Features, Categories and Checking

Towards Deriving the Epp and Abstract Case
  • Eric Haeberli
Part of the Studies in Natural Language and Linguistic Theory book series (SNLT, volume 54)


As discussed in chapter 1, the Extended Projection Principle (EPP) and the theory of abstract Case have played a central role for the analysis of the syntax of A-positions within the generative literature. However, both theoretical concepts are basically stipulations. Within the GB framework (cf. e.g. Chomsky 1981, 1986a), the EPP is a principle which states that every clause needs a subject and more precisely that [Spec, IP] has to be filled (S → NP INFL VP). Yet, it has never been entirely clear what the motivation for this requirement is. The main attempts to derive the EPP are based on the notion of predication (Williams 1980, Rothstein 1985). But to relate the EPP to predication is problematic. Given that predication is basically a semantic concept, it is unexpected that the EPP can be satisfied by semantically empty elements (expletives) which do not seem to be involved in a predication relationship. Rothstein (1985) therefore argues that the EPP is the result of a syntactic form of predication. Yet, such an extension of the semantic concept of predication is obviously again stipulative and the predication approach therefore is not able to derive the EPP in a satisfactory way.


Object Movement Categorial Feature Embed Clause Nominal Feature Case Theory 
These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.


Unable to display preview. Download preview PDF.

Unable to display preview. Download preview PDF.


  1. 1.
    The term object shift has generally been used for object movement in the Scandinavian languages. For a similiar type of movement in West Germanic languages like Dutch or German, the term scrambling has generally been used. However, I will follow much recent work (cf. e.g. Bobaljik 1995, Zwart 1997) in treating object movement past a VP-peripheral adjunct in these two language groups in a uniform way. To avoid terminological confusion, I will simply refer to the process shown in (1b, d, f) as object movement out of the VP in this chapter. This general term will not include topicalization of objects to [Spec, CP]. This movement is an A’-movement type and it therefore is not relevant for our purposes here. Cf. also chapter 3 for a more detailed discussion of object movement in the West Germanic languages (“scrambling”).Google Scholar
  2. 2.
    Support for the assumption that there may be a close relation between D and specificity comes from Creole languages. As Bickerton (1981:56ff.) observes, Creoles generally have determiner systems in which specific nominal constituents are marked by a (definite or indefinite) determiner whereas nonspecific nominal constituents do not have determiners. This contrast could be accounted for by saying that DP is present only with specific nominals but not with non-specific ones. If, as Bickerton (1981, 1984) suggests, certain aspects of Creole languages reflect fundamental properties of Universal Grammar, then specificity could be argued to be the unmarked property of D.Google Scholar
  3. 3.
    This intuition can of course also be found outside generative grammar. For example Langacker (1998:19; cited by Jackendoff 2002:124, fii. 12) says that he “personally findfs] it hard to imagine that fundamental and universal categories like noun and verb would not have a conceptual basis”.Google Scholar
  4. 4.
    But cf. e.g. Jackendoff (1977:31f.), Muysken and van Riemsdijk (1986:3f.) for problems that the attempt to define natural classes in terms of categorial features may raise.Google Scholar
  5. 5.
    For lexical categories, it could be argued though that the status of the categorial features need not be pre-specified but that it is determined through the structural context. Thus, in a way which is reminiscent of Marantz’s (1997) proposals, it could be argued that for example the nominal feature of a lexical category is only identified as uninterpretable once averbal functional element is introduced within the extended projection and that therefore the categorial status of a lexical head depends on the occurrence of specific functional elements in the structure.Google Scholar
  6. 6.
    Note however that the approach pursued here ultimately differs slightly from Grimshaw’s and van Riemsdijk’s. Grimshaw and van Riemsdijk propose that functional feature matrices consist of the two lexical features N and V and a single functional feature which they label F. As discussed above, here I will assume that there is a symmetry between the lexical and the functional level of feature matrices and that we therefore have two functional features, a nominal one and a verbal one. With respect to van Riemsdijk’s system, an additional difference should be pointed out here. Van Riemsdijk proposes that lexical categories are defined by a functional feature with a negative value (e.g. [+N,-V,-F]). However, given the analysis proposed here for the contrasts in (1), we have to assume that an NP does not have to check a functional feature and that N is therefore defined simply as [+N.-V]. Hence, I will assume here that the basic building block of a categorial feature matrix consists of the lexical features and that the functional features are added only if they contribute something to the content of the feature matrix at the interfaces, i.e. if at least one of them is interpretable.Google Scholar
  7. 7.
    Note that with respect to the data in (1e) and (If), the analysis proposed in the text implies that the nominal constituent prjár bœkur (‘three books’) can either be an NP or a DP. The only effect the DP-level has is to mark the constituent as specific. The presence of the DP-level then allows the object to move out of the TP. However, as illustrated also by (1a), movement of the object DP is optional. For (1a) or (1e) with a specific reading, I will assume that the object DP checks its T-feature non-overtly. Cf. section 5.2.10 for a more detailed discussion of non-overt object movement.Google Scholar
  8. 8.
