1 70 CHAPTER 2: THE SYNTAX OF THE TOUGH CONSTRUCTION (TC) AND RELATED STRUCTURES 2.0 Introduction The literature concerning the syntax of the tough construction (TC) is both extensive and contentious and, consequently, a full review of the various syntactic accounts of this construction that have been previously offered is outside the scope of this thesis. Instead, I limit attention in this chapter to one specific analysis of the TC that has been widely endorsed in the generative literature, and which will serve as the basis for the experimental study to be discussed in Chapter 4. This is termed the null operator analysis of the TC and related constructions, originally proposed by N. Chomsky in an influential paper published in In this paper, Chomsky argued that the derivation of the TC involves the presence of an abstract or null operator in the embedded infinitive clause, which allows co-reference to obtain between the empty object of the infinitive and the matrix subject DP. While the following discussion will illuminate the fact that Chomsky s original analysis has been subject to certain modifications in the years following 1977, I will nevertheless demonstrate that the basic aspects of this analysis remain well accepted in generative syntactic theory, and have even been adapted to suit the theoretical tenets of the Minimalist Program proposed by Chomsky (1995a/b, 2000, 2001). I pursue two primary goals in this chapter. The first is to evaluate the theoretical validity of the null operator analysis of the TC and the second is to establish the applicability of this same analysis to the object-gap degree construction (ODC), object-gap purpose construction (OPC), and infinitival relative construction (IR). I focus on these four constructions in particular as all feature in my experimental study of children s acquisition of the TC, which will be discussed in Chapter 4. As will be further detailed in that chapter, I follow a precedent established in the literature (Goodluck and Behne 1992, Goodluck, Finney, and Ling 1995) and take as a working assumption that the interpretation of constructions that involve a null operator-gap dependency requires relatively advanced syntactic ability. On this basis, I propose the following experimental hypothesis: If a child lacks the syntactic ability to assign a target-like interpretation to a construction that contains a null operator-gap
2 Chapter 2: The Syntax of the TC and Related Structures 71 dependency, then the child should demonstrate a concurrent inability to interpret the TC, ODC, OPC, and IR in a target-like manner. Alternatively, I take the null hypothesis to hold that the child who possesses adult-like syntactic ability should interpret all four structures in a target-like manner. Of course, there also remains the possibility that a child could perform inconsistently across the four constructions. This pattern of performance would suggest that each construction is acquired independently of the others and would furthermore provide support for one, or both, of the following claims. The first, and most straightforward, is that syntactic complexity is not implicated in delayed acquisition of the TC and the second is that the syntactic analysis of the TC that is endorsed in this chapter may not be valid for each of the constructions under investigation in my experimental study. As any proper evaluation of the latter possibility will require a thorough knowledge of both the similarities and differences that exist between the four constructions, I will use this chapter to detail the properties of each in turn. 2.1 Organization of chapter In the following section, 2.2, I review theoretical treatments of the syntax of the TC that have been offered in the generative literature. In 2.2.0, I begin with a consideration of the types of theoretical considerations that preceded Chomsky s 1977 analysis of the TC, which was itself developed in the framework of the Extended Standard Theory (EST). In 2.2.1, I review Chomsky s original analysis of the TC and discuss certain modifications of this analysis that were necessitated by the introduction of Government and Binding (GB) Theory. After Nanni (1980), I also advance the claim that speakers of English have two means of forming the TC, one lexical, involving formation of a complex tough predicate, and the other syntactic. As Nanni points out, it is only the second of these derivational options that allows longdistance extraction of the embedded object argument (e.g. That dress i was hard to imagine anyone wanting to wear e i ), a syntactic operation which was earlier considered in of Chapter 1. In 2.2.2, I offer my own recommendations for modification of the null operator analysis of the TC. In particular, I endorse Lasnik and Stowell s (1991) proposal that
3 Chapter 2: The Syntax of the TC and Related Structures 72 the empty object argument in the TC should be analysed as a null referential element, or null epithet. Following Rizzi (1997), I also maintain that the null operator itself is incorrectly analysed as quantificational. Thirdly, I take issue with the standard assumption that the null operator in a null operator structure (NOS) is projected in embedded Spec CP. After Heggie (1993), I instead contend that there are two types of null operators that are attested cross-linguistically, only the first of which, the zero topic operator, is projected in CP, while the second, which features in the TC and OPC, is projected in the I- (for inflectional) domain. Continuing a line of argument that began in of Chapter 1, I propose that the latter type of operator specifically adjoins to embedded IP. In 2.2.3, I consider Minimalist adaptations of the null operator analysis of the TC. As I detail, a direct movement account of the derivation of the TC, first proposed in the EST framework, once again becomes a legitimate consideration in Minimalism. In particular, Chomsky s (2000, 2001) proposal for phase-based movement of constituents through CP and vp (for light verb phrase) removes the earlier GB prohibition on combining A- and A-bar movement in a single derivation. I then offer a critical review of several Minimalist analyses of the TC, including those offered by Rifkin (2001), Hornstein (2001) and Richards (2001), noting that all these accounts face the same problem of explaining how the object argument of the embedded verb is allowed to undergo syntactic displacement in tough-movement (TM), despite the fact that the object is case-marked by the embedded verb in situ and should therefore be frozen in place (cf. Chomsky 2000:123, 2001:6). I evaluate alternative proposals for addressing this issue, ultimately advocating an analysis of the TC in which the embedded object argument is merged in, rather than moved to, matrix T (for tense). Section 2.3 contains a discussion of various aspects of the diachronic development of the TC. I include consideration of this historical data not only to provide insight into the appropriate syntactic representation of the present-day TC but also to serve as a preface for certain discussion contained in Chapter 5, where I propose an original theoretical analysis of the acquisition of the TC.
