wessen Student glaubst du wessen Student wir kennen? which student think you which student we know ‘Which student do you think we know?’ Welchen Mann glaubst du wem sie das Buch gegeben hat? which man think you who she the book given has ‘Which man do you think that she has given the book to?’ [German] [dialectal German]103 Let us assume that the reduced copy theory of movement is on the right track, and copies left behind by movement have no internal structure. Rather, these reduced copies only contain the φ-feature set present on the maximal projection of the moved item. We then predict that the movement chain of the element man die in the relative clauses under discussion, looks like (13b) instead of (13a). That is to say, the lower copies of this element are not identical to the head of the chain as in (13a) but contain only the φ-feature set of this higher copy as in (13b). 99 Note that with the reduced copy theory of movement, we need an alternative theory about reconstruction, i.e. it is no longer sufficient to assume that reconstruction is the interpretation of a lower copy in the chain. 100 For other arguments in favour of the reduced copy theory of movement, see Van Koppen (2007). 101 Notice that example (8) from section 4.3.2.1 actually forms a problem to Van Koppen’s (2007) theory since the lower copy of the chain is identical to the higher copy, i.e. the lower copy is a complex phrase (i.e. a PP). 102 As noted by Van Koppen (2007:345), another analysis of (12b) is also possible, in which the wh-phrase in the embedded clause is taken to be wh-agreement instead of the spell out of an intermediate chain link (cf. footnote 111 for a discussion of this possibility). 103 Examples are taken from Van Koppen (2007:345). 61 (13) a. b. [man die] … [man die] … [man die] [man die] … [3P.SG.M] … [3P.SG.M] Consider now the possibility that multiple chain links get spelled out. More in particular, let us assume that in addition to the highest copy, the intermediate copy is spelled out. At the level of Morphology, an element that matches the features [3P.SG.M] is then inserted into the structure. In the present case, this could very well be the demonstrative pronoun die as illustrated in (14), giving rise to the pattern of system VII.104 (14) Dat is de [man die] ik denk [die] het verhaal verteld heeft. (system VII) In sum, with the assumption of a reduced copy theory of movement together with the assumption that multiple copies can be spelled out, we easily derive the presence of system VII. However, we still need a theory that accounts for the fact that some varieties allow intermediate copies to be spelled out. In addition, we need to derive that in the dialects under discussion only intermediate copies can be spelled out, but not the lowest copies.105 In the remainder of this section, I will focus on the question of what such a theory should look like. Recall from chapter 1 the important observation that (varieties of) Dutch do not allow zero relativization, which means that there always needs to be an element in the COMP-domain: a complementizer or a relative pronoun; this is illustrated in (15). (15) Dat is de man *(dat/die) ik denk *(dat/die) het gedaan heeft. This particular requirement/observation can be restated as follows. (16) In the absence of an overt complementizer, a (copy of a) pronoun must appear at the left periphery of the embedded clause. When a sentence does not obey (16), it is ungrammatical, as (15) shows. Therefore, we could simply state that when there is no (feature bundle of a) complementizer present in the numeration106 underlying the derivation, (16) forces the spell out of the intermediate copy in the embedded Spec,CP. If this assumption is on the right track, we immediately derive the observation that the lowest copy cannot be spelled out, as this particular spell out is not necessary, and by economy, it is not allowed. In sum, we basically state that an intermediate copy should be spelled out only when there is no complementizer present in the COMP-domain. However, it should be clear that this very simple assumption will not work, as there are dialects that allow doubly filled COMP. With the above line of reasoning we would predict doubly filled COMP to be ruled out by economy. Another option to account for the spell out of intermediate copies is to follow Nunes (2004) and assume that the intermediate copy has undergone morphological reanalysis. As mentioned in section 4.3.2.1, one major argument against this analysis is the assumption that pronouns are not heads and are thus unable to undergo morphological reanalysis. However, we followed Van Koppen (2007) and assumed that lower copies have no internal structure – similar to heads – and as a consequence, this argument no longer holds. Let us therefore follow Nunes (2004) and assume that dialects differ in permitting morphological reanalysis, and that, as a consequence, some dialects allow multiple chain links to be spelled out, whereas others do not. To illustrate how Nunes’ analysis exactly works, 104 Given the feature specification [3P.SG.M], the lower copy in (13b) might as well have been spelled out as the personal pronoun hij ‘he’, instead of the demonstrative pronoun die. However, sentences like dat is de man die ik denk hij het verhaal verteld heeft are not attested in the SAND corpus. It should be clear that the question of what determines how the lower copy gets spelled out needs to be further investigated (Van Koppen 2007:346). 105 Spell out of the lowest copy might provide an account of resumptive pronouns (cf. Pesetsky 1998). I will briefly touch upon the issue of resumptive pronouns in chapter 5, section 5.2.2. 106 Within the Minimalist Program, it is assumed that the computational system does not have direct access to the lexicon, rather it has access only to a collection of lexical items that together form the starting point for the derivation: the numeration (cf. Hornstein et al. 2005:69). 62 consider again sentence (14), here repeated as (17). The ‘problematic’ copy – i.e. the copy that (at least at first sight) offends the LCA107 – is given in bold. (17) Dat is de man die ik denk die het verhaal verteld heeft. Recall that according to the LCA a lexical item α precedes a lexical item β iff α asymmetrically ccommands β (cf. footnote 97). This means that a structure with two heads which are in a mutual ccommand relation, as given in (18), cannot be linearized since neither α nor β asymmetrically ccommands the other. (18) α β (19) β α #β# β We could overcome this problem by assuming that α and β get morphologically fused, as represented in (19), i.e. two heads that are sisters can be fused together to form a single terminal node. As a result, the structure no longer constitutes a problem for linearization. Given Nunes’ proposal that successivecyclic wh-movement proceeds by adjunction to the intermediate C0,108 and if we further assume that in example (17) the intermediate copy of the relative DP and the most deeply embedded C0 undergo morphological fusion, the intermediate copy die becomes invisible to the LCA, as illustrated in the structure in (20). The highest copy of the relative DP (in Spec,CP2) gets pronounced (and the lowest copy – die in TP1 – gets deleted) because this copy has the most formal features checked, and is therefore more optimal in terms of the interfaces. Notice that in order for Nunes’ analysis to work, we need to assume that reduced copies – just like ‘normal’ copies – are also subject to the LCA, and therefore, also undergo deletion. 108 As already mentioned in section 4.3.2.1, this assumption is problematic as it should not be possible for an adjoined copy to move any further (cf. Felser 2004). Although it should be clear that we need a solution to this problem, I will not go into the matter here. 107 63 (20) CP3 dat is DP FFi + de CP2 DPk C’ D’ mani C0 die TP2 ti ik denk CP1 # C0 # TP1 diek 0 C diek het verhaal verteld heeft With Van Koppen’s (2007) proposal of reduced copies and Nunes’ (2004) proposal of morphological reanalysis we derive the grammaticality of system VII, in which multiple copies are spelled out. Furthermore, we could follow Nunes in claiming that dialects differ with respect to whether they allow morphological reanalysis or not. We thus have an additional micro-parameter, as indicated in (21). (21) +/- spell out of multiple chain links (+/- morphological reanalysis) The question now arises how to deal with the pattern of system VIII. I will argue that this pattern is the result of partial copying – a syntactic operation that copies a proper subset of the structure/feature bundle of the maximal projection. In order to understand this analysis of system VIII, in the following paragraph, I will briefly discuss the possibility of partial copying, and the predictions that such an assumption makes. 4.3.2.3 Partial copying – Barbiers, Koemenan & Lekakou (2008) Based on dialectal variation on syntactic doubling as given in (22), Barbiers, Koeneman & Lekakou (2008) (henceforth BKL) put forward the generalization that in a syntactic movement chain,109 the higher chain link can never be more specified than a lower chain link.110,111 109 The two pronouns in the sentences in (22) are assumed to be part of the same movement chain. Notice that, essentially, this claim is the opposite of the claim made by Van Koppen (2007). Whereas she argues that the highest chain link is always more specified than lower chain links (since lower chain links only contain the φ-features of the moved maximal projection), Barbiers, Koeneman & Lekakou (2008) on the contrary, claim that the highest chain link is always less specified than lower chain links. The question immediately arises whether and how these two analyses can be reconciled. Unfortunately, I do not have an answer to this question. However, for present purposes, it is sufficient to state – as a description – that only in case of full/identical copying, the lower copy is reduced (but see footnote 111). 110 64 (22) a. b. c. d. Wie denk je dat ik gezien heb? who think you that I seen have ‘Who do you think that I have seen?’ Wie denk je wie ik gezien heb? who think you who I seen have Wat denk je wie ik gezien heb? what think you who I seen have *Wie denk je wat ik gezien heb? who think you what I seen have non-doubling identical doubling non-identical doubling adding structure/features This generalization follows from (i) the copy theory of movement, (ii) a phrasal analysis of pronouns112, and (iii) partial copying, a syntactic operation that copies a subconstituent and (re)merges it in a higher position. Given that wie is not a subconstituent of wat, the ungrammaticality of (22d) is 111 A problematic issue for this claim is formed by sentences like (12b) in which we are dealing with moved XPs that leave a non-identical/reduced copy. In other words, such sentences seem to show evidence for the opposite claim: higher chain links are more specified than lower chain links (Van Koppen 2007). A potential solution to this problem is to assume that in cases like (12b) the wh-phrase in the embedded CP is not a spelled out copy of the moved wh-phrase, but rather an instance of wh-agreement (cf. footnote 102). For instance, Felser (2004:565566) suggests that the possibility of wh-copying (which implies successive-cyclic wh-movement) might be restricted to non-D-linked wh-phrases. That is to say, it has often been mentioned in the literature that D-linked wh-phrases behave differently from non-D-linked wh-phrases with respect to locality, in that the former but not the latter may undergo non-local movement. Given that Felser’s suggestion is on the right track, and given the fact that the moved wh-phrase in (12b) is in fact a D-linked wh-phrase, we could assume that the wh-expression in (12b) escaped successive cyclic wh-movement, and that the overt wh-phrase in the embedded clause is an instance of wh-agreement instead of a spelled out copy. Sentences like (12b) then no longer constitute counterexamples to the generalization put forward by Barbiers, Koeneman & Lekakou (2008), as there is no chain formation between the two (bold-faced) elements in the first place. Notice, however, that if we take Felser’s suggestion (2004) seriously, this has consequences for Van Koppen’s (2007) reduced copy theory of movement, in the sense that the wh-element in the embedded clause in sentences like (12b) is no longer taken to be a reduced copy of the moved wh-phrase. Naturally, the question then arises whether we can take the lower instance of die in the relative clause in (14) to be a reduced copy of the moved relative DP (which in turn is dependent on the question whether we are dealing with a D-linked or a nonD-linked element). I will not go into this issue, but just want to have mentioned that if it is assumed that with Felser’s (2004) assumption we can give an adequate analysis of the sentences under discussion, it might turn out that it is necessary to abandon Van Koppen’s (2007) reduced copy theory of movement, in which case we need an alternative explanation of the presence of the element die in (14). 