    In Chomsky (1995), Case is considered as the trigger for object movement. In Chomsky (2000, 2001), however, the link between object movement and Case is less direct. Case allows an object to enter an Agree relation with v and, as a consequence of this relation, the object can be selected as the element satisfying the EPP-feature of v. Thus, the actual trigger of object movement is EPP on v, but Case on the object is a prerequisite for movement.Google Scholar
  9. 9.
    Pesetsky and Torrego derive their proposal from a domain of the syntax which differs considerably from the one discussed in this book. While my discussion will mainly deal with the inflectional domain, Pesetsky and Torrego focus on phenomena related to the C-domain. In section 6, I will briefly compare the two approaches in more detail.Google Scholar
  10. 10.
    For simplicity’s sake, I will represent feature hierarchies in the text by putting subfeatures in parentheses next to the feature they are associated with.Google Scholar
  11. 11.
    So far, I have only considered the role of the EPP in finite clauses. In sections 5.2.5 to 5.2.7, it will be shown that the approach pursued here can also be extended to infinitival clauses.Google Scholar
  12. 12.
    Cf. also Martin (1999) for the proposal that it may be desirable to unify the EPP and abstract Case. However, Martin’s analysis remains very tentative. A connection between the EPP and abstract Case is also suggested in Lasnik’s (1992) analysis of expletives.Google Scholar
  13. 13.
    But note that the claim here is not that all feature combinations are necessarily realized in every language. Thus, it may well be that some gaps remain for the languages considered here, even if aspects of the internal syntax of nominal constituents were taken into account.Google Scholar
  14. 14.
    An alternative for defining C would be to postulate an additional functional feature C. However, general theoretical considerations like simplicity or economy favor an option which reduces the inventory of categorial features to a minimum. Hence, the option explored in the text seems preferable since it makes use of a feature combination which is already available within a more restrictive system of features.Google Scholar
  15. 15.
    The idea would be that prototypical cases of a functional category have lexical categories whose values correspond to those of the functional features (cf. also the inflectional category of a matrix clause (finite T) or D in a simple nominal consituent).Google Scholar
  16. 16.
    The feature matrix [+D(+N), +T(+V)] will be relevant in particular for the analysis of the highest C-projection in embedded CPs. It is conceivable that different lexical feature specifications can be found with other types of C-heads (e.g. C in VI or V2 constructions), but since phenomena related to the CP-domain lie outside the scope of this book, I will not pursue this issue here. For standard cases of movement to CP (wh, negation, topicalization, focus), 1 will assume that they are generally not triggered by uninterpretable categorial features but by non-categorial features such as Wh, Neg, Top or Foe. This distinction could be a potential source for differences between checking within the A-system and the licensing configurations which are required within the A’-system (cf. e.g. Haegeman 1995, Rizzi 1996, 1997 on licensing within the A’-syntax (criteria) and, more particularly, Haegeman 1995:232f, Laenzlinger 1998:25 on differences with respect to the A-and the A’-system). It may also be that certain projections within the CP-layer, in particular TopP and FocP (cf. Rizzi 1997), are not specified in terms of the categorial features D, T, N and V, but that they are proxy categories of the type discussed for NegP, AspP or AgrP below.Google Scholar
  17. 17.
    However, as the discussion in section 5.2.3 will show, the syntax of PPs is a bit more complex than suggested here. In particular, it will be argued in section 5.2.3 that PPs are complex constituents consisting of two functional heads, the higher one being defined as([+D(+N), +T(+V)]), the lower one being defined as [+D(-N), +T(+V)]. Thus, PPs realize two of the four possible feature combinations that are available with the functional features [+D, +T] rather than just one. Cf. also section 6 for a similar possibility for C with a lower projection with an uninterpretable V-feature.Google Scholar
  18. 18.
    Cf. also Fraud (2001) for evidence from aphasia which supports the view that prepositions are functional elements. For our purposes, the traditional definition of P would be problematic for another reason. Given the proposal that negative values in feature matrices stand for uninterpretable features, [-N,-V] would be a category with a contentless feature matrix at the interfaces. It would be plausible to assume that categories must be defined in terms of something which can be interpreted at the interfaces, i.e. in terms of at least one interpretable feature. [-N,-V] would therefore not be a possible feature matrix for P.Google Scholar
  19. 19.
    As for “little V” (v) in Chomsky’s (1995) system, I will assume, as already pointed out in chapter, that it is simply an independent V head with its own [-N, +V] matrix.Google Scholar
  20. 20.
    What we may assume for the NP-complement of D is that its own features are unified with the lexical features of the DP and that as a consequence the uninterpretable V on N is checked in the same checking process.Google Scholar
  21. 21.
    A possibility suggested by Pesetsky and Torrego (2001) would be that the value of uninterpretable T is Nominative. This may be plausible for languages without a productive morphological case system such as English (cf. chapter 3.2.7). For languages with productive case morphology, however, I will argue in chapter 3 that the role of case goes beyond being a value of categoriel features.Google Scholar
  22. 22.