4 Chapter 2: The Syntax of the TC and Related Structures 73 In 2.4, I propose a syntactic representation of the TC, which is supported by certain previously reviewed empirical evidence, as well as by my own evaluation of various theoretical proposals. As I detail, my analysis of the base and logical form (LF) representations of the TC differs in certain key respects from other analyses that have been offered in the literature, particularly with regard to the base representation of the theme and experiencer arguments of the tough adjective. I go on to offer a number of reasons why the analysis I propose can be considered superior to alternative accounts. Finally, in 2.5, I examine the syntactic properties of the ODC, OPC, and IR, each of which has been argued to share the null operator analysis proposed for the TC, and each of which features in my experimental study of children s acquisition of the TC. I review standard analyses of the syntax of the three constructions, pointing out certain aspects of these analyses that I consider problematic or that have been identified as being problematic in earlier treatments. On the basis of my evaluation of both theoretical and empirical evidence, I propose a syntactic representation for each construction which serves as the basis for the experimental hypotheses that I present in Chapter Existing theoretical analyses of the TC Some controversial issues In the generative literature, the interest in TM as a syntactic construct is longstanding, dating at least back to Lees (1960) study of multiply ambiguous adjectival constructions, which included structures of the form NP is hard to VP. Lees proposed a transformational derivation for constructions of this type, which involved extraposition of the clausal subject of a sentence such as To convince him is hard and stranding of the direct object of the infinitive verb in matrix subject position (e.g. [Him] is hard to convince) (ibid.:217-8). Certain subsequent proposals, such as the transformational accounts of TM offered by Rosenbaum (1967), Postal (1971), and
5 Chapter 2: The Syntax of the TC and Related Structures 74 Bresnan (1971), modified Lee s original analysis by allowing for direct movement of the embedded object to the matrix subject position of the sentence. 1 There were also proposals advanced in the early generative literature for transformational derivations of the TC in which no syntactic movement was posited. For example, Lasnik and Fiengo (1974:539) credit Ross (1967:231) with originally advancing the idea that the TC is transformationally derived through application of a rule of object deletion, rather than through object movement. In their own formulation of the deletion account of TM, Lasnik and Fiengo proposed that the TC is derived from a base structure such as that illustrated in (1), below, with the object argument of the embedded infinitive verb deleted at phonetic form (PF) under identity with the matrix subject DP (op.cit.:536): (1) Franz is easy to please Franz. One crucial way in which this type of analysis differs from a movement or raising account of TM is with regard to the assumption that the matrix subject argument of the TC is generated in situ, as is the object argument of the embedded infinitive verb. In the EST, these assumptions were generally considered unproblematic (although see Postal and Ross 1971 for discussion of certain theory-internal complications). Yet, as we shall later see, these same assumptions carry different theoretical import in the GB framework and in Minimalism. One piece of evidence that Lasnik and Fiengo offered against a movement analysis of the TC concerns the types of sentences that were previously presented in 1.2.0, Chapter 1. These are tough sentences that involve either: (a) matrix verb be inflected for the progressive aspect (cf. (2), below), (b) a matrix adverb that expresses intentionality (e.g. deliberately) or (c) a modal auxiliary, such as must. (2) That dog is being difficult to control. 1 Although each of the three referenced authors offered slightly different analyses of TM, these differences do not affect the general claim made here, which is that each advocated a derivation of the TC that involved direct movement of the embedded object.
6 Chapter 2: The Syntax of the TC and Related Structures 75 As Lasnik and Fiengo observed, selection restrictions hold between the matrix subject DP and matrix predicate in each of the three types of sentences. Such evidence, they argued, is inconsistent with a movement account of the TC, which postulates basegeneration of the subject argument in an embedded, rather than main, clause. They furthermore pointed out that according to a standard movement account of TM, the derivational source of a sentence such as (2) would be a sentence such as (3a) or (3b), below (or possibly both, depending on the precise formulation of the movement transformation). This assumption is problematic, however, in that neither of the purported source sentences is grammatically well formed: (3) a. *It is being difficult to control that dog. b. *To control that dog is being difficult. Personally, I do not believe that the type of evidence noted above need necessarily rule out a movement derivation of the TC. This is because, as discussed in of Chapter 1, I contend that (2) involves an entirely different derivation than that of a standard TC. For example, Partee (1977) has argued that a sentence such as (2) involves joint θ-marking of a subject DP by both progressive main verb be and a complex adjectival predicate (e.g. difficult-to-control). Moreover, she maintains that the copular verb in (2) can be distinguished in type from the copular verb that appears in the standard TC; specifically, she proposes that the former is featurally specified as [+active] while the latter is specified as [+stative] (ibid.:300-11). According to this line of argument, it is not surprising then that matrix be in (2), as an active verb, can impose selection restrictions on the subject argument that it also θ-marks (see also the discussion in Williams 1984:138-44). As regards Partee s claim that sentences such as (2) involve the presence of a complex tough adjective, I submit that the comparatively ungrammatical status of a sentence such as (4) serves as supportive evidence: (4) *That dog i is being difficult to expect Bill to be able to control e i.