112 As mentioned briefly in chapter 1, it is not directly evident how a promotion theory of relative clauses along the lines of De Vries (2002) can be reconciled with a phrasal analysis of pronouns. That is to say, in De Vries’ (2002) analysis, the relative DP consists of the relative pronoun die (D0) which takes the NP man as its complement, see (ia), and in the course of the derivation this NP moves to Spec,DP to check φ-feature agreement with D0 (cf. section 1.3.3.1). This movement operation yields the correct word order in which the relative pronoun follows the relative head noun, see (ib). With a phrasal analysis of pronouns on the other hand, the relative DP is more complex, as the pronoun is assumed to have internal structure. Given the phrasal analysis of pronouns as provided by BKL (2008), the relative DP should look something like (ii), cf. (25). The question arises how this complex will be spelled out, because whereas the noun man is the spell out of a terminal node, the pronoun die is taken to be the spell out of the whole DP-phrase. Suppose (ii) is the correct structure of the relative DP die man. We run into problems when we extend the analysis of De Vries (2002) to the complex in (ii). That is to say, we would predict the NP man to move to the specifier position of DP before the DP as a whole moves up higher in the structure. However, when partial copying then targets a subpart of the DP, the NP man will never end up in front of the relative pronoun. A potential solution to this problem is to assume that the NP man does not move to Spec,DP, but rather, it moves out of the relative DP to the specifier position of a higher C; see also footnote 119 (but see section 1.2.2 for a point of criticism to this proposal). Needless to say that future research is necessary to see whether and how De Vriesʼ (2002) promotion theory of relative clauses can be reconciled with a phrasal analysis of pronouns. In the main text I will not further discuss this issue and I will simply assume that the structure in (ii) is the correct representation of the relative DP die man. (i) a. b. [DP die [NP man]] [DP [NP man]i die ti] (ii) 65 [DP = die [PhiP [QP [NP man]]]] explained. If we assume, in addition to the possibility of partial copying, that Phonology/Morphology can spell out more than one chain link, we derive the following options that are available to the grammar (and the possible outputs). Table 5; interaction between Syntax and Phonology/Morphology Phonology → Spell out one chain link Spell out multiple chain links non-doubling -113 identical doubling non-identical doubling Syntax ↓ Full copying Partial copying Note that the data on wh-chains in (22) are reminiscent of the data on relativization. In particular, they seem to match the data on long-distance relativization of systems I, VII and VIII, here repeated as (23). (23) a. … de man die ik denk dat het verhaal verteld heeft b. … de man die ik denk die het verhaal verteld heeft c. … de man dat ik denk die het verhaal verteld heeft non-doubling (SD, system I) identical doubling (system VII) non-identical doubling (system VIII) If we follow the line of reasoning summarized above, the important question that now needs to be answered is what are the feature specifications of dat and die? That is to say, the analysis of (23c) as a manifestation of non-identical doubling will only work if we can show that dat is the spell out of a part of die. BKL assume the following feature specifications, related to the hierarchical structures in (25)114 – for the arguments in favour of these feature specifications and structures, see BKL (2008). (24) a. b. (25) a. die = wie + definiteness dat = wat + definiteness DP D0 b. = die PhiP Phi0 DP D0 = wie QP Q0 [+Q, +φ, +def] [+Q, +def] QP Q0 = wat = dat = wat NP NP Let us assume that the morphosyntactic feature specifications of die and dat in (24) are on the right track. We then see that the features of dat are a proper subset of the features of die. However, given the syntactic structures in (25) – and given the assumption that φ-features are situated in the PhiP layer – the question arises whether the relative pronoun can move, while ‘stranding’ its PhiP layer. As such an operation is generally taken to be impossible, we are now faced with a problem, because the structure of dat is not a subpart of the structure of die and we predict (23c) to be out. Although it 113 As noted by BKL, the option of partial copying and spelling out more than one chain link is in principle possible. However, whereas partial copying and spelling out the higher chain link faces the problem of recoverability, nothing prohibits partial copying while spelling out the lower chain link. Regarding the latter, BKL (2008:79) note that “whether or not this option can be used depends on the parameter setting regulating whether or not a language has obligatorily “overt” wh-movement.” 114 This analysis of the D-pronoun die is very similar to the one proposed by Wiltschko (1998), and Déchaine & Wiltschko (2002) for German d-pronouns, e.g. der Mann is analyzed as [DP d- [AgrDP er [NP Mann]]]. However, whereas their analysis assumes that der consists of different morphemes which each occupy different heads in the structure, BKL (2008) assume that die is the spell out of a phrase (DP). 66 seems tempting to abandon the analysis of system VIII as an instance of partial doubling and instead argue that dat in (23c) is a complementizer, rather than a subpart of the relative pronoun die, I will nevertheless argue that the pattern of system VIII is the result of the syntactic operation partial doubling.115 This assumption is primarily based on the observation that partial copying (in combination with the feature specifications and structures in (24)-(25)) actually makes the right predictions regarding possible patterns of long-distance relativization in varieties of Dutch that involve the pronouns wie, wat and die, as indicated in table 6, i.e. we predict the presence/absence of the following chains of relative markers in varieties of Dutch.116 Table 6; predictions of (partial) doubling and structures in (25) Chain wie – wie wie – die *die – wie wat – wat wat – wie wat – die *wie – wat *die – wat wat – dat *dat – wat Occurrence in SAND data + (Groesbeek) + (Hooghalen) + (Rijckholt, Vaals) + (Meterik, Nieuwehagen) + (Didam) - Prediction borne out + + + + + + + + + Except for the chain wat-wie, all (and only) logical possibilities are attested. The fact that this chain is never attested is not very strange because there are only a few dialects that make use of wat or wie as a relative marker in the first place (cf. Appendix I). The results in table 6 thus seem to suggest that the assumption that movement can target a subpart of an XP is on the right track. However, we are still faced with the problem that dat is not a subpart of die. A possible solution to this problem is to assume that there is an additional layer of structure between PhiP and QP that is similar to the DP layer, and in 115 A different explanation/analysis for the pattern of system VIII might be found in sociolinguistic factors, instead of in syntactic (and phonological/morphological) factors. Recall from section 4.2 that the majority of instances of system VIII is found in the south(west) of Oost-Vlaanderen. Close to this area several other systems of (long-distance) relativization are attested. That is to say, this area is adjacent to both the area in which the φagreement relation between C0 and the (copy of the) relative subject DP in Spec,CP is spelled out – resulting in a subject/object asymmetry (cf. chapter 3) – and to the area in which this subject/object asymmetry is not attested. We could therefore argue that the area in Oost-Vlaanderen in which (amongst others) system VIII occurs, is a transition area, in the sense that the system it displays shows features/properties of surrounding systems (OostVlaanderen is in fact a well-known transition area). In that sense the pattern dat-die, dat-die might be argued to be the result of the influences from dialects that have no relative pronouns and do not spell out the φ-agreement relation (dat-dat, dat-dat) and dialects that do spell out the φ-agreement relation (die/dat-die, die/dat-dat). In sum, it should be clear that the existence of system VIII might also be explained by making appeal to extralinguistic factors. A possible way to distinguish between this analysis and the analysis presented in the main text is to investigate whether there is a correlation between the pattern of system VIII and partial doubling in long wh-questions, i.e. we should look at the geographic distribution of the chain wat-die in long-distance whformation. When the distribution of this structure is on a par with the distribution of system VIII, we might have an indication that we are really dealing with partial doubling. Unfortunately, in the SAND-project sentences like Wat denk je die ik in de stad heb gezien ‘Who think you die I in the city seen have’ were not tested (however, in four locations in (the north-eastern part of) the Netherlands this sentence was spontaneously given as a translation of another sentence). Further research seems necessary to distinguish between the two alternative analyses. 