    This proposal is similar to a proposal made by Zwart (1993a:373, 1997:178f.) according to which licensing relations are sisterhood relations. As for the notion of dominance, I will use it here simply in terms of nodes, i.e. every node X which is connected downwards in the tree to node Y dominates Y (cf. also e.g. Haegeman 1994:85). The segment-category distinction will therefore not be adopted here. As for immediate dominance, I will assume that it stands for a dominance relationship in which no other node intervenes.Google Scholar
  23. 23.
    Since we assume that the crucial structural relation for identifying feature checkers is dominance, it is conceivable that locality could also be defined in a slightly different way than suggested by Chomsky. Instead of using c-command, we could simply say that the closest element is that for which the number of intervening structural nodes is the smallest. Thus, locality could simply be measured by counting the nodes intervening between the node bearing the attracting feature and the node bearing the potential checker.Google Scholar
  24. 24.
    Chomsky (2001:12) makes a distinction between weak and strong phases which I will leave aside here. The phases that are of importance here are strong ones.Google Scholar
  25. 25.
    This approach could also be adopted if we used Chomsky’s (2000, 2001) cyclic Spell-Out model discussed earlier. In Chomsky’s system, non-overt movement is generally not necessary because features can be checked in non-local configurations. However, as discussed in section 4.3, I will maintain here that feature checking occurs only in a local configuration (immediate dominance). Integrating this assumption into a cyclic Spell-Out model would also mean that certain elements move without carrying their phonological features along.Google Scholar
  26. 26.
    This assumption is reminiscent of Pesetsky’s (1989) Earliness principle. Cf. also Bobaljik (1995:350, fn. 6) for concluding that processes within a Single Output Model proceed in a way which corresponds basically to Pesetsky’s Earliness principle. Chomsky (2001:15) reaches the same conclusion for his cyclic Spell-Out model.Google Scholar
  27. 27.
    This is in contrast to Chomsky’s (1995, 2000, 2001) assumptions. In these frameworks, Merge preempts Move. However, the evidence for this claim is very scant. It mainly concerns expletive constructions, but, as I will argue in particular in chapter 5, expletive constructions can be analyzed without giving priority to Merge over Move. In addition, the “Merge preempts Move” hypothesis leads to the problem that Move nevertheless seems to take place sometimes even if there is an expletive present in a lexical array which could be merged. It is in this context that Chomsky introduces the notion of phase (cf. 2000:106). If Move precedes Merge however, the same type of problem does not arise and, at least for the analysis of the data discussed in this book, the concept of “phase” therefore does not have to be added to our inventory of theoretical devices. Finally, the theoretical motivation that Chomsky gives for “Merge preempts Move” (Merge being a more economical operation) is irrelevant in the system proposed here. As pointed out in the text, within the system proposed here, the Search space for an uninterpretable feature is the structure already built. Hence, if an adequate, structurally lower element is present in the syntactic structure, it is this element which is designated as a checker. Merge is simply not an option as long as Search has not established that no adequate element is contained within the structure already built, and economy considerations therefore do not play a role.Google Scholar
  28. 28.
    I am adopting this specific condition here rather than Chomsky’s (2000:123, 2001:4) more general proposals for the following two main reasons. First, Chomsky’s system seems to be too restrictive for the case considered here. As discussed below, I assume that fronted subject CPs bear a topic feature and that it is this topic feature which allows the CP to be fronted in examples like (11). It seems plausible to assume now that a feature which marks a CP as a topic is interpretable at LF. Hence, a fronted subject CP never bears any uninterpretable features. In terms of Chomsky’s system, an element has to carry uninterpretable features to be active, and it would therefore not be entirely clear why a subject CP can undergo movement. A second problem that Chomsky’s proposals raise for our purposes is the status of constituents with uninterpretable features such as subject DPs. In Chomsky’s system, such constituents become inactive once their uninterpretable features have been checked. What I will assume here, however, is that constituents with uninterpretable features can remain active even after checking (cf. e.g. (22k) below for an illustration). As for “defective intervention effects” which Chomsky derives from his proposals, they can be accounted for in alternative ways within the system developed here (cf. e.g. the analysis of (42)/(43) below accounting for Chomsky’s (2000:129) example (48b)).Google Scholar
  29. 29.
    As shown by Levin and Rappaport (1995:268) however, there are CPs that can be fronted independently of topicalization. Thus, wh-CPs can occur in a preverbal position in the context of subject-auxiliary inversion (ia), i.e. in a context where topicalization and hence preposing of that-CPs is impossible (ib). (i) a. ? Does why Paul came continue to surprise you? b. * Does that Paul came continue to surprise you? As pointed out to me by Liliane Haegeman and Ian Roberts (p.c.), it may be that (ia) involves a hidden relative (the reason why...) and hence a DP rather than a CP. But then the question arises, as an anonymous reviewer observes, why a that-CP as in (ib) does not also have the option of being realized as part of a DP with omitted nominal material (e.g. the fact that...). It may be that the distinct syntactic status of the CP within the DP (relative vs. complement) has an influence on whether the head noun can be omitted or not, but I will have to leave this issue as an open problem here for the moment.Google Scholar
  30. 30.