7 Chapter 2: The Syntax of the TC and Related Structures 76 Specifically, I would argue that, in (4), it is not the long-distance referential relationship that holds between the matrix subject DP and embedded object gap that is problematic per se; this is because I have identified this as one of the defining characteristics of TM (see , Chap. 1). Rather, I contend that (4) is ill-formed because it requires the presence of a complex tough adjective, which may consist only of a tough adjective and a single infinitive complement (cf. (a) difficult-to-control (dog) and (a) difficult-to-expect-bill-to-be-able-to-control (dog)). As earlier noted, Nanni (1980) has explicitly argued that English provides two distinct means of forming TCs, one of which is strictly lexical, and which therefore cannot involve long-distance extraction of an object argument, and the other syntactic. I will entertain further consideration of these claims later in this section. Another piece of evidence that has been cited as being problematic for a movement account of the TC concerns the contrast in grammaticality between (5a) and (5b), below, where an entire gerund phrase has been tough-moved: (5) a.?it has its advantages being easy to please. b. Being easy to please has its advantages. Once again I contend that a sentence such as (5b) does not involve the same derivation as the standard TC. Specifically, after Schacter (1981), I maintain that (5b) involves direct merge of a complex tough adjective in the subject position of the sentence, rather than upward movement of an embedded constituent. And, as regards (5a), my interpretation and preferred pronunciation of this sentence are consistent with a right-dislocated subject; thus, I think the derivation of (5a) most likely involves extraposition of a complex tough adjective that is initially merged in the subject position of the sentence. (Interestingly, Schacter ibid.:432 cites the very same evidence as being problematic for a deletion account of the TC, since he argues that there is no plausible object nominal that can be deleted under identity with the matrix subject argument in a sentence such as (5b).) Turning now to a third oft-cited piece of evidence that is held to be problematic for a movement derivation of the TC, this concerns a purported lack of syntactic
8 Chapter 2: The Syntax of the TC and Related Structures 77 connectivity in constructions of this type. 2 As noted by, inter alios, Jacobson (1992), Wilder (1991), and Mulder and Den Dikken (1992), the subject of the TC can be a constituent that the embedded infinitive verb does not c-select (cf. (6a&b), below, adapted from Jacobson s #49, ibid.:286): (6) a. [That language is learnable] is hard for any theory to capture/express. b. *It is hard for any theory to capture/express [that language is learnable]. As Jacobson notes, the verbs to capture and to express subcategorize for DP rather than clausal complements, and therefore it is not clear how a movement account could explain the presence of a fully clausal subject in (6a). However, according to my own judgment, both (6a) and (6b) involve a nominal rather than clausal complement, with ellipsis of a DP that serves to introduce the clause (i.e. The fact that language is learnable ). Thus, I think Mulder and Den Dikken (ibid.:307) provide a more appropriate example of the problem when they cite the contrast illustrated in (7a&b), below (adapted from their #13a): (7) a. For him to be top of his class is hard to believe. b. *It is hard to believe for him to be top of his class. For me, (7b) is acceptable only with a complement clause headed by that rather than by for, while (7a) is acceptable, if somewhat awkward. Given that the verb believe does not c-select the type of complement clause that appears in the subject position of (7a), the grammaticality of the sentence would thus require some special explanation 2 I have chosen not to discuss evidence from idiom chunk extraction above, although this type of evidence is standardly considered relevant in determining whether syntactic connectivity holds between the subject of the TC and the embedded object gap. This is because I find that judgments regarding this type of evidence are not consistent; for example, Goh (2000b) judges structures such as (i), below, grammatical (although questionable) (i)?headway should be easy to make in cases like this, but I ve gotten nowhere. (his #43a, ibid.:9) while Chomsky (1981) and Lasnik and Fiengo (1974) judge them completely unacceptable. Since I find no clear consensus of opinion in the literature with regard to availability of idiom chunk extraction in TM, I have opted not to consider this type of evidence here.