116 The observant reader might have seen that the doubling patterns in table 6 (i.e. wie-wie, and wat-wat) cannot be accounted for by the analysis that was proposed for doubling pattern die-die (system VII) in section 4.3.2.2. That is to say, given the assumption that lower copies of moved XPs only contain the φ-features of these XPs, we predict that in all cases in which multiple chain links are spelled out, the lower copy has the form die. However, this problem is easy to overcome by relaxing the requirement that lower copies only contain the φfeature set of the moved XP. Rather, we could assume that lower copies contain for example the entire feature bundle of the moved XP. 67 which dat is spelled out. This is illustrated in (26), in which the additional layer is labelled XP. Note that the structure in (26) can also account for the presence/absence of certain chains of pronouns as given in table 6. However, structure (26) makes an additional prediction, that is to say, it predicts the grammaticality of the chain dat-wie. In the SAND corpus, this chain is never attested. Moreover, with the structure in (26) we no longer derive the grammaticality of the chain wat-wie (cf. SAND data on long wh-questions) because there is always an intervening head. Both problems will be solved if we can relate the presence of XP to the presence of DP, i.e. XP is licensed iff DP is present. (26) DP D0 = die PhiP Phi0 = wie XP X0 = dat QP Q0 = wat NP It should be clear that assuming that pattern VIII should be analyzed as involving partial copying raises many issues. Many of these issues remain unsolved and have to await further research.117 However, let us for the sake of argument suppose that structure (26) is on the right track. Then we can account for the existence of pattern VIII by the simple assumption that partial copying targets a subconstituent of die, namely XP, and remerges it higher in the structure.118 The relevant part of the derivation of (23c) is (somewhat simplified) given in (27).119 117 The idea that the internal structure of pronouns is more elaborated than given in the structures in (25) is not uncommon. As noted by Cardinaletti & Starke (1999:183), there is probably a large number of functional heads associated with DP (based on the study of adjective placement and pronominal modifiers). Moreover, given the parallelism between the DP and the clausal structure (cf. footnote 66), it is not strange to assume an elaborated internal structure of DPs/pronouns. Note, however, that we still need to answer the question about the nature of the XP-layer in (26). I will not go into this matter here, but leave this issue open for future research. 118 Notice that the partial movement account of BKL (2008) is similar in nature to the so-called peeling approach by which an element moves up in the tree, and in each movement step it may strand the highest layers, i.e. projections can be missing higher in the structure, by peeling off from the top (cf. Cardinaletti & Starke 1999). 119 As already mentioned in footnote 112, in order for the analysis to work, we need to reconcile De Vries’ (2002) promotion theory of relative clauses with a phrasal analysis of pronouns. It becomes even more apparent from the structure in (27) that we need to make additional assumptions to do so. That is to say, as the analysis of system VIII involves partial copying, we cannot simply follow De Vries (2002) and assume that the NP moves to Spec,DP because in that case, the NP will never end up in front of the relative pronoun, i.e. if the NP man moved to Spec,DP, the higher XP in (27) cannot be spelled out as man dat. This problem can be solved by assuming that the NP ends up in front of the pronoun by means of movement to a higher C-head, i.e. the NP does not move DP internally (to Spec,DP). 68 (27) CP2 XPk CP1 X 0 QP man dat DP Q NP D0 die PhiP Phi0 XPk X0 QP Q0 NP 4.3.2.4 Intermediate summary In the foregoing, I have argued that the pattern of system VII is the result of the spell out of more than one chain link. In order to account for the fact that the intermediate copy in Spec,CP is not an identical spell-out of the highest copy, i.e. the NP man is never spelled out in the intermediate copy, I have followed Van Koppen (2007) and argued that lower copies are reduced in the sense that they have no internal structure, but only contain the features of the moved XP. Some dialects allow the intermediate copy to undergo morphological reanalysis (Nunes 2004) and as a consequence, this copy can be spelled out. The option of morphological reanalysis, or more general, the option of the spell out of multiple chain links, is micro-parameter 3. The account of system VIII I have proposed in section 4.3.2.3 raised many questions that need to be the subject of further research. In this section, I basically followed Barbiers, Koeneman & Lekakou (2008) and assumed that Syntax has, in addition to the operation copy, an operation called partial copying that targets a subpart of the structure/features of an XP; the option of partial movement/copying in a given dialect is taken to be micro-parameter 4. Although with this simple assumption I was able to derive several systems of relativization (cf. table 6) that are otherwise difficult to account for – indicating that this assumption is on the right track – it is not clear whether we can plausibly argue that dat is a subpart of die. Needless to say that further research into the matter should reveal whether this is the case. In the following section, I will discuss some of the predictions that the analysis of systems VII and VIII makes. 4.3.2.5 Predictions In the previous sections, I argued that languages/dialects differ with respect to the possibility of spelling out more than one chain link and with respect to the presence/absence of the syntactic operation partial copying that targets a subpart of the moved element. These two assumptions were able to capture the existing doubling patterns in long-distance relativization structures. Given that these structures are analyzed as involving successive cyclic wh-movement, we predict that dialects that have the option of spelling out more than one chain link in relativization structures, also have this option with other long wh-movement structures like the ones in (28). 69 (28) a. b. Wie denk je wie ik in de stad heb gezien? who think you who I in the city seen have ‘Who do you think I have seen in the city?’ Wat denk je wie ik in de stad heb gezien? what think you who I in the city seen have Given the SAND data and the data of Schippers (2006), it needs to be concluded that this prediction is not borne out. Dialects that show the (partial) doubling configuration in relative clauses, often do not exhibit the structures in (28). This seems to force the unwarranted conclusion that allowing (partial) doubling is dependent on the construction. Moreover, the (partial) doubling structure with whquestions is attested very frequently – 113 attestations of (28a) and 151 attestations of (28b) (Barbiers et al. 2005) – whereas this is not the case for the relativization structures. In short, there are a lot of dialects that exhibit the structure(s) in (28), but do not have similar constructions for long-distance relativization. In chapter 3, section 3.4.3, I already concluded that long wh-movement and longdistance relativization should probably not be analyzed on a par. This conclusion is thus corroborated by the facts presented here. Another prediction we make when we assume that systems VII and VIII are the result of (partial) doubling, reduced copies and the spell out of multiple chain links, is that since both die and dat in these systems are not complementizers (as opposed to systems IV and V, cf. chapter 3), an additional complementizer in C0 is not excluded (as it is in systems IV and V). We thus predict that dialects that allow spell out of multiple copies may make use of an additional overt complementizer in the COMP-domain, as illustrated in (29). (29) a. b. Dat is de man die (dat) ik denk die (dat) het verhaal verteld heeft. Dat is de man dat (dat) ik denk die (dat) het verhaal verteld heeft. (system VII) (system VIII) It would be particular telling when we find a dialect that, in addition to the d-pronoun dat makes use of the complementizer dat. This prediction is hard to test, as the use of a complementizer in addition to a relative pronoun is predominantly found in Friesland and the north-eastern and western part of Antwerp (cf. Barbiers 2005 et al.), whereas system VII (and VIII) is not found in these areas. Moreover, natural languages seem to disallow the occurrence of two adjacent identical morphemes (cf. Neeleman & Van de Koot 2006). It is not surprising then, that we did not find the occurrence of (29b) with dat dat in the highest COMP-domain. Nevertheless, we found two dialects that show the presence of a complementizer in addition to the pronoun die, as illustrated in (30). This result strengthens the conclusion that we are dealing with the relative pronoun die instead of complementizer die. (30) a. b. c. Da is de man die an ik denkn die at dat verhaal verteld eit. Dat is de man die ank denke die asse geroepen ei. Da s de man die ak denk die asse geroepen emme. [Hoek Dutch] [Hoek Dutch] [Lier Dutch] When we take the analyses of chapter 3 and 4 together, in total, we now have the following four micro-parameters. (31) a. b. c. d. +/- presence of relative pronouns120 +/- spell-out of φ-agreement with C0 (in case of subject relativization) +/- spell-out of multiple chain links (+/- morphological reanalysis) +/- full/identical copying/movement 120 For the sake of simplicity, I take this parameter to state that a dialect/language either makes use of the relative pronoun die for subject and objects or it does not. That is to say, although there are also dialects that make use of the relative pronoun die for subject relativization and the pronoun dat for object relativization (and there are even more possibilities, like the relative pronoun wie for both subjects and objects), for the present discussion on the interaction between the different micro-parameters, this is not relevant. 70 With four parameters, we predict 24 (=16) possible patterns. However, given the assumption that when there is partial copying/movement part of the intermediate copy will be spelled out, the value of the parameter +/- spell out multiple chain links is irrelevant. This reduces the number of possible patterns to 12, as indicated in the following table. Table 7; predicted patterns of the interaction from the four micro-parameters in (31) relative pronouns + (die,die) + (die,die) + (die,die) + (die,die) + (die,die) + (die,die) - spell out agreement + + + + + + - spell out multiple copies + xx + xx + xx + xx - full/identical copying + + + + + + + + die-die(-die), die-die(-dat) dat-die(-die), dat-die(-dat) die-die, die-dat die-die(-dat), die-die(-dat) dat-die(-dat), dat-die(-dat) die-dat, die-dat dat-die, dat-dat dat-die, dat-dat dat-die, dat-dat dat-dat, dat-dat dat-dat, dat-dat dat-dat, dat-dat √ √ V VII VIII I √ √ IV √ √ VI When abstracting away from the potential presence of a complementizer, the most right row of table 7 shows that all the predicted patterns of relativization are attested. However, it seems impossible to see what the exact parameter setting of a given dialect is, i.e. it is impossible to distinguish the different analyses in the sense that sometimes the same output follows from different parameter settings.121 A way to circumvent this problem is to assume a particular hierarchy of parameters. It was already noted that the +/- spell out multiple chain links parameter is a subparameter of the +/- full/identical copying/movement parameter, and it might be possible to extent such an analysis of dependency relations between parameters. I think it is plausible to assume the hierarchy of parameters as given in (32), which basically states that in dialects/languages the syntactic operation copy can either target the full structure of an XP or a subpart of it. In the latter case, we derive system VIII (and the other patterns involving partial copying/movement as presented in table 6). In case dialects have full/identical copying/movement, there needs to be made a distinction between dialects that allow the spell out of more than one chain link (system VII and the two patterns involving full/identical doubling in table 6), and dialects that do not allow the spell out of multiple chain links. In the latter case the difference between dialects/languages boils down to (i) whether dialects/languages make use of relative pronouns, and (ii) whether dialects/languages have overt spell out of the φ-agreement relation between the most deeply embedded C0 and the (copy of the) relative subject DP in its specifier. It should be noted that the +/- relative pronoun parameter and the +/- spell out agreement parameter are interchangeable, i.e. there is no hierarchy between those two parameters. 