    A similar kind of analysis could also be argued to hold for example for participle agreement in the Romance languages (Luigi Rizzi, p.c.; cf. e.g. Kayne 1989 for discussion). Participle agreement in these languages is generally a side effect of another syntactic phenomenon (e.g. wh-movement or cliticization of the object) and it could therefore be argued that movement to a participle agreement projection depends on an independent, structurally higher attractor.Google Scholar
  31. 31.
    The simplest assumption to make here is that extraposition is rightward movement to a VP-peripheral position. In terms of a system which bans adjunction to the right (cf. Kayne 1994), we could assume that “extraposition” involves leftward movement of the CP and subsequent leftward movement of a projection containing the VP past the CP (cf. Haegeman 1998 for a proposal along these lines). However, it would lead us too far afield here to explore the consequences of these leftward movement operations within the system proposed here. For simplicity’s sake, I will therefore assume a traditional analysis of extraposition here, and I will leave the issues that arise within a framework without rightward movement for future research.Google Scholar
  32. 32.
    A problem that arises for Stowell’s framework (cf. Stowell 1981:167ff.) is the fact that infinitival object clauses do not behave exactly like finite object clauses. For example, extraposition of non-finite complement clauses does not seem to be obligatory, as shown in (i) (Stowell 1981:170). (i) a. John has promised repeatedly [to help us] a’. John has promised [to help us] repeatedly b. Frank wants very badly [to visit us] b.’ Frank wants [to visit us] very badly Non-finite object clauses can remain in a position which is adjacent to the verb when other subcategorized elements are present. Such structures are bad with finite clauses (cf. 15a’/b’). Within the framework proposed here, the data in (ia’/b’) raise the question how the N-feature on V can be checked given that the object clause is a CP (i.e. a category to which condition 14 would apply) and given that the object clause does not seem to have been attracted by a higher attractor. One possibility for dealing with this issue would be to say that C of a non-finite clause can incorporate into V. Thus, C is attracted to V for incorporation and, as a side effect, N on V can be checked. Cf. also fn. 35 below. If we assume furthermore that it is only V which can incorporate a C-head, the incorporation approach would immediately explain why infinitival CPs do behave like finite CPs when they are subjects. This is shown in the examples in (ii) which are parallel to those in (12) (from Stowell 1981:168). (ii) a. * John’s belief [that [to take this course] would help you] is unfounded b. * [Although [to take this course] would help you]... Non-finite subject clauses are banned from a preverbal position in contexts where topicalization is ruled out. Whereas this contrast between object and subject non-finite CPs is unexpected within Stowell’s system, the approach proposed here can distinguish the two cases in terms of incorporating heads. If we assume that T, contrary to V, cannot incorporate C, then an independent attractor has to trigger CP movement to allow feature checking in T. Hence, a non-finite subject CP can only precede the verb if topicalization is possible. One question that remains in terms of these proposals is why non-finite C would differ from finite C with respect to incorporation into V. A possible explanation for this could be based on the observation that whereas a finite clause is able to exist as an independent entity, non-finite clauses generally depend on the presence of some higher clause in their syntactic environment. It therefore could be argued that a finite clause does not interact with other (higher) clauses in the same way as a non-finite clause does and that the absence vs. presence of C-incorporation into V is a consequence of this difference.Google Scholar
  33. 33.
    Cf. Stowell (1981) for an analysis in terms of movement of the locative through the subject position in the functional domain. Cf. also Bresnan (1994) for a non-derivational analysis in the same spirit.Google Scholar
  34. 34.
    Note however that there seems to be one major exception to the claim that PPs cannot be attracted to [Spec, TP] (cf. e.g. Jaworska 1986). (i) a. [Under the chair] is a nice place for the cat to sleep b. Is [under the chair] a nice place for the cat to sleep? The PP under the chair can occur in a position immediately following the auxiliary in the context of auxiliary movement to C. In this respect the fronted PP in (i) differs from the fronted PP in the standard cases of locative inversion (cf. 19d). This suggests that PP movement in (i) simply targets [Spec, TP] and that no independent trigger (topicalization) for movement beyond TP is necessary for PP fronting here. Thus, the PP subject in (i) seems to behave like a nominal argument. As proposed by Bresnan (1994:110), the simplest analysis would therefore be to analyze the PP in (i) as a nominal constituent in which the nominal material is not realized overtly (e.g. the location under the table). Yet, an analysis along these lines is not unproblematic. The main problem is basically identical to that discussed at the end of fh. 29, i.e. what would have to be determined is exactly under what conditions material in the nominal domain can be omitted (cf. also Bresnan 1994:121).Google Scholar
  35. 35.
    An additional issue that arises here is the status of PP complements. Contrary to finite CP complements, PP complements seem to be able to remain in their base position. Consider the following contrast: (i) a. *? Mary said [that she wanted to drive ] quietly (=15a’) b. Mary talked [about it] quietly Example (ia) is discussed by Stowell (1981) as evidence for the claim that finite CPs cannot remain in the complement position of V. (ib) suggests that the same does not hold for complement PPs. At first sight, this conclusion is surprising. Due to their positive feature matrix, PPs cannot be attracted for categorial feature checking (cf. condition 14). Hence, although V selects its PP complement as the feature checker for its uninterpretable N-feature, the PP cannot check N on V in its base position. PP complements would therefore be expected to behave like finite CP complements in the sense that they should have to be attracted independently (for topicalization or extraposition) so that categorial feature checking can occur as a side effect of attraction to some other position. The analysis that I propose here is identical to that proposed already for non-finite CP complements (cf. fn. 32). More precisely, I argue that N-feature checking on V is the result of (possibly non-overt) Pincorporation into V. Due to incorporation, P is attracted independently and categorial feature checking then can occur as a side effect of incorporation. Cf. also Neeleman (1997) for the claim that the head of PP complements incorporates into V.Google Scholar
  36. 36.