9 Chapter 2: The Syntax of the TC and Related Structures 78 if, as is assumed in standard movement accounts of the TC, it is the object argument of the embedded verb believe that is held to be tough-moved to the matrix subject position of the sentence. Mulder and Den Dikken (ibid.:307-8) account for the discrepancy between (7a) and (7b) by challenging the assumption that there is a direct referential link between the embedded object gap and the subject argument of the TC. Instead, they propose that the subject of the TC is licensed by a complex tough predicate, which minimally consists of a modal operator, a tough adjective, and a tough infinitive that contains an unbound gap. Thus, c-selection of the subject clause in (7a) is effected in a different manner than c-selection of the clausal complement in (7b), where the subcategorial properties of the verb believe take precedence. While I personally think this is a plausible explanation of the discrepancy, the significance of the type of evidence presented in (7) is likely to remain a source of contention in the literature, given the diversity of opinion that exists more generally with respect to the appropriate structural analysis of the TC. I return to a consideration of Mulder and Den Dikken s thesis in the following section The null operator analysis of the TC In this section, I examine one particular syntactic treatment of the TC, termed the null operator analysis, which has achieved widespread currency in the generative literature. Originally described as the wh-movement analysis of the TC, the null operator account of the construction offered an interesting alternative to previously proposed analyses, as it incorporated aspects of both movement and deletion accounts of TM. Chomsky first presented the account in an influential paper written in 1977 within the framework of the EST. He was inspired by certain similarities, first observed by Ross (1967), in the syntactic behaviour of a number of different constructions; these included comparatives, topicalization constructions, clefts, infinitival indirect questions, and infinitival relatives. Additionally, Chomsky included in this comparison set what I have termed the TC, the ODC (e.g. The boys are too young to teach), and the AODC (e.g. Butterflies are pretty to look at). In particular, he demonstrated that the same island effects that are observed in overt wh-
10 Chapter 2: The Syntax of the TC and Related Structures 79 movement constructions are attested in all of the listed constructions, even though none feature an overt wh-operator. Accordingly, he proposed that all involve whmovement but that the overt wh-element is obligatorily deleted at some stage from the phonetic representation of the sentence. 3 In the 1977 paper Chomsky provided a list of specific syntactic criteria for identifying the aforementioned set of constructions, and I will follow Contreras (1993) in terming these null operator structures (hereafter, NOS). As these criteria were formulated in the EST framework, some of the terminology used and concepts referenced have now become obsolete. For this reason, I offer the set of criteria outlined in Browning (1987), who adapted Chomsky s proposals in conformance with developments in GB Theory. After Browning, then, structures evidencing either overt or covert whmovement are held to be subject to the following conditions (ibid:17): (8) a. it (i.e. wh-movement) leaves a gap; b. it (i.e. wh-movement) obeys the Subject Condition (Huang s 1982 Condition on Domain Extraction (CED)) 4 ; c. the trace left by wh-movement is subject to the Empty Category Principle (ECP) and Binding Condition C. It was anticipated that any wh-movement structure that failed to meet one, two, or all three of these three conditions would be ungrammatical. In (9) and (10) below, I 3 It is important to acknowledge before proceeding that the null operator analysis of the TC and related constructions was formulated only with reference to English data. Chomsky (1982), for example, has openly questioned whether this particular analysis is correct for Romance TM. And Comrie and Matthews (1990:45-50) have argued that German TM lacks certain characteristic features of wh-movement constructions, such as long distance extraction and licensing of parasitic gaps, and is therefore unlikely to be derived through wh-movement of a null operator (see McDaniel, Chiu, and Maxfield 1995 for an opposing view, however.) Moreover, Comrie and Matthews suggest that the tough lexical items that feature in German TCs may actually be adverbs rather than adjectives, although they concede that this claim is difficult to verify given that the relevant German adverbs and adjectives cannot be morphologically distinguished. For further information on cross-linguistic differences in the syntactic analysis of the TC, see the references cited in ftnt. 24 of Chapter 1. See also Massam (1985, 1992) for a discussion of contrasts in the syntactic representation of TM in English as compared to certain non-indo-european languages. 4 Huang (1982:505, as cited in Contreras 1993) formulates the CED as follows: A phrase A may be extracted out of a domain B only if B is properly governed. As Contreras observes, this condition thus effectively bars movement out of subjects or adjuncts.
11 Chapter 2: The Syntax of the TC and Related Structures 80 provide examples of the predicted types of violations, which are adapted from examples originally provided by Browning (ibid.:20-4). The (a) examples illustrate violations of the CED, the (b) examples, ECP violations, and the (c) examples, violations of the subjacency condition on movement of a wh-constituent. 5 (NB: The notation t stands for trace.) (9) Degree construction a. *John i is too well liked for us to think that friends of t i would betray him. b. *John i is too incompetent for us to convince Mary that t i should get the job. c. *Who i was the job j good enough for us to offer t j to t i? (10) Tough construction a. *John i would be difficult to convince Mary that pictures of t i should appear on the front page. b. *John i would be difficult to convince Mary that t i should get the job. c. *What j is John i difficult for us to give t j to t i? One important assumption of the proposed wh-movement analysis of the TC and related constructions was that syntactic movement was restricted to the embedded clause only. That is, the idea that the direct object of the embedded verb moves to the subject position of the sentence was never entertained. Instead, adherents of the analysis maintained that the DP subject of the TC is base-generated in its S-structure (for surface structure) position and that it is in this position that the subject receives its referential index. The details of the specific analysis were as follows. On the basis of the types of evidence reviewed in (9) and (10) above, it was proposed that movement of the direct 5 Browning (1987) adopts the version of the subjacency condition developed in Chomsky s Barriers (1986a) in which optimal movement is held to cross neither IP nor NP. This is a considerable simplification of the definition of subjacency proposed therein, however, and so the interested reader is referred to the original source for a more detailed discussion.