121 As table 7 shows, some patterns can only be distinguished on the basis of the form of the complementizer (die or dat), i.e. some patterns would only differ from each other when the complementizer would be spelled out. Further research into the possible patterns of doubly filled COMP might thus shed more light on the question whether the hierarchy in (32) is correct, or whether we need less dependency relations to account for the differences between the patterns of doubly filled COMP. 71 (32) full copying - + system VIII & patterns table 6 (partial doubling) spell out more than one chain link + - system VII & patterns table 6 (full doubling) relative pronouns (die,die) + - spell out agreement + - system V system I spell out agreement + system IV system VI 4.4 Summary In this chapter, I have discussed the 4 patterns of relativization in table 8. I have shown that patterns I and VI can be simply accounted for by appealing to the two micro-parameters that I formulated in chapter 3 to account for dialects that show subject/object asymmetries in relativization: (i) +/- presence of relative pronouns, and (ii) +/- spell-out of φ-agreement with C0 (in case of subject relativization). Whereas the systems of relativization discussed in chapter 3 both spell out the φ-agreement with C0, systems I and VI do not spell out the φ-agreement relation with C0; they differ with respect to the first parameter: system I makes use of relative pronouns (die for both subjects and object), whereas dialects that display system VI do not make use of relative pronouns. In order to account for pattern VII, I assumed that some dialects allow multiple chain links to be spelled out, whereas others do not. To capture the observation that the intermediate copy in Spec,CP never gets fully spelled out, I followed Van Koppen (2007) and assumed that lower copies are reduced in the sense that they only contain the features of the moved XP, but do not have internal structure. With this assumption and the micro-parameter +/- spell out of more than one chain link, I was able to account for pattern VII. Based on SAND data – different from those presented in table 8 – that indicated that partial copying/movement seems to be an option in some dialects (micro-parameter 4), I assumed that system VIII is also a manifestation of partial copying. In order for system VIII to be the result of partial copying, we need to assume that dat is a subpart of die. Whether this is indeed the case still needs to be investigated. Finally, I took all four micro-parameters and showed that we could argue that there are dependencies between them, i.e. there is a parameter hierarchy. Table 8; four systems of relativization without subject/object asymmetry system I (47) system VI (16) system VII (9) system VIII (7) short subject die dat die die short object die dat die die long subject die-dat-Ø dat-dat-Ø die-die-Ø dat-die-Ø 72 long object die-dat-Ø dat-dat-Ø die-die-Ø dat-die-Ø CHAPTER 5 Conclusion and topics for future research 5.1 Introduction This final chapter summarizes the most important findings of this thesis. In addition, it briefly discusses some remaining topics and suggests a few avenues for future research. The organization of the chapter is as follows. First, in section 5.2, I will summarize the major results of this thesis. In section 5.3, I will touch upon a construction that has been mentioned in chapter 2, but has not yet been discussed in this thesis: the PP-relative. Finally, section 5.4 deals with resumptive pronouns in constructions other than PP-relatives and suggests some questions that will guide further investigation into the issue of resumption. 5.2 Conclusion The main goal of this thesis was to provide a description and a unified analysis of the attested variation in long-distance relativization in varieties of Dutch. I primarily focussed on the following six systems of relativization – that each involve long A’-movement without a resumptive pronoun at the extraction site – and I showed that these systems can be accounted for by the four (micro-) parameters in (1). Table 1; six systems of long-distance relativization - long A’-movement without resumptive pronoun short subject die die die dat die die I IV V VI VII VIII (1) a. b. c. d. short object die dat die dat die die long subject die-dat-Ø dat-die-Ø die-die-Ø dat-dat-Ø die-die-Ø dat-die-Ø long object die-dat-Ø dat-dat-Ø die-dat-Ø dat-dat-Ø die-die-Ø dat-die-Ø +/- presence of relative pronouns +/- spell-out of φ-agreement with C0 (in case of subject relativization) +/- spell-out of multiple chain links (+/- morphological reanalysis) +/- full/identical copying/movement Whereas the latter parameter (1d) is clearly a syntactic parameter – indicating that partial or full copying is true syntactic microvariation – the locus of (micro-) variation induced by the first three parameters (1a)-(1c) is the lexicon. More specifically, the locus of variation – caused by these three parameters – is the level of Morphology (i.e. insertion of Vocabulary Items). This is in line with Chomsky (1995) who argues that the level of microvariation is the lexicon. In this thesis, I have shown that the interaction between these four parameters can account for (at least) the six patterns of relativization as given in table 1. In the remainder of this section, I will briefly summarize the most important assumptions and findings of this thesis. I assumed that Standard Dutch long-distance relative clauses should be analyzed with a head raising analysis of relative clauses as proposed by De Vries (2002), by which it is assumed that the relative DP – consisting of the relative head noun and the relative pronoun (if present) – undergoes successive-cyclic wh-movement to the specifier of the highest CP, thus leaving an intermediate copy. To account for subject/object asymmetries (systems IV and V), I followed Mayr (to appear) and assumed that subjects always need to enter into a φ-agreement relation with the local C0 in order to be extracted. Dialects differ with respect to whether they spell out this agreement-relation, giving rise to subject/object asymmetries, or whether they do not spell out this agreement relation, resulting in identical subject and object relative clauses. Moreover, to account for the difference between system IV and V, I assumed that the latter makes use of relative pronouns whereas the former does not. With this particular micro-parameter (1a), the grammaticality of systems I and VI followed straightforwardly. Importantly, this analysis of the subject/object asymmetries and the two microparameters make the right predictions regarding the existence and non-existence of particular patterns of long-distance relativization, i.e. the predictions are borne out by the SAND data on relativization. In order to account for systems VII and VIII, I needed to makes some additional assumptions. First, I assumed that system VII is the result of the spell out of more than one chain link. Moreover, to 73 account for the fact that the spell out of the intermediate copy is never identical to the highest copy, i.e. the relative NP that is overt in the highest copy is never spelled out in the intermediate copy, I followed Van Koppen (2007) and assumed that lower copies are reduced in the sense that they only contain the feature bundle of the moved XP. With these assumptions, I was able to account for the doubling pattern of system VII. Finally, I assumed system VIII to be the result of the syntactic operation partial copying (Barbiers, Koeneman & Lekakou 2008) that targets only a subconstituent of the relative DP and (re)merges it in a higher position. With this assumption, I was able to give an analysis of system VIII, and, in addition, several other (rather marginal) systems of relativization as found in the SAND corpus on relativization (cf. table 6, chapter 4), were immediately accounted for. However, the exact internal structure of the pronominal expressions discussed in this thesis, needs to be the subject of further research. In other words, not until a well argumented structure of (relative) pronouns is proposed, we are able to see whether the analysis of partial movement can adequately account for the empirical data. This thesis has contributed to a better understanding of the formation of relative clauses in general and the microvariation regarding long-distance relativization in particular. I showed that (a hierarchy of) four micro-parameters can account for the attested variation in varieties of Dutch. Hopefully, this analysis can be extended to microvariation in other empirical domains. 5.3 Topics for future research Recall from chapter 2 that the SAND data show the existence of (at least) ten patterns of relativization in Dutch. The previous section showed that the analysis of six of these patterns – long A’-movement without resumptive pronouns – was the major topic of this thesis. Based on the four remaining systems of relativization as given in table 2, in this final section, I will discuss two topics – PP-relatives and resumption – and discuss some interesting points for future research. Section 5.3.1 deals with PPrelatives, a construction that differs from the long A’-movement construction without resumptive pronoun in that in PP-relatives the relative clause is introduced by a prepositional phrase and a coreferring pronoun appears at the extraction site. Section 5.3.2 briefly touches upon the syntax (and semantics) of resumptive pronouns in constructions different from PP-relatives. Both sections will mention some important issues that need to be the subject of further research. Table 2; four remaining systems of relativization in varieties of Dutch II III IX X short subject die die die die short object die die die die long subject waarvan-dat-hij/die waarvan-dat-hij/die die-dat-hij/die waarvan-dat- Ø long object waarvan-dat-die/’m die-dat-Ø die-dat-Ø waarvan-dat-die/’m 5.3.1 PP-relatives As presented in chapter 2, many informants spontaneously translated the test sentences with the Standard Dutch long-distance relativization pattern (die-dat-Ø) as constructions in which the relative clause is introduced by a prepositional phrase and the subject/object in the lower clause is retained, i.e. a personal or demonstrative pronoun appears at the extraction site. This is illustrated in (2a). (2b) shows that preposition stranding is also possible with this construction. (2) a. b. Dat is de man that is the man Dat is de man that is the man waarvan whereof waar ik van where I of ik denk dat hij/die het verhaal I think that he/die the story denk dat hij/die het verhaal verteld think that he/die the story told verteld heeft. told has heeft. has In addition to (2a), in which the preposition follows its object (waarvan ‘whereof’), it is also possible for the preposition to appear in front of its object, giving rise to the structure van wie ‘of whom’ in (3). (3) Dat is de man van wie that is the man of whom ik denk I think dat hij/die het verhaal verteld heeft. that he/die the story told has 74 The question that needs to be answered is of course what is the status and structure of PP-relatives and how do these configurations relate to other types of long-distance relativization? Salzmann (2005, 2006) addresses this issue, and in this section, I will briefly present and discuss his 2006 analysis.122 As Salzmann (2005,2006) notes, there are two possible analyses of the PP-relative construction: (i) a base-generation + binding approach in which it is assumed that the PP is basegenerated as a complement of the matrix verb, and the coreferential pronoun in the embedded clause is bound, and (ii) a movement approach in which it is assumed that there is long A’-movement (i.e. it is assumed that the proleptic object – the relativized constituent/dislocated