    Locative inversion is found not only with PPs but also with locative adverbs (Cf. e.g. Here lives the world’s oldest man). So far I have not discussed the categorial status of adverbs. The conclusion from the fact that locative adverbs can satisfy the EPP would be that adverbs are partly defined in terms of functional features, and more particularly in terms of a D-feature. One possibility would be that adverbs are also of the category [+D(+N), +T(+V)]. This assumption might get some support from the fact that certain words can be used both as an adverb or as a preposition (like before in English). As for the relationship between adverbs and adjectives which are generally defined as [+N, +V], it could be obtained for example by assuming that the adverbial [+D(+N), +T(+V)] head takes a [+N, +V] complement. Hence, a typical adverbial morpheme like-ly in English could be analyzed as the [+D(+N), +T(+V)] head (cf. nice — nicely). However, since the details of the analysis of adverbs is not crucial for our purposes, 1 will not pursue this issue here.Google Scholar
  37. 37.
    Cf. also e.g. Stowell (1981:269) for the assumption that the subject in locative inversion constructions has undergone movement. Note that the claim here is not that overt movement to a focus position is the only option for focusing a constituent. The alternative is to assign stress to a particular element in its normal syntactic position as for example in JOHN walked into the room (not Peter). In this case, the subject DP John is prosodically marked as focus once it has been raised to [Spec, TP]. Thus, we can distinguish a syntactic type of focus assignment and a purely prosodie one. It is only the former which interferes with syntactic derivations in the way described below.Google Scholar
  38. 38.
    There is one issue that the analysis in the text does not capture, namely the way in which the nominal argument checks its verbal features (T/V). Suppose that, as suggested in the text, the nominal argument occupies [Spec, VP] and the locative PP is the complement of V. Hence, V selects the PP as the checker for its N-feature once V and the locative are merged. The nominal argument therefore does not enter a checking relation with this V. Furthermore, T is excluded as a checker, too, because, as argued in the text, T selects and attracts the PP as its feature checker. Given this problem, it could be proposed that the structural representation of arguments in locative constructions is slightly different from what I proposed in the text. More precisely, we could follow the basic idea of Hoekstra and Mulder (1990:29) according to whom the relation between the PP and the DP is a predicative relation. These authors argue that the DP and the PP form a small clause where the DP is the argument of the PP predicate as represented in the following structure. Thus, we get the following structure: (i) [vp V [sc DP PP] It would lead us too far afield here to consider the exact categorial status of small clauses within the framework proposed here, but if we assume that a structure along the lines of (i) can be maintained, we would get the result that the DP can be selected as a categorial feature checker by V and that it therefore can satisfy its categorial feature checking requirements without having to undergo overt movement to TP (cf. section 5.2.10 for non-overt movement to TP).Google Scholar
  39. 39.
    Thanks to Liliane Haegeman for pointing this out to me. Nominal adjuncts could be analyzed as elements containing an empty preposition.Google Scholar
  40. 40.
    To be precise, note that 1 will assume that, if an element X occurs in a position which is not a categorial feature checking position, it is actually the entire X which is inaccessible for the categorial Search process. This is to exclude the possibility that for example a DP-complement of a PP (i.e. a DP in a categorial checking position) contained within a postverbal DP with locative inversion could be attracted by T.Google Scholar
  41. 41.
    The GB and Minimalist analyses presented in the following subsections are mainly those discussed in Chomsky’s work. In the literature, various alternatives and developments of Case Theory can be found (cf. e.g. Bittner and Hale 1996a, b, de Hoop 1992, Torrego 1998). The problem these approaches raise is the same one as that discussed for the standard GB/Minimalist analyses in section 1, i.e. they are based on the concept of abstract Case. I will therefore not discuss these developments since the ways in which they differ from Chomsky’s analyses is not central to our concerns here.Google Scholar
  42. 42.
    For simplicity’s sake, 1 do not distinguish different VP-shells for the two arguments here (cf. chapter With respect to the derivation of (22a) the presence of an additional VP would not make any difference because the object would then simply be selected as a feature checker by two V-heads rather than one. But the subject still would require a categorial feature checker from outside the VP.Google Scholar
  43. 43.
    Given the proposal made in section 2 that nominal arguments can either be DPs or NPs, the question might arise at this point what would happen if the subject was an NP. Since an NP cannot check the Dfeature of T, the only possibility to save a derivation with a subject NP would be to insert an expletive as in constructions like There arrived a man (cf. also Chomsky 1995, Frampton 1995). This type of construction will be discussed in more detail in chapter 5. Note finally that in a sentence like A man arrived the subject would have to be analyzed as a DP, suggesting that the content of D may not necessarily be [+specific] in English (cf. also the discussion in section 2). But since the indefinite determiner a occurs both in constructions in which the subject is fronted and in expletive constructions, the conclusion for the status of a would have to be that it occurs in a position below D (i.e. on a proxy head between D and the NP or NP-adjoined).Google Scholar
  44. 44.