12 Chapter 2: The Syntax of the TC and Related Structures 81 object argument of the embedded infinitive verb results in the formation of an operator-gap dependency in the lower clause. In earlier work, Chomsky (1977) had suggested that the operator took the form of an overt wh-phrase, which was later deleted at the level of phonological specification, along with the clause-introducing complementizer for (ibid.:85-6). In later work, however (see Chomsky 1980, 1982:31), he maintained that phonological deletion of the operator need not take place since the operator was inherently null, that is, devoid of phonological features. Movement of the null operator to the Spec of CP would put the operator in an appropriate position to bind the object trace, which, according to Chomsky s later formulation of the analysis, was assumed to share the featural specification of an overt trace of wh-movement (i.e. [-a(naphor)], [-p(ronominal)]). 6 According to Chomsky, the infinitive clause could thus be seen to qualify as a syntactic predicate given that it involves an open position (i.e. the position of the whtrace), which requires co-indexation with some external argument in order to be considered saturated or referentially satisfied (see also Faraci 1974, Bresnan 1979, Aoun and Clark 1985, and Browning 1987:51). (Note that, for others, such as Mulder and Den Dikken 1992 and Rothstein 2001, it is not the presence of the gap itself that confers predicate status on the embedded clause of the TC but rather the fact that the gap is bound by a null operator.) In the TC, the matrix subject DP was held to satisfy the referential requirements of the predicative infinitive clause, with the subject 6 This latter assumption in fact proved to be a contentious point in the GB formulation of the analysis, with competing arguments advanced for both the featural specification of the empty object and null operator. For example, Cinque (1990) argued that the object gap of the TC is occupied by pro, while Browning (1987), Epstein (1989), and Contreras (1993) assigned this same analysis to the null operator. Since the significance of these arguments is particular to the GB framework alone, I will not pursue these various claims here.
13 Chapter 2: The Syntax of the TC and Related Structures 82 receiving a referential index on the basis of the predication relation that was presumed to exist between the two constituents. 7 An obvious question arises, though, which is why Chomsky and those subscribing to the null operator analysis were unwilling to entertain the possibility of a direct movement account of the TC. The answer - or, more correctly, answers - to this question had to do with certain theoretical considerations that were particular to the GB framework. One problem concerned the fact that direct co-indexation of the empty object and matrix subject argument would constitute a binding theory violation. Recall that Chomsky (1977) had proposed that the element occupying the embedded object gap should be analysed as a variable, consistent with its status as a wh-trace. He later recognized this proposal as being problematic, however, since a wh-trace shares the features of a lexical R- (for referential) expression (i.e. [-a, -p]), and, accordingly, the distribution of this element would be subject to Condition C of the binding theory. As Condition C prohibits A-binding of a variable within the domain of the head of its chain, direct binding of the embedded variable trace by the matrix subject argument would thus be illicit (Chomsky 1981, 1986b). 7 For Browning (1987:2), the syntactic relationship between the two constituents is defined somewhat more precisely in terms of the existence of an agreement chain, described as: two categories (which are) connected by an unbroken sequence of independently motivated instances of agreement, e.g. subject-predicate agreement, SPEC-HEAD agreement, etc. In the case of the TC, the agreement chain was postulated to involve multiple relationships of this sort, which are detailed in Browning (ibid.; see also the discussion in Rothstein 1991a:146-7). While I will not explore the theoretical import of Browning s proposals here, I wish to note that her analysis highlights the often overlooked consideration that there may be agreement relations attested in the syntax of the English TC beyond that of simply subject-copula agreement. As case in point, I am aware that in Romance languages such as French and Spanish, the tough adjective itself is inflected to agree with the subject argument of the TC, as illustrated in (ia&ib) below: (i). a. Ces jeux sont difficiles à jouer. These games (3P-PL-M) are difficult (3P-PL-M) to play. (Where 3P-PL-M stands for 3 rd person, plural, masculine gender. ) b. Los vestidos fueron faciles de lavar. The clothes (3P-PL-M) were easy (3P-PL-M) to wash. Rather atypically, the examples cited above do not display gender agreement between subject and adjective, but only number agreement. Nevertheless, because there is clear evidence for an agreement relation that holds between the two constituents, it is reasonable to consider that the same type of relation holds in the English TC, even though it remains morphologically unmarked.