phrase – originated in the position of the coreferring pronoun in the embedded clause), the preposition is taken to be inserted as a Case-marker and the coreferential pronoun in the embedded clause is a resumptive pronoun. It is difficult to argue for one of the two approaches as the proleptic construction has paradoxical properties, i.e. some properties (like reconstruction effects) point towards a movement analysis, whereas other properties (like the fact that the proleptic object has matrix clause properties) point towards a base-generation + binding approach. I will only present some of the main arguments in favour of each approach; for a comprehensive overview of the (reconstruction) properties of the proleptic construction, I refer the reader to Salzmann (2006; chapter 3). The main argument in favour of the movement analysis of the proleptic construction comes from reconstruction effects.123 (4a) shows reconstruction for variable binding, and (4b) shows reconstruction for Principle A. In both sentences the antecedent needs to undergo reconstruction into the position of the italicized pronoun in the embedded clause, suggesting that we are dealing with movement.124 (4)125 a. b. … de [periode van zijni leven] waarvan ik denk dat niemandi er graag the period of his life whereof I think that no.one there likes.to aan terug denkt is de puberteit to back thinks is the puberty ‘… the period of hisi life I think no.onei likes to remember is puberty’ ... het [gerucht over zichzelfi] waarvan ik denk dat Pieti het niet wil horen the rumor about SE-SELF whereof I think that Piet it not wants to hear ‘… the rumor about himselfi that I think that Pieti does not want to hear’ Evidence in favour of a base-generation + binding approach, or in other words, evidence that the proleptic object is base-generated in the matrix clause, comes from the observation that there is a construction in which the proleptic object is in a non-derived position, as indicated in (5). Salzmann refers to this variant of the proleptic construction as the in-situ variant. (5) Ik hoop van I hope of [dit boek] dat het this book that it een succes wordt. a success becomes [Salzmann 2006:193] Based on the Dutch examples in (6), Salzmann argues that the ex-situ variant of the proleptic construction is based on the in-situ variant, as the complex PP waarvan ‘whereof’ can be separated, stranding the preposition van ‘of’ in the verbal domain. This means that at a given point, in the ex-situ proleptic construction, the proleptic object must have been in a middle-field position. 122 Salzmann (2006) refers to this particular construction with the terms resumptive prolepsis or proleptic construction. In the remainder of this section I will use both terms PP-relative and proleptic construction to refer to constructions like (2a). 123 The validity of this argument is of course dependent on the analysis of reconstruction, i.e. whether reconstruction correlates with movement. Salzmann (2006) models reconstruction by means of the Copy Theory of Movement (Chomsky 1993), i.e. reconstruction is taken to be the interpretation of a lower copy of the chain. 124 Arguments against the movement approach of PP-relatives are the following (cf. Salzmann 2005:110): (i) there is a category mismatch between the head of the chain (PP) and the tail of the chain (DP), (ii) the status of the preposition is unclear, (iii) as the preposition projects a PP, movement would take place into a non ccommanding position, and (iv) the obligatoriness of the coreferring pronoun is unexpected. 125 Examples are taken from Salzmann (2006:163), and Salzmann (2006:166) respectively. 75 (6) a. b. … het [boek] waarvani ik __i denk dat Piet het leuk vindt the book whereof I think that Piet it cool finds ‘… the book that I think Peter likes’ … het [boek] waari ik __i van denk dat Piet het leuk vindt [Salzmann 2006:193] Moreover, there are some strong semantic arguments in favour of a base-generation approach (cf. Salzmann 2005), e.g. if long A’-movement were involved, we would predict that the proleptic object is sensitive to lexical restrictions, quod non. Based on these (and other) paradoxical properties, Salzmann (2006) proposes the following analysis of the proleptic construction.126 He assumes that the proleptic object is licensed by means of operator movement in the embedded clause, in the sense that the embedded CP merges with the matrix verb yielding a complex predicate, which in turn is saturated by the (DP in the) proleptic object PP. The operator is linked to the proleptic object by means of ellipsis. That is to say, Salzmann extends the matching analysis of relative clauses (cf. section 1.2.3) to the proleptic construction, and the operator in the embedded clause is in fact a full copy of the proleptic object. This is illustrated in (7). ellipsis (7) operator movement [CP P [DPi] V [CP [DPi]j subject [DPi]j V ]] predicate predication (7) shows the structure of sentence (5) – the in-situ variant. To derive the ex-situ variant, the proleptic object moves further, and is deleted under identity with the external head. The derivation of sentence (8) is given in (9).127 (8) … de man van wie the man of whom ik denk I think dat hij/die intelligent is. that he/die intelligent is 126 For the details of this analysis, see Salzmann (2006:232-276). The proleptic object is assumed to be generated higher than the verb, either in a higher projection of V or adjoined to VP, as indicated in structure (9) (Salzmann 2006:194). 127 76 DP (9) de man linked through ellipsis CP2 C’ PPk van wie man C0 TP ik T’ T0 VP PPk VP van wie man V0 denk linked through ellipsis CP C’ [OP man] C0 dat TP [OP man] intelligent is coreferring pronoun hij/die This analysis of the proleptic construction explains all its syntactic and semantic properties, e.g. the assumption that the proleptic object is merged within the matrix clause explains its matrix clause properties. Moreover, reconstruction effects like reconstruction into the complement clause as in (4) can be straightforwardly accounted for given the Preference Principle: only the operator in Spec,CP is retained whereas the restriction is interpreted in the base position. The question arises how Salzmann’s (2006) analysis of the proleptic construction relates to the analysis of long-distance relativization without resumptive pronouns that is the subject of this thesis. It should be clear that the basic assumptions of both analyses are different, i.e. whereas I assume a promotion theory of relative clauses (in line with De Vries 2002), Salzmann (2006) argues for a matching analysis of relative clauses. Suppose that Salzmann’s analysis of PP-relatives is on the right track, we still might argue that whereas PP-relatives receive a matching analysis, other relatives should be analyzed with a promotion theory. Naturally, this issue should be subject to future research. Moreover, the properties of PP-relatives in varieties of Dutch should be further investigated, as they probably can shed more light on the question about the analysis of the proleptic construction. For example, as already noted in chapter 2, the SAND data show that in the north-eastern part of the 77 Netherlands, we find dialects that allow PP-relatives without resumptive pronoun, as illustrated for the dialect of Schiermonnikoog in (10). (10) a. b. Dat is de man weervan that is the man whereof Dat is de man weervan that is the man of whereof ik tink dat __ het verhaal vertoald het. I think that __ the story told has ik tink dat ja __ ruppen hewwe. I think that they __ called have According to Salzmann (2006:159) constructions like these are unexpected, and he argues that these data should be further investigated. To illustrate why this is the case, it is necessary to lay out the basic assumptions of the theory of resumption Salzmann assumes. In the literature, it is often argued that resumption is related to base-generation (cf. Shlonsky 1992). However, Salzmann follows Pesetsky (1998) and Boeckx (2003) amongst others, and argues that resumption is not incompatible with a movement approach. On the basis of sentences like those in (11), which involve reconstruction into wh-islands, Salzmann argues that movement is always involved and that the presence of a resumptive pronoun makes movement from islands possible, i.e. resumption voids locality constraints.128 (11)129 a. b. … de [foto van zichzelfi] waarvan ik weet waarom Pieti er zo trots op is the picture of SE-SELF whereof I know why Piet there so proud on is ‘… the picture of himselfi that I know why Peteri is so proud of’ … de [periode van z’ni leven] waarvan ik weet waarom niemandi er the period of his life whereof I know why no.one there graag aan terugdenkt likes.to at remember ‘… the period of hisi life that I know why no onei likes to remember’ The distribution of resumptive pronouns in Dutch and German is accounted for by assuming a constraint that states that only one chain link can be overt. Given that wh-movement and relativization involve overt operators, resumption is correctly predicted not to be possible according to this constraint.130 However, in the proleptic construction there is no overt operator, and resumption is predicted to be possible. With respect to the motivation of resumption,131 Salzmann simply assumes that there is a resumptive strategy because nothing rules it out. In other words, he assumes that for German and Dutch, resumption is an option, but its distribution is very limited for independent reasons. Returning to the issue of Dutch varieties that allow PP-relatives without resumptive pronouns, it would be particularly interesting to look at these varieties and investigate whether they allow movement from islands without a resumptive pronoun, as this can shed more light on the question whether we need to distinguish resumptive pronouns that occur inside islands and resumptive pronouns that occur in transparent structures.132 128 Interestingly, wh-extractions corresponding to the structures in (11) are ungrammatical, irrespective of the presence of a resumptive pronoun. Instead of taking this as evidence that resumption does not void locality, Salzmann argues that those sentences are ungrammatical due to independent factors, i.e. resumption is not available in such contexts. 129 Example is taken from Salzmann (2006:279). 130 But recall from chapter 2, section 2.3.2, that some dialects allow resumptive pronouns with the long A’movement construction of relativization (cf. section 5.3.2). 131 Although sentences like those in (11) seem to force the conclusion that the motivation for resumption is to repair locality violations, Salzmann (2006:284-285) takes that conclusion to be incorrect, as the presence of resumptives in the proleptic construction is not dependent on whether there is an island, i.e. resumptive pronouns are obligatory in the proleptic construction, whether there is an island or not. Moreover, given that resumptive pronouns inside islands and resumptive pronouns outside islands behave the same, Salzmann assumes that they should receive the same analysis. Thus, the motivation for resumption is not to void locality. In other words, although it can be shown that resumption voids locality, we still need to explain why there is resumption. 