    As observed by an anonymous reviewer, the analysis of P proposed here is reminiscent of the use of top and bottom features in Tree Adjoining Grammar (cf. Vijay-Shanker 1987 and subsequent work in TAG). The structure in (24) is also very similar to that proposed by Starke (1993:41) in that it contains two C-like heads (i.e. two [+D, +T]-heads in our terms). However, Starke also argues that between these two heads a lexical category occurs which he identifies as A (adjectival/adverbial, 1993:42). Such an assumption would not be incompatible with the proposals made here. On the contrary, its presence may even have a desirable consequence in the sense that it could account for what looks like a lexical (i.e. non-functional) property of at least some prepositions, i.e. the ability to assign thematic roles. However, for the purposes of “Case” checking this head would not play a role and I will therefore leave it aside here. As for the additional heads that Koopman (1997) proposes within the prepositional domain, their status would presumably mainly be of the proxy-type as discussed in section 4.1.Google Scholar
  45. 45.
    As for the exact structure of examples like (25), it is important to note first that the prepositions like auf or unter in (25) seem to be the Case assigners (cf. van Riemsdijk 1990:236f). In terms of the system adopted here this would mean that they have to occupy the [-N] head in the structure in (24) and the postpositions would have to be the heads of the higher P projection. Thus, the examples in (25) can be analyzed in two ways. First, as proposed by van Riemsdijk, we could assume that the lower projection is head-initial and the higher one is head-final. Alternatively, in terms of a purely head-initial structure (cf. Kayne 1994), the lower projection would have to move to the specifier of the higher one. However, it would not be entirely clear what the trigger for this movement would be.Google Scholar
  46. 46.
    An anomyous reviewer observes that the point for because of and instead ofmay be weakened by the fact that historically these elements are of the form P-N-P (e.g. by cause of) and that therefore they may simply retain the syntactic properties of their origin. However, it could be argued that these properties were retained precisely because P is a complex category. Once because or instead were reanalyzed as individual lexical items (which they presumably are in Modern English, rather than combinations of a P and an N), there still was a structural position available for of in their prepositional uses and of could therefore be retained. Prepositions like onto or into could also be argued to be comparable to (25). However, for Modern English, it seems difficult to prove that these elements provide evidence for a complex PP structure since it could be argued that they are independent lexical items requiring only a single syntactic head. Yet, the fact that items like into and onto, i.e. combinations of two prepositions, emerged in the history of English could be considered as additional evidence for the availability of more than one P-head within a PP.Google Scholar
  47. 47.
    In recent Minimalist work, it is proposed that Agree relations and hence feature deletion are only possible with a φ-complete probe (Chomsky 2001:6). The claim could therefore be that A and N cannot enter an Agree relation with a nominal complement because they are not φ-complete. However, the questions arising then are (i) why feature deletion is restricted to relations involving a φ-complete probe, and (ii) what determines whether an element is φ-complete or not (e.g. why are A and N not φ-complete whereas P seems to be φ-complete since it licenses nominal complements?).Google Scholar
  48. 48.
    Nominal constituents productively occur as complements of adjectives for example in a language like German. As suggested by van Riemsdijk (1983), there seems to be a correlation between this property and the occurrence of morphological case. In chapter 3, I will argue that this correlation can be explained within the framework proposed here.Google Scholar
  49. 49.
    Apart from the phenomenon in (28b), the internal syntax of DPs/NPs raises several additional issues within the framework proposed here which 1 cannot address in detail here. Let me just briefly speculate on two main points. First, in connection with (28b), one may wonder what the status of possessives is as in John’s description of Steve. It seems clear that the possessive also cannot be licensed by any verbal features DP-internally. One possibility would therefore be to assume that possessives are similar to the complement in (28b) in the sense that they are also licensed by a dummy [+D(-N), +T(+V)] licenser. In this case however, licensing would be related to the N-feature on the functional D-head rather than to N and this difference could then be argued to be the source of the different spell-out form (’s instead of of, cf. also Chomsky 1986:194 for a similar proposal in the context of inherent Case assignment). Another more general issue that arises with respect to the NP/DP-syntax concerns the status of the verbal features of NPs and DPs. In order to rule out the occurrence of NPs/DPs in certain positions in the clausal structure, it is crucial for the approach adopted here that NPs/DPs cannot check their verbal features through NP/DP-internal elements bearing verbal features such as APs ([+N, +V]), PPs ([+D(+N), +T(+V)]) or CPs ([+D(+N), +T(+V)]). If such NP/DP-internal checking processes were possible, we might expect certain nominal constituents to be legitimate for example in the complement position of an adjective. Such a conclusion has to be avoided and the question is how this could be done. There are basically two options here. First of all, at least certain DP/NP-internal elements could be argued to occupy positions which are not categorial feature checking positions. Thus, as in the case of clausal adjuncts (cf. section 5.1.2), such elements occur in specifier positions of proxy categories and they therefore do not qualify as categorial feature checkers (cf. Alexiadou 1997, Cinque 1999 for the parallelism between clausal and nominal adjuncts). There may be an additional reason though why DP/NP-internal elements do not check verbal features of the DP/NP. The typical DP/NP-internal constituents mentioned above (AP, PP, CP) all have entirely positive feature matrices. Thus, given Attraction Resistance (cf. 14), they can be selected as feature checkers but they cannot get attracted. If, as suggested in section 5.1.1, this means that the feature checking process is always interrupted as soon as Search has identified a positively specified feature matrix as a checker, we would obtain the result that APs, PPs or CPs do not check verbal features on N/D even when they occur in a potential categorial feature checking position. As for the dummy licenser of which I have analysed as a [+D(-N), +T(+V)] head and which therefore is not defined as an entirely positive feature matrix, we could assume that it is created through the interaction of N and its complement and that it therefore cannot be taken into consideration for checking of N’s own features.Google Scholar
  50. 50.