14 Chapter 2: The Syntax of the TC and Related Structures 83 A second objection to a direct movement account of the TC centred on the observation that creation of a CHAIN (Chomsky 1986b) which includes only the matrix subject and object gap would violate the theta criterion, since the CHAIN soformed would include two case-marked positions. As Kim (1995) observes, this particular problem is encountered regardless of whether the TC is analysed as involving direct NP movement or direct wh-movement of the embedded object. Finally, a third objection to a direct movement account of the TC was that the CHAIN formed would contain two theta-positions (hereafter, θ-positions), that of the matrix subject and that of the embedded object, in violation of the uniqueness requirement of the theta criterion (Chomsky 1986b:97). With regard to the last of these concerns, there are two main ways in which this particular problem was addressed in the literature. The first is exemplified by Brody (1993) and Koster (1987), who have argued that the uniqueness requirement of the theta criterion should be weakened to allow a single θ-role to be associated with more than one argument. The second approach to the problem was taken by Chomsky (1981), who proposed a syntactic means by which the grammatical status of the whtrace could change from variable to anaphor. Since I think Chomsky s proposal has important implications for any syntactic analysis of the TC, including those developed outside GB Theory, I will review the main points of his argument below. Chomsky offered what I will henceforth term his reanalysis account of the derivation of the TC as a means of addressing several perceived weaknesses of the original whmovement account of the construction. The first weakness he identified as a paradox of θ-theory (1981:309). As he explained, the projection principle holds that an argument is inserted in D-structure (for deep structure) in the same syntactic position in which it is assigned a θ-role by a licensing predicate; this principle would thus mandate that the subject argument of the TC originates as an embedded object argument and only later comes to occupy the matrix subject position of the sentence. As Chomsky observed, the latter supposition is given further support by the fact that sentences containing a tough adjective may feature expletive it as subject, a constituent which cannot occupy a thematic position in the syntax.
15 Chapter 2: The Syntax of the TC and Related Structures 84 Nevertheless, as Chomsky noted, the subject position of the TC is also reasonably analysed as being a thematic position, and therefore one capable of hosting a lexical argument at D-structure, when the distribution of idiom chunks is considered. Specifically, because idiom chunks and other non-arguments, such as expletive there, are generally prohibited from occupying the matrix subject position of the TC, it is reasonable to analyse this as a θ-position, that is, as a position that cannot host non-arguments (ibid.). (Recall, however, that in ftnt. 2 of 2.2.0, I previously questioned the validity of the idiom chunk extraction evidence that Chomsky relies on here.) For Chomsky, the apparently contradictory properties of the subject position of the TC - that is, the paradox of θ-theory - could only be reconciled according to the assumption that the TC undergoes obligatory syntactic reanalysis at LF. He proposed that the TC, with a structural representation as in (11a) below, would, after reanalysis, take the form illustrated in (11b), where the tough infinitive is absorbed into a complex tough predicate (his examples #3ii and #19, ibid.:309 & 312; note that S stands for sentential projection in the EST): (11) a. John i is [ AP easy [ S PRO i ] [ S PRO to please t i ]]. b. John i is [ AP [ A easy to please] e i ]. After structural reanalysis, Chomsky proposed that the embedded trace would no longer be A-bar bound within the lower clause but would instead be A-bound by the matrix subject argument. Accordingly, the empty element in (11b) would have the status of an anaphor rather than a wh-trace, and the binding relation observed in the reanalysed structure would satisfy Condition A of the binding theory. 8 In this way, the Condition C violation evidenced in (11a) could be circumvented. Furthermore, as 8 Implicit in the account described above is the assumption that the featural specification of an empty category is not an inherent property of the empty element itself but is instead established on the basis of the particular syntactic function that is served by the empty element (Chomsky 1981, 1982). This assumption has been termed the contextual definition of empty categories and was ultimately abandoned as a legitimate principle of GB Theory (see, e.g., Brody 1984, 1985, or Chomsky 1986a:57). Consequently, Kim (1995) has argued that the reanalysis account of the TC is seriously compromised, being predicated on what is now considered a flawed assumption. Since I do not believe that this particular criticism has any relevance outside the scope of GB Theory, however, I will not consider it further here.
16 Chapter 2: The Syntax of the TC and Related Structures 85 contrasted with a direct movement account of the derivation of the TC, the reanalysis account complied with GB rules for licit CHAIN formation. Specifically, it was proposed that after reanalysis of the tough adjective and infinitive as a complex constituent, the empty object would no longer receive case. Consequently, the CHAIN in (11b), consisting of the matrix subject as head and anaphoric element as foot, would contain only one case-marked position and would therefore be considered well-formed in comparison to the CHAIN postulated in (11a). The reanalysis account was also designed to address a second well-recognized weakness of the original wh-movement analysis, which concerns the type of contrast illustrated in (12a&b), below; specifically, wh-extraction of an object argument is allowed in (12a), even though such extraction should give rise to a wh-island (i.e. subjacency) violation, as in (12b) (adapted from Chomsky s examples #135a&b, 1977:105; Op stands for null operator): 9 (12) a. Which violin i is the sonata k easy to play t k on t i? from: The sonata k [ IP is easy [ CP Op i [ IP to play t k on this violin]]]. b. *Which sonata k is this violin i easy to play t k on t i? from: This violin i [ IP is easy [ CP Op i [ IP to play the sonata on t i ]]]. If, however, reanalysis is assumed to take place in the derivation of the TC, with the reanalysed adjective-infinitive phrase assuming the status of a lexical constituent, then the underlying structure of a question such as (12a) would not be as depicted above but rather would be as depicted in (13) (adapted from Chomsky s #20, 1981:313). (13) The sonata i is [ AP [ A easy to play] t i on [wh-violins]]. In (13), the relevant point to note is that extraction of the wh-element is proposed to take place independently of the reanalysed complex tough adjective, that is, from a 9 Given space limitations, I have chosen to focus on only one problematic aspect of wh-extraction when it occurs in the context of the TC. For a more detailed discussion of related issues, the interested reader is referred to Chomsky (1977, 1981).