132 Aoun et al. (2001) make such a proposal. They make a distinction between true resumptives, which are resumptives that repair locality violations (i.e. resumptives that occur inside islands), and apparent resumptives. Sentences involving the latter type of resumptive pronoun are analyzed as involving movement, whereas 78 Finally, it is worth noting that it would be interesting to investigate the difference in the amount of occurrences of subject PP-relatives on the one hand and object PP-relatives on the other, i.e. PPrelatives occur more often with subject relativization than with object relativization (cf. maps 1 and 2, chapter 2), and the question arises how we should interpret this observation. As far as I can see, there are two logical possibilities: either it is the case that somehow subject PP-relatives are ‘easier’ than object PP-relatives, or the observation should be explained from the perspective of the relation between PP-relatives and the other types of relativization. That is to say, one could imagine that PPrelatives are used more often as an alternative strategy of relativization because other types of subject relativization are ‘more difficult’ than object relativization. In short, an explanation for the difference might be found in the construction itself or in the relation of the construction with other constructions. 5.3.2 Resumptive pronouns Abstracting away from resumptive pronouns in PP-relatives – which were briefly discussed in the previous section – this section discusses the presence of resumptive pronouns in restrictive relative clauses without PPs, as illustrated in (12). (12a) and (12b) illustrate subject relativization with a resumptive pronoun, and (12c) illustrates object relativization with a resumptive pronoun. (12) a. b. c. Dat is de man die ik denke dat die het verhaal that is the man die I think that die the story Dat is de man dei ik denk dat hij het verhaal that is the man die I think that he the story Da s de man diek denk dasem that is the man die-I think that-they-him het verteld. [Oldemarkt Dutch] has told verteld het. [Roswinkel Dutch] told has geroepen emme. [Lier Dutch] called have The question arises how to account for these patterns of relativization, i.e. what is the status of these relative clauses involving resumptive pronouns? In this section, I will briefly focus on some analyses of resumption and provide some potential research questions for future research. As already mentioned in the previous section, within the literature on resumptive pronouns, there is little consensus on the syntax and semantics of these elements. That is to say, there is some debate as to whether resumption involves base-generation or movement (cf. section 5.3.1). Based on data like the following,133 which show that resumption is insensitive to locality constraints (i.e. resumption is not subject to constraints on extraction)134, it has often been argued that resumption involves base-generation (cf. Shlonsky 1992). (13)135 a. b. c. d. … de man waarvan ik denk dat Marie < elk boek leest dat hij schrijft > the man whereof I think that Marie every book reads that he writes … de man waarvan ik denk dat niemand weet < hoe hij heet > the man whereof I think that no.one knows how he is.called … de man waarvan ik denk dat < zijn moeder > gezond is the man whereof I think that his mother healthy is … de man waarvan ik denk dat niemand < met hem > wil praten the man whereof I think that no.one with him wants talk sentences with resumptive pronouns that seem to void locality (true resumptives) are assumed to involve basegeneration (and binding). 133 As resumptive pronouns do not occur in long-distance relativization without PPs in Standard Dutch, I have taken Dutch PP-relatives to illustrate the insensitivity to islands. The question whether resumptive pronouns in PP-relatives and resumptive pronouns in long-distance relative clauses different from PP-relatives should be analyzed on a par, remains to be answered. 134 As noted by Sharvitt (1999:589), most discussions on resumption are primarily concerned with syntactic constraints that regulate the distribution of resumptive pronouns, instead of semantic/pragmatic constraints that govern them. For some observations regarding the semantics and interpretation of sentences involving resumptive pronouns (in Hebrew), see Sharvitt (1999). 135 The examples are taken from Salzmann (2006:207). The sentences in (13) show that a resumptive pronoun can occur in a strong island (13a), in a weak island (13b), in a possessor (13c), and inside a PP (13d)-(13e). 79 e. … de man waarvan ik denk dat ik < met een zuster van hem > op school heb gezeten the man whereof I think that I with a sister of him at school have been On the other hand, primarily based on reconstruction effects, it has also been argued that resumption is compatible with movement (cf. Pesetsky 1998, Aoun et al. 2001, Salzmann 2006). Given my assumption that all variants of long-distance relativization attested in the Dutch speaking language area (excluding PP-relatives) involve successive cyclic movement of the relative DP, the most straightforward option regarding the analysis of resumption in Dutch restrictive relative clauses would be the assumption that resumptive pronouns are simply the spell outs of traces/lower copies. Dialects would then differ with respect to whether they allow multiple spell out of chain links and if so, which lower copy they spell out.136 Interestingly, Pesetsky (1998:367) argues that “resumptive pronouns are partial pronunciations of traces”, more specifically, he assumes that only the φ-features of the trace get pronounced.137,138 Notice that this is very reminiscent of Van Koppen’s (2007) approach to the spell out of lower copies (cf. section 4.3.2.2). Suppose that this line of reasoning is on the right track, then we run into problems when faced with the empirical observation that resumptive pronouns often block the violation of some locality constraint, as illustrated for Hebrew in (14) and for English in (15). (14)139 a. b. (15)140 a. b. *ha-iSa Se dibarnu the-woman Op we-talked ‘The woman we talked about arrived.’ ha-iSa Se dibarnu the-woman Op we-talked al __ higia about arrived ale-ha about her __ higia arrived [Hebrew] *There is one worker who the company fired the employee [that had treated __ badly] There is one worker who the company fired the employee [that had treated him badly] That is to say, given the assumption that resumptive pronouns only appear to avoid a violation of some syntactic constraint, the variation in the presence/absence of resumptive pronouns in restrictive relative clauses in varieties of Dutch (and sometimes even within a single dialect) is unexpected. In other words, if the assumption that resumptive pronouns are only inserted if the structure crashes otherwise is correct, then we either expect no variation in the distribution of resumptive pronouns, or we expect an interpretation difference between structures involving resumptives and structures without resumptives. It is thus very important to investigate the following issues in Dutch dialects: (i) the 136 The implementation of this idea is not without problems. Recall that I followed Nunes (2004) and assumed that dialects differ with respect to the degree in which they allow morphological reanalysis. However, as the lowest (reduced) copy cannot undergo morphological reanalysis, given this approach, it cannot be spelled out. Thus, additional machinery is needed to account for the spell out of the lowest copy in the chain. 137 Recall that Van Koppen (2007) also assumed that reduced copies only contain the φ-feature set of the moved XP (but see footnote 116), and recall that this assumption was not sufficient to account for the fact that the intermediate copy could only be ‘spelled out’ as a d-pronoun, not as a personal pronoun (cf. footnote 104). In case of resumption, however, the lower copy can be spelled out either as a d-pronoun or as a personal pronoun. 138 Pesetsky’s (1998) proposal is stated within the framework of Optimality Theory. He argues that the constraints that govern movement interact with the constraints that determine pronunciation. More specifically, based on the observation that resumptive pronouns are a ‘repair strategy’ to avoid a violation of an island constraint, Pesetsky assumes the following two interacting constraints to account for the distribution of resumptive pronouns. (i) (ii) Island Constrants: *α ... [island ... β ... ], where β is the trace of α and unpronounced Silent Trace (SILENT-t): do not pronounce traces The first constraint ensures that the lower copy is spelled out in movement-out-of-island-contexts, and the second constraint makes sure that this spell out is as minimal as possible (partial pronunciation of the phrase (i.e. only φ-features) leads to a lesser violation of SILENT-t than full pronunciation of the phrase). 139 Example is taken from Sharvit (1999:590). 140 Example is taken from Pesetsky (1998: 364). 80 interpretation of (structures involving) resumptive pronouns (and those without) and (ii) the syntactic constraints that restrict their distribution. Moreover, it might be interesting to investigate whether there is a difference in semantics/syntax between resumptive pronouns in PP-relatives and resumptive pronouns in restrictive relative clauses without PPs (cf. note 133), i.e. resumptive pronouns in PPrelatives are obligatory in the majority of dialects that allow this construction (and in Standard Dutch), whereas resumptive pronouns are mostly absent in other structures of long-distance relativization. 81 REFERENCES ALEXIADOU, A. et al., eds, (2000). The Syntax of Relative Clauses. John Benjamins, Amsterdam. AOUN, J., L. CHOUEIRI & N. HORNSTEIN (2001). ‘Resumption, Movement, and Derivational Economy’. Linguistic Inquiry 32, 371-403. AOUN,J. & A. LI (2003). Essays in the representational and derivational Nature of Grammar: the Diversity of wh-constructions. MIT: MIT Press (= Linguistic Inquiry Monographs 40). AUWERA, J. van der (1984). ‘Subject vs. non-subject asymmetries in the relativization of embedded NP’s’. In: W. de Geest & Y. Putseys (eds.), Sentential complementation. Dordrecht/ Cinnaminson: Foris Publications. 257-269. BARBIERS, S. et al. (2005). Syntactic Atlas of the Dutch Dialects, Volume I. Amsterdam: Amsterdam University Press. BARBIERS, S. et al (2006). Dynamische Syntactische Atlas van de Nederlandse Dialecten (DynaSAND). Amsterdam, Meertens Instituut. URL: http://www.meertens.knaw.nl/sand/. BARBIERS, S. & O. KOENEMAN & M. LEKAKOU (2008). ‘Syntactic Doubling and the Structure of Chains.’ In: C.B. Chang & H.J. Haynie (eds.), Proceedings of the 26th West Coast Conference on Formal Linguistics, 77-86. Somerville, MA: Cascadilla Proceedings Project. BENNIS, H. (1983). ‘De PRO-drop-parameter en subjektloze zinnen in het Nederlands’. Spektator 126, 409-427. BENNIS, H. & L. HAEGEMAN (1984). ‘On the status of agreement and relative clauses in WestFlemish’. In: W. de Geest & Y. Putseys (eds.), Sentential complementation. Dordrecht/ Cinnaminson: Foris Publications. 33-53. BHAT, D.N.S. (2004). Pronouns. Oxford: Oxford University Press. BHATT, R. (2002). ‘The Raising Analysis of Relative Clauses: Evidence from Adjectival Modification’. Natural Language Semantics 10, 43-90. BIANCHI, V. (1999). Consequences of Antisymmetry: Headed Relative Clauses. Mouton de Gruyter, Berlin. BIANCHI, V. (2000). ‘The Raising Analysis of Relative Clauses: A Reply to Borsley’. Linguistic Inquiry 31, 123-140. BOECKX, C. (2003). Islands and Chains. Resumption as Stranding. Amsterdam: John Benjamins (= Linguistik Aktuell/ Linguistics Today 63). BORSLEY, R. (1997). ‘Relative Clauses and the Theory of Phrase Structure’. Linguistic Inquiry 28, 629-647. BRACHIN, P. (1973). ‘Een in dubbel opzicht “vreemde” constructie?’. De Nieuwe Taalgids 66, 59-60. BRACHIN, P. (1974). ‘Een in dubbel opzicht “vreemde” constructie?’. De Nieuwe Taalgids 67, 228234. CARDINALETTI, A. (1994). ‘On the internal structure of pronominal DPs’. The Linguistic Review 11: 195-219. CARDINALETTI, A. & M. STARKE (1999). ‘The typology of structural deficiency: A case study of the three classes of pronouns’. In: H. van Riemsdijk (ed.), Clitics in the Languages of Europe, 145-233. Berlin: Mouton de Gruyter. CARSTENS, V. (2003). ‘Rethinking Complementizer Agreement: Agree with a Case-checked Goal.’ Linguistic Inquiry 34:4, 393-412. CHOMSKY, N. (1977). ‘On Wh-movement’. In: P. Culicover et al. (eds.), Formal Syntax. Academic Press, New York, 71-132. CHOMSKY, N. (1981). Lectures on Government and Binding. Foris Publications, Dordrecht. CHOMSKY, N. (1982). Some Concepts and Consequences of the Theory of Government and Binding. Cambridge, MA: MIT Press. CHOMSKY, N. (1986). Barriers. Cambridge, MA: MIT Press. CHOMSKY, N. (1995). The Minimalist Program. Cambridge, MA: MIT Press. CHOMSKY, N. (2000). ‘Minimalist Inquiries: The Framework’. In: R. Martin, D. Michaels & J. Uriagereka (eds.), Step by Step: Essays on Minimalist Syntax in Honor of Howard Lasnik. Cambridge, MA: MIT Press. 89-155. CHOMSKY, N. (2001). ‘Derivation by phase’. In: M. Kenstowicz (ed.), Ken Hale: A Life in Language. Cambridge, MA: MIT Press. 1-52. CHOMSKY, N. & H. LASNIK (1977). ‘Filters and Control’. Linguistic Inquiry 8:3, 425-504. 82 CITKO, B. (2001). ‘Deletion Under Identity in Relative Clauses’. NELS 31, 131-145. CORNIPS, L. & W. JONGENBURGER (2001). ‘Het design en de methodologie van het SAND-project.’ Nederlandse Taalkunde 6, 215-232. CRAENENBROECK, J. VAN & M. VAN KOPPEN (2002). The locality of agreement and the CP domain. Handout Glow 2002, Amsterdam. DÉCHAINE, R.M. & M. WILTSCHKO (2002). ‘Decomposing Pronouns’. Linguistic Inquiry 33, 409-442. DEKKERS, J. (1999). Derivations & Evaluations. On the Syntax of Subjects and Complementizers. Doctoral Dissertation, Universiteit van Amsterdam. Holland Academic Graphics, The Hague. FELSER, C. (2004). ‘Wh-copying, phases and successive cyclicity’. Lingua 114, 543-574. FIENGO, R. & R. MAY (1994). Indices and Identity. Cambridge, MA: MIT Press. FOX, D. (1999). ‘Reconstruction, Binding Theory, and the Interpretation of Chains’. Linguistic Inquiry 30:2, 157-196. GALLEGO, Á.J. (2004). ‘T-to-C Movement in Relative Clauses’. In: J. Doetjes & P. González (eds.), Romance Languages and Linguistic Theory 2004. John Benjamins, 143-170. GROSU, A. & LANDMAN, F. (1998). ‘Strange Relatives of the Third Kind’. Natural Language Semantics 6, 125-170. HAEGEMAN, L. (1983). ‘Die en dat in West-Flemish relative clauses’. In: H. Bennis & W. van Lessen Kloeke (red.), Linguistics in the Netherlands 1983, 83-91. HAEGEMAN, L. (1992). Theory and description in generative syntax; A case study in West Flemish. Cambridge: Cambridge University Press. HAESERYN, W. et al. (1997). Algemene Nederlandse Spraakkunst. Groningen: Wolters-Noordhoff. 2e druk. HALLE, M. & A. MARANTZ (1993). ‘Distributed Morphology and the Pieces of Inflection’. In: K. Hale & S. J. Keyser (eds.), The View from Building 20. MIT Press, Cambridge, 111-176. HALLE, M. (1997). ‘Distributed Morphology: Impoverishment and fission’. In: B. Bruening, Y. Kang, M. McGinnis (eds.), MITWPL 30: Papers at the Interface. 425-449. HARLEY, H. & R. NOYER (1999). ‘Distributed Morphology’. Glot International 4:4, 3-9. HORNSTEIN, N., J. NUNES, & K.K. GROHMAN (2005). Understanding Minimalism. Cambridge University Press. KAYNE, R. (1976). ‘French relative “que”’. In: F. Hensey & M. Lújan (eds.), Current Studies in Romance Linguistics. Georgetown University Press, Washington, D.C. 255-299. KAYNE, R. (1994). The Antisymmetry of Syntax. Cambridge, MA: MIT Press. KAYNE, R. (2000). ‘Microparametric Syntax: Some Introductory Remarks’. In: R. Kayne, Parameters and Universals. Oxford Studies in Comparative Syntax. New York: Oxford University Press, 3-9.’ KOOPMAN, H. & D. SPORTICHE (2008). ‘The que/qui Alternation: New Analytical Directions’. Available at: http://ling.auf.net/lingbuzz/000638. KOPPEN, M. VAN (2005). One Probe – Two Goals: Aspects of agreement in Dutch dialects. Doctoral Dissertation, Leiden University. KOPPEN, M. VAN (2007). ‘Agreement with (the internal structure of) copies of movement’. In: N. Corver & J. Nunes (eds.), The Copy Theory of Movement. John Benjamins: Amsterdam/ Philadelphia. LARSON, R. K. (1985). ‘Bare-NP adverbs’. Linguistic Inquiry 14, 595-621. MALING, J. & A. ZAENEN (1978). ‘On the non-universality of a surface filter’. Linguistic Inquiry 9, 475-497. MAYR, C. (to appear). ‘On the Necessity of φ-features: The Case of Bavarian Subject Extraction’. In: P. Panagiotidis (ed.), The Complementizer Domain: Subjects and wh-dependencies. Oxford University Press. NEELEMAN, A. & H. VAN DE KOOT (2006). ‘Syntactic Haplology’. In: M. Everaert, H. Van Riemsdijk (eds.), The Blackwell Companion to Syntax, vol IV. Oxford: Blackwell Publishers, 684-710. NUNES, J. (2004). Linearization of Chains and Sideward Movement. Cambridge, MA: MIT Press. PERLMUTTER, D. M. (1971). Deep and Surface Structure Constraints in Syntax. New York: Holt, Rinehart and Winston. PESETSKY, D. (1982). ‘Complementizer-trace Phenomena and the Nominative Island Condition.’ Linguistic Review 1, 297-343. 83 PESETSKY, D. (1998). ‘Some Optimality Principles of Sentence Pronunciation’. In: P. Barbosa, D. Fox, M. McGinnis & D. Pesetsky (eds.), Is the best good enough? Optimality and Competition in Syntax. Cambridge, MA: MIT Press. 337-383. PESETSKY, D. & E. TORREGO (2001). ‘T-to-C Movement: Causes and Consequences’. In: M. Kenstowicz (ed.), Ken Hale: A Life in Language, 355-426. Cambridge, MA: MIT Press. PESETSKY, D. & E. TORREGO (2004). ‘Tense, Case and the Nature of Syntactic Categories’. In: J. Guéron & J. Lecarme (eds.), The Syntax of Time. Cambridge, MA: MIT Press. PLATZACK, C. (2000). ‘A Complement-of-N0 Account of Restrictive and Non-Restrictive Relatives: The Case of Swedish’. In: A. Alexiadou et al. (eds.). The Syntax of Relative Clauses. John Benjamins, Amsterdam, 265-308. RIZZI, L. (1990). Relativized Minimality. Cambridge, MA: MIT Press. RIZZI, L. & U. SHLONSKY (2007). ‘Strategies of subject extraction’. In: U. Sauerland & H.-M. Gärtner (eds.), Interfaces + Recursion = Language? Chomsky’s Minimalism and the View from Syntax-Semantics. Walter de Gruyter, 115-160. ROSS, J.R. (1967). Constraints on variables in syntax. Doctoral dissertation, MIT. Reprinted as Infinite Syntax! (1986). ABLEX Publishing Corporation. SAFIR, K. (1999). ‘Vehicle Change and Reconstruction in Ā-Chains’. Linguistic Inquiry 30:4, 587620. SALZMANN, M. (2005). ‘On an alternative to long A’-movement in German and Dutch’. In: M. Salzmann & L. Vicente (eds.) Leiden Papers in Linguistics 2.3, 107-128. SALZMANN, M. (2006). Resumptive Prolepsis; A study in indirect A’-dependencies. Utrecht: LOT (LOT Dissertation Series 136). SALZMANN, M. (2006b). ‘Reconstruction for Principle C in German relatives. In favour of the Matching Analysis’. Proceedings of SAM2, Utrecht Institute of Linguistics Working Papers, 65-79. SALZMANN, M. (2006c). ‘Reconstruction in German restrictive relative clauses’. In: J. van der Weijer & B. Los (eds.), Linguistics in the Netherlands 2006. John Benjamins, Amsterdam, 186-196. SAUERLAND, U. (1996). ‘The late insertion of Germanic inflection’. Manuscript, MIT. SAUERLAND, U. (2003). ‘Unpronounced Heads in Relative Clauses’. In: B. Schwabe & S. Winkler (eds.), The Interfaces: deriving and interpreting omitted structures. John Benjamins, Amsterdam/Philadelphia, 205-226. SCHIPPERS, A. (2006). ‘Variation in the CP-domain: Long-distance dependencies in the Dutch dialects’. Unpublished manuscript, Meertens Instituut – Rijksuniversiteit Groningen. SHARVIT, Y. (1999). ‘Resumptive Pronouns in Relative Clauses’. Natural Language and Linguistic Theory 17, 587-612. SHLONSKY, U. (1992). ‘Resumptive Pronouns as a Last Resort’. Linguistic Inquiry 23:3, 443-468. SMITH, C (1964). ‘Determiners and Relative Clauses in a Generative Grammar of English’. Language 40, 37-52. SMITS, R. (1988). The relative and cleft constructions of the Germanic and Romance languages. Foris Publications, Dordrecht. SPORTICHE, D. (2008). ‘Inward Bound: Splitting the wh-paradigm and French Relative qui’. Available at: http://ling.auf.net/lingbuzz/000623. TARALDSEN, K. T. (2001). ‘Subject extraction, the distribution of expletives and Stylistic Inversion’. In: A. Hulk & J.Y. Pollock (eds.), Inversion in romance and the theory of Universal Grammar. New York: Oxford University Press. VANDEKERCKHOVE, R. (2003). ‘De relativa die en dat in het dialect en de standaardtaal van WestVlamingen’. Nederlandse Taalkunde 8, 115-129. VRIES, M. de (2002). The Syntax of Relativization. Doctoral dissertation, Universiteit van Amsterdam. LOT Dissertations 53. WILTSCHKO, M. (1998). ‘On the internal and external syntax and semantics of (relative) pronouns’. Journal of Comparative Germanic Linguistics 2, 143-181. ZWART, J.-W. (2000). ‘A head raising analysis of relative clauses in Dutch’. In: A. Alexiadou et al. (eds.), The Syntax of Relative Clauses. John Benjamins, Amsterdam, 349-385. 84 The SAND data on relativization Kloeke Location nr. short subject (oral+ tel) short object (oral+ tel) long subject relative (372: oral data) long object relative (373: oral data) System(s) APPENDIX I A001p A001q Midsland/Midslân Lies die die die-die-Ø die-dat-Ø die-die-Ø die-dat-Ø VII I A002p West Terschelling/ West Skylge Oosterend die die die wie die weervan-dat-ie weervan-dat-die II die wie die die wie - (diet) die - weer van-dat-hem weervan-dat-die weervan-dat-Ø weer van-dat-m weervan-dat-Ø - (diet) dat (diet) - (diet) die wie die die(t) wert van-dat-er diet-dat-m dat-dat-Ø - diet-dat-die die-dat-die waart van-dat-die diet-dat-Ø - daart van-dat-m diet-dat-Ø - die-dat-Ø - diet-dat-Ø - A006p B001a B004p II B007p Hollum Schiermonnikoog/ Skiermûntseach Ferwerd/ Ferwert B013b B035p Anjum/Eanjum Kollum - (diet) - (diet) B041p B046b Visvliet Oosterbierum/ Easterbierrum Beetgum/ Bitgum die(t) die(t) - (diet) - (diet) - (diet) - (diet) - (diet) - (diet) wervant-dat-hij - (wervan-diet-Ø) wervan-dat-Ø - (diet) - (diet) wervan-dat-hij wervan-dat-m II C023p C029p C041a C108p Bergum/Burgum Kimswerd/ Kimswert Jorwerd/Jorwert Bakkeveen/ Bakkefean Waskemeer/ De Waskemar Kloosterburen Warffum Leermens Groningen wie (diet) - (diet) - (diet) dee dee dee die dei-dat-Ø dat-dee-Ø dei-dat-ij woar van-dad-ie dei-da-Ø dei-da-dei woar van-da-Ø I C123p C146t C148p Nieuw Scheemda Langelo Paterswolde dee dee dee die wie dein die die wie dei die dat dei-dat-Ø wovan-dat-Ø die-dat-hij wovan-dat-hij die-dat-tie die-die-Ø waarvan-dat-die wovan-dat-Ø dei-dat-Ø wovan-dat-hum I X wovan-dat-hom wovan-da-die die-dat-die II waarvan-dat-die wovan-dat-em II die-die-Ø die-dat-Ø die-dat-Ø die-da-die die-dat-Ø die-dat-Ø B052a B062p B076a B085c B127p B128a die C181p Spijkerboor dei die die wie dei die dat wie die C192p D003p E004p Onstwedde Monster Hippolytushoef dee die die C165p C175p Bellingwolde Een die wie die die die wie 85 II I I X I I E008p E014a E016p Schagerbrug Opperdoes Warmenhuizen E017p E031a E038a E051p E067q E079p E091p E092p E109p Schoorl Oostwoud Lutjebroek Schermerhorn Oost Knollendam/ West Knollendam Beverwijk Monnickendam Marken Amsterdam E121p E127p Weesp Huizen E130p E134p E147a Eemnes Buiten Katwijk aan den Rijn De Kwakel E170p Zoeterwoude E183p E192p E206p Kamerik Utrecht Boskoop F002p Hindeloopen/ Hylpen Heerenveen Makkinga/ Makkingea Rijs/ Riis F013p F022p F036q F038p die die die wie die die die die die die-dat-hij - die-dat-Ø die-die-Ø die-dat-Ø weervan-dat-ie dat-dat-die die-die-Ø die-dat- Ø weervan-dat-um die-die-Ø die-dat-Ø die-da-Ø die die die die wie die die wie die die die wie die dat die die die wie - (diet) die die die die wie die die wie die die die wie die wie die die die wie - (diet) - (van wie-dat-hij) - die-die-Ø - (van wie-dat-die) die-dat-Ø die-die-Ø die-dat-Ø die-dat-tie - die-die-Ø die-dat-Ø die-die-Ø die-da-Ø die-dat-Ø die-dat-hij - waarvan-dat-hij waarvan-dat-ie die-dat-Ø waarvan-dat-hem die-da-Ø II I wurt van-dat-tie wurt van-dat-die II - (diet) die - (diet) die die-dat-Ø die-dat-Ø I - (diet) die(t) wie die(t) die-dat-Ø