    As for potential intervening proxy categories between T and V (cf. section 4.1), they could be argued to be able to transmit the categorial features of V since they are directly related to V (V being the head on which the features of proxy categories are generated).Google Scholar
  51. 51.
    This analysis has two immediate consequences. First, constructions like (31) involve movement of the embedded subject into the matrix domain as originally proposed by Postal (1974). If we assume that the structure of transitive verbs consists of two VPs (cf. chapter, the derivation looks as follows. First, the subject moves to the [Spec, VP] of the lower VP. Then the higher V (i.e. the V which is responsible for external theta role assignment) is merged. This head also has an uninterpretable N-feature and, given the structure available so far, it selects the subject of the embedded clause for N-feature checking. Thus, the embedded subject moves to [Spec, VP] of the higher V. Then the external argument is merged in the outer Spec of the higher VP. The second consequence of this analysis is then that, since the matrix verb precedes the subject of the embedded clause, we have to assume that main verbs in English undergo short movement (cf. also e.g. Johnson 1991, Koizumi 1993).Google Scholar
  52. 52.
    This idea would be similar to the proposal made by Rizzi(1986) in the context of pro, i.e. that a null category has to be licensed and its content recovered in a formal configuration. The licensing property is related to categorial feature checking with C. As for recovering the content of PRO, we could say for example that non-ftnite C has the capacity to assign an arbitrary interpretation, but that this arbitrary intrepretation can be suppressed by control relations.Google Scholar
  53. 53.
    Some additional observations can be made in this context. Given that Acc-ing constructions have clausal properties, I assume that the TP defined as [-D(+N), +T(+V)] is generally dominated by a CP. The C-head of this projection can then be argued to have (at least optionally) the properties of C in other non-finite contexts. PRO can therefore be licensed in-ing clauses (cf. e.g. PRO leaving now would be a mistake). A second observation concerns the status of-ing more generally. The assumption here is that-ing can realize several distinct functional heads. The functional head involved in poss-ing constructions (e.g. John’s leaving his wife was unexpected) can be assumed to be more D-like because it licenses Genitive subjects. The minimal assumption to make for this functional head isthat it differs from Acc-ing simply with respect to its functional features and that it is therefore defined as [+D(+N),-T(+V)], a feature combination we have not used in any other context yet. Finally, with respect to-ing in the progressive (He was reading a book), we could assume that it shares its feature specification with past participles, which, as I will suggest in the next subsection, are defined as [-D(+N), +T(-V)]. Finally, it should be pointed out that, apart from their ability to license overt subjects, Acc-ing constructions have several additional interesting syntactic properties (cf. Reuland 1983). However, it would go beyond the scope of this subsection to explore the issues these properties raise for the framework proposed here.Google Scholar
  54. 54.
    Note that this analysis depends on the observation that Move precedes Merge within the system adopted here (cf. section 4.4 fn. 27). But cf. chapter 5 for a more detailed discussion of the syntax of expletives.Google Scholar
  55. 55.
    As often pointed out, it in (40) may not be a pure expletive but might rather have the status of an argument which is somehow associated with the CP (cf. e.g. Bennis 1986, Vikner 1995). Hence, the option for satisfying the EPP shown in (40) might actually consist of insertion of it in the argument position, subsequent movement of it to TP and insertion of the CP in a non-argumental position. This option is only legitimate with CPs because, due to the lack of uninterpretable categorial features, CPs can occur in positions which remain unaffected by categorial feature checking. DPs however have to be involved in the categorial feature checking process so that their verbal features can be checked and they can only do that by occurring in argumentai positions (cf. section 5.1.2).Google Scholar
  56. 56.
    Cf. chapter 3.3 for a slight refinement of this proposal. As discussed below, the restriction discussed in the text may account for what has been referred to as Holmberg’s generalization. In addition, this restriction ensures that D cannot attract an unlimited number of arguments. If D had this capacity, it would also be able to attract for example DP complements of adjectives, thereby licensing such complements productively. However, a language like Dutch licenses object DP movement out of the VP but DP complements of adjectives are not licensed productively (cf. chapter 3). In terms of the restriction formulated in the text, this contrast follows because it is only with object movement out of the VP that a secondary attractor is available under T.Google Scholar
  57. 57.