17 Chapter 2: The Syntax of the TC and Related Structures 86 peripheral position in the clause. Thus, it is anticipated that questioning of the prepositional object will be unproblematic in terms of generally accepted constraints on wh-movement. In (14), however, which Chomsky proposed as the derivational source for question (12b), it can be seen that wh-movement of the DP sonata takes place from within the complex adjective that is formed by reanalysis: (14) The violin i is [ AP [ A easy to play wh-sonata] on t i ]. As Chomsky explains, it is thus anticipated that wh-movement will be barred in (14) (cf. (12b)), as extraction would have to take place out of a lexical constituent (ibid.:313). However, one problem with this particular explanation of the contrast illustrated in (12) is that it is predicated on the ill-defined assumption that certain linguistic elements are more easily incorporated in terms of tough lexical reanalysis than others. For example, the structural representation of the TC depicted in (13), above, requires that the PP headed by on resist incorporation with its licensing verb, and, in turn, resist being reanalysed into the complex tough adjective, as only in this way can the prepositional object remain eligible for legitimate wh-movement. Yet, Chomsky explains the comparative ungrammaticality of (15b), below, by proposing just the opposite: In this case, he argues that a closer relationship obtains between the embedded verb and PP than in (13), which allows incorporation of the prepositional object into the complex tough adjective (ibid.:312; examples below adapted from Chomsky s #143a&b, 1977:106): (15) a. the book is [easy to put t i on the table]. b.?*what table i is the book [easy to put on t i ]? The validity of this latter argument, and thus the validity of the reanalysis account of the TC, was challenged by, inter alios, Bach (1977:149-50) and Schacter (1981:434), who maintained that sentences such as (15b) are simply awkward rather than ungrammatical. Returning to the contrast noted in (12), Bach proposed an alternative explanation of the data, in which the lesser acceptability of (12b) could be explained
18 Chapter 2: The Syntax of the TC and Related Structures 87 in terms of processing considerations. On the basis of the contrast between the acceptability of a sentence such as (16a), below, (his #8a, ibid.), and the nonacceptability of (12b) (repeated below as (16b)), Bach argued that the problem more likely relates to the fact that a sentence such as (16b) involves crossed or non-nested dependencies (cf. Fodor 1978), which place exceptional demands on the memory resources that support sentence processing: (16) a. Which students k is this subject i easy to talk about t i with t k? b. *Which sonata k is this violin i easy to play t k on t i? In contrast, Jacobson (1992:294-5, 2000:10), who has addressed this issue more recently, claims that the problem has to do with the level of embedding of the element that is to be wh-moved from the tough infinitive, rather than the existence of crossed dependencies as proposed by Bach. Comparing a grammatical question such as (12a) with the ungrammaticality of (17) below, she notes that wh-extraction of a deeplyembedded prepositional object in (17) produces ungrammaticality similar to that
19 Chapter 2: The Syntax of the TC and Related Structures 88 observed in (12b), even though the processing of (17) does not involve any crossed dependencies (her #3, ibid.:9) 10 : (17) *Which violin i is that sonata k hard to imagine (anyone) playing t k on t i? Jacobson (2000) posits that there are actually two mechanisms for licensing an object gap in the case of the TC: In the case of long-distance extraction, the type of whmovement postulated by Chomsky (1977, 1981) is invoked, whereas in short-distance extraction function application is the appropriate mechanism, with the tough adjective c-selecting for a bare verb phrase complement. Jacobson was not the first theorist, however, to propose distinct mechanisms for the derivation of the TC according to whether short or long-distance extraction of the embedded object is involved. In a paper published in 1980, Nanni argued that the derivation of more complex TCs (i.e. those involving more than one embedded infinitive clause) is necessarily a non-lexical process, while the derivation of the standard TC (i.e. involving a single infinitive) more likely involves lexical restructuring. Nanni thus maintained that the standard TC is derivationally 10 Chomsky offers a similar example, which is reproduced in (i) below (his #iii, 1981:318-9, ftnt. 32, which he attributes to T. Stowell): (i) *Which oven i did the cake k take you all day to bake t k in t i? On the basis of the ungrammaticality of (i), which involves no crossed dependencies, Chomsky proposes that whatever problems arise as a consequence of crossed dependencies must arise independently of the problems associated with illicit wh-movement. Furthermore, he questions the cross-linguistic legitimacy of the type of processing constraint that is proposed by Bach (1977) above. This is because he observes that in languages such as Italian, in which wh-island violations are not attested, non-nested filler-gap dependencies such as that illustrated in (12b) are not considered problematic.