die-die-Ø wer van-dat-er dert van-dat-him II die die dat wie die wie die die wie die die-dat-die waarvan-dat-hij waarvan-dat-die waar van-dat-die II die-dat-die die-dat-Ø IX waarvan-dat-die waarvan-da-um waarvan-dat-die II die-dat-ie woarvan-dat-ie die-die-Ø dat-da-die - (van wie-dat-Ø) woarvan-dat-um die-da-Ø dat-da-die waarvan-dat-hem - (diet) F042p F051p Lemmer/ De Lemmer Wolvega Diever F056p Oldemarkt die F060p F070p Steenwijk Giethoorn die die F077p Urk die wie die (as) die die die die F087p F102a F107a F112p F121p Staphorst Wezep Wechterholt Epe Spakenburg IX die die die wie die die die die die die die die (as) die die die dee 86 II V I VII I IX II V Nijkerk Hoog Soeren Gorssel die die die die die die F171p Lunteren die die F174p F178p F204p Beekbergen Brummen Doesburg die die die wie wovank-dat-tie waarvan-dat-tie G001b Appelscha/ Appelskea Bovensmilde die die die wie dat die die-dat-Ø dat-die-Ø die-da-Ø die-dat-die die-die-Ø die - - die die G001p die-dat-Ø die-dat-Ø wovan-dat-Ø die-dat-Ø (van wie-dat-hem) die-dat-Ø die-da-Ø F142p F149p F161p G007p Hooghalen Rolde die wie die wie die die G009p Gasselte die wie dei G039p Roswinkel G055p Zuid Sleen G058p Emmen G062a die G080p Emmer Compascuum Erica die die wie die G081a Klazienaveen die wie die wie G081c Zwartemeer G094p Gramsbergen die wie den G095p Coevorden die wie die den die G140b Sibculo G171p Vriezenveen G173p G177p Almelo Ootmarsum G180p G181q G198p G211p G255p Rossum Noord Deurningen Enter Losser Eibergen die wie die die wie die der den die der den die die dei dee die die die der den die die den die die die dee die 87 I III wovan-dat-die wovan-dat-Ø dei-dat-hij die-dat-Ø - die-dat-Ø die-dat-Ø die-dat-hij wovan-dat-hij die-dat-Ø III dat-die-Ø dat-die-Ø waarvan-dat-em VIII wor van-dat-Ø (wie-dat-Ø) die wie die wie die I die-die-Ø waarvan-dat-die die-dat-Ø worvan-dat-Ø (wie-die-Ø) die-dat-Ø wovan-dat-die - waarvan-dat-tie G006p I II II III II IX III IX I wovan-dat-hij die-dat-tie die-dat- Ø - die-dat-Ø - wovan-dat-Ø wovan-dat-hem X die-dat-die die-dat-Ø IX die-die-Ø die-dat-die woos-dat-Ø die-dat-Ø waarvan-dat-em V IX X wovan-dat-er die-dat-Ø wovan-dat-em II waar van-dat-ie wor von-dat-tie waarvan-dat-Ø waarvan-dat-dee waarvan-dat-e worvan-dat-Ø waarvan-da-m waarvan-dat-den - X II G257p Haaksbergen G280p Groenlo H013p Oostkerke H016p Oostende der die wie den wie die dat die H036p H046p H054p Brugge Nieuwpoort Gistel die die die H071p H081a die die H084p H108p H116p Hertsberge Bray Dunes/ Bray Duinen Veurne Diksmuide Torhout da dan die dan da(n) dan die die die H123p Tielt I002p I019p I023p I049p die wie die(t) wie(t) dan waarvan-dat-hij/die die-dat-Ø III waarvan-dat-e (van wie-dat-Ø) da-die-Ø die-dat-Ø III da-da-Ø IV dan da-da-Ø da-die-Ø da-die-Ø da-die-Ø die-da-Ø da-dan-Ø VI IV IV IV da-die-Ø da-diet-Ø da-dan-Ø da-dan-Ø IV IV dan dan da da-die-Ø da-die-Ø die-da-Ø da-dan-Ø da-dan-Ø da-dan-Ø IV IV die dan waarvan-dan-em II Oostvoorne Ouddorp Dirksland Zierikzee die die die die Stavenisse Steenbergen I061p I069p I074p Oostkapelle Goes Scherpenisse die die wie die die die die-da-Ø wervan-da-em dat-dat-Ø die-da-Ø da-da-die die-as-Ø die-da-Ø III II I054p I057p die die die die wie die (as) die(t) waarvan-da-die da-die-Ø waarvan-dat-ie wervank-da-die die-dat-Ø waarvan-dat-die I118p I123p I137p I142p Ossendrecht Oostburg Hoek Hulst die die die die I148p Doel I156p Sint Laureins I158p Eeklo die dat wie die dat die I163p Oosteeklo die I168c I175p Koewacht Be Sint Niklaas die dat wie die (an) die die die wie die die die (as) die (an) wie die (as) dat wie dien dan dien dan dien dien dat wie die (an/as) 88 die-dat-Ø waarvan-dat-tie die-dat-Ø die-dat-die die-dat-Ø die-dat-Ø waar van-dat-tie die-die-Ø die-dat-Ø da-dan-Ø da-dan-Ø da-dan-Ø die-da-Ø die-da-Ø die-dat-Ø daar van-da-em die-dat-Ø die-die-Ø die-da-Ø III I III I IX I I VII I die-dat-die die-die-Ø da-da-Ø die-dat-Ø die-dan-Ø I die-die-Ø die-dan-Ø V die-die-Ø da-dan-Ø die-dan-Ø die-dan-Ø V I die-da-Ø da-die-Ø die-die-Ø die-dat-Ø V die-die-Ø I178p Beveren die(t) dat I183a Oostveld I187p Aalter die(t) dat die dat I193p Lovendegem die I202p I204p I208p Oostakker Zaffelare Lokeren die die die (da) I241p Gent I251p I257p I VI die (an/as) dat die da(n) die dan die-dat-Ø da-dat-Ø die-da-Ø da-da-Ø waarvan-da-die - da-die-Ø da-dat-Ø da-die-Ø da-dan-Ø die-die-Ø die-dan-Ø die-die-Ø waarvan-dat-ij die-da-Ø die-die-Ø die-dien-Ø die-dan-Ø die-dan-Ø VII III I V die dien dat dien dien die (da) da(n) wie die die-die-Ø Laarne die dien VII V I Berlare die dat die (at) dat die (da) dat die da die dat die(ën) (da/as) dat die da die die die-dat-Ø da-dat-ij da-die-Ø da-die-Ø die-dien-Ø die-dan-Ø die-dan-Ø die-da-Ø da-da-Ø da-da-Ø IV IV da-da-Ø VI I260p Moerzeke I264p Sint Gillis Bij Dendermonde I274p Malderen K005p K023p Rotterdam Lopik die dat die die K039p Culemborg die K080p K096p Geldermalsen Sliedrecht K124a K124p die-die-Ø da-die- Ø da-dat-Ø waar van-dat-em da-da-Ø die-dat-Ø dat-dat-ie dat-da-die woarvan-dat-die dat-die-Ø die die die wie die die die-dat-ie dat-die-Ø dat-dat-ie die-die-Ø Moerdijk Strijen die die die die K131p K147p K152p K182p Waspik Vlijmen Fijnaart Gilze die die die die die die die die K183p Tilburg K189b K190p Nieuwmoer Zundert die wie die die die wie die (as) die die-dat-ie die-dat-hij waarvan-dat-hij die-dat-tie die-die-Ø wovan-dat-tie dat-dat-Ø waarvan-dat-tie K192p K209p Meer Sint Lenaarts K211p Weelde die (da) die wie die die (as) die (as) wie die (da) 89 waaroank-dat-tie die-da-Ø die-die-Ø die-da-Ø die-dat-Ø die-da-Ø die-dat-Ø die-die-Ø die-da-Ø die-da-Ø da-die-Ø die-da-Ø die-dat-Ø - die-die-Ø waar van-da-m der-da-Ø - waar van-dat-hij waarvan-da-m VIII VI, IV V VIII VII V IX III IX IX VII III III II III II Kapellen die die K229p ’s Gravenwezel dien dat K235p Beerse die wie K240p Arendonk K244p Antwerpen dien (da) dat die (at) dieën (as) dat die wie dat dien (da) dat die (as) K258p Zandhoven die(ën) (as/dat) dieë (as) dieë-da-Ø K274a Sint Jozef Olen dien die-dat-Ø die-da-Ø K276p Mol die - da-da-Ø K278p Lommel die-dat-Ø diene-da-Ø K291p Lier diene (da) dat die (da) die dat die dat diene (da) dat dieën (as) die-da-Ø die(ën)-die(ë)-Ø - da-die-Ø die-da-em die-die-Ø das-da-em - die(ë) dat da-dat-Ø da-da-Ø die-da-Ø die dieë dat die die (da) wie die dat da-dat-hij wovan-dat-n da-da-Ø da-da-Ø da-die-Ø dien-dat-Ø die-die-Ø dien-da-Ø die-die-Ø die-dat-Ø die dan den-dat-Ø die-den-Ø die-da-dem dat-da-Ø (wat-da-Ø) die-da-Ø die die dat die wie die woarvan-da-ij die-die-Ø woarvan-da-m die-da-Ø - (wie-wie-Ø) - (wie-wie-Ø) die-dat-Ø die-da-Ø I die waar van-dat-tie da-dat-tie die-dat-Ø die-da-Ø III K309p Tongerlo K320p Willebroek K330p K339p Mechelen Heist op den Berg K349a K353p Wolfsdonk Tessenderlo L034p Didam L038p Kilder L054p L071p Druten Nijmegen L119p Groesbeek L153p Reek L159p Cuijk dieen (da) dieë (da) dat die die dat die die (da) die dat wie die den dat wie die die dat die wie die wie die 90 die-da-Ø die-die-Ø dien-dat-Ø die-da-Ø - - da-dat-ie dat-da-Ø - die-da-Ø da-da-Ø dieë-da-Ø dieën-dieë-Ø I V I K221p die-da-Ø I I I VII VI I I V II V I L180b Vorstenbosch die die L199p Liempde die die die-die-Ø waar van-dat-ie die-dat-Ø die-dat-Ø L207p L245p L250p L255p L263p L267p Gemert Meterik Arcen Veldhoven Asten Maasbree die die den die die der wie die der die die die der wie die-die-Ø dat-dien-Ø den-dat-Ø die-dat-Ø dat-die-Ø woevan-dat-der L270p L286p L290p Tegelen Hamont Panningen der die der der die der L300p L318b Beesel Tungelroy L353p L360p L372p L377p L414p L416p L423p M009p Eksel Bree Maaseik Maasbracht Houthalen Opglabbeek Stokkem Aalten M013p Winterswijk M043p Gendringen der dee der die die die die die die die die wie die wie die N015p Izenberge der dee den die die die die die die die den wie dee wie den wie die N031p N034p Poelkapelle Hooglede die die N038p N054p N067p N072p N087p Roeselare Rubrouck/ Rubroek Poperinge Ieper Moorsele N102p hoe van-dat-er da-dat-ie oe van-dat-tee dee-dat-Ø waarvan-dat-der doa van-dat-Ø die-dat-Ø die-da-Ø waarvan-dat-Ø die-da-Ø - (wat-der-Ø) den-der-Ø die-da-Ø dat-der-Ø der-der-Ø woevan-dat-der woe van-dat-Ø da-da-m oe van-dat-um waarvan-dat-em - V III I I V I VIII II II II dat-die-Ø die-dat-Ø de-det-Ø dien-dat-Ø wovan-dat-e waarvan-da-m - (van wie-dat-m) die-dat-Ø de-det-Ø wovan-dat-um - - - - (wie-da-Ø) dan dat-die-Ø IV da-die-Ø da-die-Ø die die die die die dan die da (da) da dan dan da da-dan-Ø waarvan-dan-Ø da-dan-Ø da-da-Ø da-da-Ø da-dan-Ø da-da-Ø IV die an dat an - an-an-Ø N125p N130p Sainte Marie Cappel/ Sint Maria Kappel Steenvoorde/ Steenvoorde Wulvergem Wervik da-die-Ø dat-die-Ø waarvan-dat-n waarvan-dat-n da-die-Ø - die die da-die-Ø da-da-Ø da-dan-Ø da-dan-Ø IV VI N132p Menen die dan die dan dien dan da-da-Ø (van wien-dat-n) da-dan-Ø VI N104p 91 I I II IV IV IV - N141p N149p O014p Kortrijk Wallon Cappel/ Waals Kappel Morbecque/ Moerbeke Deinze O022p Merelbeke N154p O056p Mere O061p Aalst O080p O097p Waregem Bevere die die dan die da-die-Ø - da-dan-Ø - die die die-die-Ø - die dat die dat die dat die dat wie die wie dien dan dien dan die da die da wie die da(n) wie dien dan die-da-Ø da-dan-Ø die-dat-Ø die-die-Ø die-die-Ø da-dat-Ø die-dat-Ø da-dat-Ø die-dien- Ø die-dat-tie da-da-Ø da-die-Ø die-dat-ie da-die-Ø da-dan-Ø dien dan dien dan - die O104p Meilegem die O116p Strijpen die dat O152p Ninove O177p O228p Dilbeek Geraardsbergen die dat dat die O257p Kooigem die O265p Ronse die O275p Herne O286p P002p P018p Halle Humbeek Kamperhout P022p P051p Tremelo Lummen P065p Brussel P088p Leuven die dat dat dat diene (dat) die diene (da) die dat die P102p P112p P124p Boutersem Zoutleeuw Bosvoorde P133p P145p P176p Overijse Tienen Sint Truiden daa(ne) die die dat da die die die-die-Ø da-dat-Ø die dat da die(n) dan dan die-dat-Ø da-die-Ø da-dat-ij da-die-Ø die-da-Ø IV VII V da-da-Ø VI VI IV da-dan-Ø da-dien-Ø die-dan-Ø da-dan-Ø da-dien-Ø die-dan-Ø die-da-Ø da-die-Ø die-dat-Ø da-die-Ø VIII IX VI V I VIII VIII die-da-Ø IV VI I V III da-da-Ø da-dat-Ø da-dat-Ø da-da-Ø da-da-Ø da-da-Ø VI VI VI die-dat-ie die-dat-Ø dien-da-Ø I die da die dat da die da da-die-Ø die-die-Ø die-die-Ø - (van wie-da-Ø) da die die da-die-Ø da-da-Ø die-dat-Ø die-die-Ø woevan-dat-en dien dan die dat da da die da die die 92 da-dan-Ø die-dan-Ø V da-da-Ø die-da-Ø waarvan-dat-em die-da-Ø woevan-da-m da-da-Ø die-da-Ø waarvan-da-m da-da-Ø die-da-Ø da-da-Ø da-da-Ø die-da-Ø die-da-Ø VI I VI I II P212p Walshoutem da(t) woevan-dat-er woevan-da-er die da die - - Hasselt die dat die da die P219p Jeuk Q002p moevan-dat-em Q003p Q036p Q086p Genk Nuth Eigenbilzen die der die die der die Q096a Borgharen der Q096b Itteren Q112b Q113p Ubachsberg Heerlen die wie der der der dat die wie der der boevan-dat-er die-dat-er boevan-dat-er woevan-dat-er moevan-da-m dien-da-Ø da-der-Ø die-dat-Ø Q117p Q121p Q156p Nieuwenhagen Kerkrade Borgloon Q162p Tongeren Q170p Grote Spouwen Q193p Q194p Gronsveld Rijckholt dee dei Q222p Vaals die Q240p Lauw die S014a Merckeghem/ Merkegem die I II III IV V VI VII VIII IX X der dee die dat wie (wa(t)) wie (wat) die-dat-Ø waar van-dat-hij/die waar van-dat-hij/die dat-die-Ø die-die-Ø dat-dat-Ø die-die-Ø dat-die-Ø die-dat-hij/die waarvan-dat-Ø der-dat-Ø IX III III da-dat-Ø der-dat-Ø woevan-dat-der I II die dem die waar van-dat-hij woavan-dat-er moevan-dat-er wie (wa) die die dat (woe) dee die wie (wat) dem waarvan-dat-hij waarvan-dat-hem boechvan-dat-er - dee-dat-Ø dei-dat-Ø (wat-wat-Ø) dee-dat-Ø dei-dat-Ø waarva-dat-ten (wat-wat-Ø) - dem-dem-Ø - - die-dat-Ø waarvan-dat-die/’m die-dat-Ø dat-dat-Ø die-dat-Ø dat-dat-Ø die-die-Ø dat-die-Ø die-dat-Ø waarvan-dat-die/’m 93 II III die-dat-Ø der-dat-Ø da-da-Ø woevan-dat-em - (wat-dat-Ø) woa van-dat-m moevan-da-m die (boe) dan II - II II II I I APPENDIX II The SAND data on short relativization – geographic distribution Map A; short subject relative Map B; short object relative 94