20 Chapter 2: The Syntax of the TC and Related Structures 89 ambiguous in Modern English, involving either creation of a complex adjectival predicate of the type proposed by Chomsky (1981) or syntactic movement. 11 Nanni s proposal was driven by the observation that the combination of tough adjective and single infinitive clause shares a different syntactic distribution than the combination of either tough adjective and for-pp, or tough adjective and multiple embedded clauses. In particular, Nanni argued that it is only in the former case that the combination of elements can be shown to behave as a single syntactic constituent. Space considerations prevent a full discussion of all of the supportive evidence offered in Nanni, and consequently I offer only selected examples of the contrasts she cites, which are listed in (18), below (adapted from Nanni s #3-8, ibid.:570-2). (NB: The (a) examples contrast wh-movement in the standard TC and in the long-distance version, the (b) examples contrast the two versions of the TC with regard to rightnode raising, and the (c) examples contrast the two versions in exclamative contexts.) (18) (a) i. How easy to tease is John? ii. *?How hard to convince Bill to marry was the girl? (b) (c) i. John certainly is and Mary may well be easy to tease. ii. *?John certainly is and Mary may well be hard to convince Bill to marry. i. How easy to tease John is! ii. *?How hard to convince Bill to marry the girl was! On the basis of the evidence reviewed in (18), Nanni proposed the operation of a complex adjective rule in English which can only take as its input the combination of tough adjective and single infinitive clause. I think this claim is intuitively attractive, even if I do not wholly agree with the theoretical analysis of the phenomenon offered 11 As Wilder (1991:118) notes, while Chomsky (1981) and Nanni (1980) both invoke the notion of lexical reanalysis of the tough adjective and complement, Chomsky never made the type of distinction that Nanni makes here between the derivation of standard versus long-distance versions of the TC. Thus, as Wilder explains, there is nothing in Chomsky s account to prohibit the formation of an improbable lexical item, such as [A difficult to convince anyone that they ought to read] book. As Wilder points out, this represents a serious weakness of Chomsky s analysis and one particularly appealing aspect of the account offered by Nanni.
21 Chapter 2: The Syntax of the TC and Related Structures 90 by Nanni. It is true, for example, that it is only the combination of tough adjective and single infinitive which is licensed to appear in syntactic positions that are normally reserved for adjectives in English (cf. an easy to clean bath and *an easy to ask Bill to clean bath) (ibid.:573). Moreover, as Mair (1987, 1990) has observed, native speakers of English sometimes use the combination of tough adjective and infinitive in coordination with a regular predicative adjective or adjectives, as illustrated in (19), below (Mair s #20, 1987:65): (19) There was an irony in the fact that the final pain killing bullet should come from a team like Stoke, a little long in the tooth, but wily and hard to put down. The empirical evidence cited above thus provides support for the claim that speakers allow lexical reanalysis in the case of the tough adjective and single infinitive. Nevertheless, as the foregoing discussion will have made clear, there is as yet no consensus as to the appropriate theoretical analysis of this option. Certainly, it has long been recognized that the reanalysis account proposed by Chomsky (1981), which I return to here, is flawed in certain respects. Perhaps the most serious problem with the account, as Chomsky himself acknowledged, has to do with the issue of how the subject of the TC comes to receive a θ-role that is more appropriately associated with an embedded object. Repeating example (11) as (20) below, the reader will recall that Chomsky initially proposed that the referential relationship that holds between the subject and object anaphor in (20) is similar to that which holds between an NP trace and its antecedent: (20) John i is [ AP [ A easy to please] e i ]. However, since he did not wish to entertain an account of the TC that involved direct NP movement, he suggested instead that the subject and anaphor in (20) are referentially linked according to what he termed the free indexing condition (1981:312). Yet the above account of the derivation of the TC raised certain problems of its own. For example, as Chomsky noted, while the projection principle holds that lexical
22 Chapter 2: The Syntax of the TC and Related Structures 91 insertion must take place at D-structure, with an argument merged in the position in which it is assigned a θ-role, the subject argument in a sentence such as (20) receives its θ-role only indirectly, specifically, as a consequence of reanalysis rather than through assignment by a licensing head. He maintained that the only logical conclusion that could be drawn was that insertion of the subject argument takes place at S- rather than D-structure, and that such an exception to the projection principle is warranted precisely on the basis of the grammatical status of sentences such as (20) (ibid.:313). Chomsky (1993) was in fact later forced to reject the rather radical modification of the reanalysis account outlined above. This is because, as noted by, inter alios, Lasnik and Uriagereka (1988), the subject of the TC can be a phrase that has a derivational history of its own (e.g. [The cat that ate the rat that ate the cheese] is difficult to describe). Such cases, the authors argued, could not be explained according to the proposed weakening of the projection principle, since the subject is a syntactically derived phrase rather than a lexical argument. (Notably, it is partly on the basis of this specific consideration that the representational levels of S- and D-structure have been abandoned in the Minimalist Program; see Chomsky 1993:20-1.) There is one further theory-internal issue that is also traditionally cited as proving problematic for the reanalysis account of the TC that Chomsky (1981) proposed and this concerns the behaviour of parasitic gaps (PGs) in such constructions. As was discussed in of Chapter 1, GB Theory standardly holds that the licensing of a PG is an S-structure phenomenon, requiring the presence of an operator in an embedded clause which A-bar binds both an antecedent trace and the PG, as illustrated in (21), below: (21) This article i is hard [ CP Op i [to file t i ]] [without reading PG i ]. Additionally, according to what is standardly termed the anti-c-command constraint (Chomsky 1982, 1986a:60-1), the antecedent trace must not c-command the PG or ungrammaticality will result. As Chomsky observed, both of these requirements are considered met if the S-structure representation of a TC reflects movement of an