    1 argued in section 5.2.6 that in ECM contexts an N on V can attract a DP with an unchecked T-feature. There is one major difference between ECM and the context discussed here. In ECM contexts, the DP has already been selected for feature checking with a T. This ensures that the DP can check its T-feature on its way to the attracting V and the requirement proposed in section 5.9 that a functional feature has to be checked before its subfeature(s) is therefore met.Google Scholar
  58. 58.
    Note that V-movement cannot be triggered by categorial feature checking within the framework proposed here since V2 does not have an uninterpretable categorial feature which could be checked by V1 or similarly T does not have an uninterpretable categorial feature which could be checked by the complex V-head. I will therefore assume that V-movement is triggered by an independent feature. However, I will have to leave a more detailed investigation concerning the status of this feature for future research. As for the timing of head movement, 1 will assume that it occurs right after insertion of the attracting head, i.e. before the head projects further in order to create specifier positions. Cf. also fh. 60 below for some additional observations concerning head movement.Google Scholar
  59. 59.
    For N on V1/V2 to be attractors in T, I assume that features in complex heads can percolate within a projection. Thus, N on V1/V2 can dominate the object DP from TP and, thus, they are able to attract and are checked (cf. the definition of attraction in terms of dominance and feature checking in terms of immediate dominance in (7) above)Google Scholar
  60. 60.
    The assumption would have to be then that in those cases in which object movement is not possible because the verb has not moved overtly, the verb also does not move non-overtly. Otherwise, reactivation of D on T could be triggered through non-overt movement. Thus, we could follow Bobaljik (1995) and Bobaljik and Thráinsson (1998) in assuming that in languages without overt verb movement the relevant checking requirements can be satisfied by V in its base position. However, as pointed out above already, I will have to leave a more detailed investigation of the status of V-movement for future research.Google Scholar
  61. 61.
    The way the parameter in (47) is formulated implies that multiple attraction is optional if the parameter is set positively (i.e. a feature can (but does not have to) act as a multiple attractor). This reflects the situation found in Icelandic with respect to object movement since this movement seems to be possible but not obligatory (cf. examples 44/45). However, there would be an alternative, more restrictive way of formulating (47), i.e. as a parameter determining simply whether a head does or does not act as a multiple attractor. In terms of such a parameter, cases of optionality would have to be analyzed as cases where a head can realize both parameter settings. The assumption that two parameter settings may be available in a single language would be reminiscent of the idea of‘competing grammars’ developed by Kroch (1989) and in much subsequent work. However, in Kroch’s work, the availability of two parameter settings generally characterizes periods of change in which one parameter setting replaces another one. Whether such an assumption would be appropriate for the case of object movement discussed in the text has to be left open here, and I will therefore continue using the parameter in (47). However, nothing in the remaining discussion hinges on this choice.Google Scholar
  62. 62.
    One issue that is not addressed in the text is the status of N-feature checking on V and V-feature checking on the DP. For these processes, there are two possible scenarios. If the verb moves out of the VP as for example in (48), then N/V-feature checking can also be done in TP. However, if the verb does not move, then we could argue that the DP checks its V-feature and the VP’s N-feature VP-internally once the D-head in T has checked the primary categoriel feature (T). Thus, the idea would be that in the second context categorial feature checking is distributed over two positions (i.e. the head of the chain and the spell-out position). Both of these options may also occur with pronouns. Thus, in (48), all the feature checking takes place in TP. However in constructions involving non-finite verb forms where the verb does not seem to move (e.g. SU-Aux-V-DO(pronoun)) and where therefore it has generally been assumed that the pronoun does not move, either, the pronoun would also check its categorial feature in two positions (TP for T, and VP for V). This contrast could then be argued to account for the overt/non-overt distinction with pronouns. 1f all categorial feature checking takes place in TP (i.e. if the main verb undergoes movement, as in 48), the pronoun moves. If only the T-feature is checked in TP and the V-feature is checked in VP, then the pronoun does not move. This proposal could then be extended to account for the properties of object pronouns in English. Since a main verb never moves in English, pronominal objects do not move, either.Google Scholar
  63. 63.
    Thanks to Luigi Rizzi for suggesting an approach along these lines. Note that the basic idea mentioned in the text is similar to some extent to a proposal made by Cardinaletti and Starke (1994) for pronouns. In their system, certain elements have to move in order to compensate for structural deficiency. Here, the movement would be triggered by a deficiency in the feature matrix of a category.Google Scholar
  64. 64.
    However, the extension to the CP-domain discussed in section 6 would be problematic in terms of this system because uninterpretable features were assumed to be able to check other uninterpretable feature at least under certain circumstances. This option would not be available here since the deficiency caused by a negatively specified feature could not be compensated for by another negatively specified feature.Google Scholar
  65. 65.
    Given these independent motivations, categorial distinctions play a role even in models of Universal Grammar such as Jackendoffs (2002) in which UG has a very limited degree of prespecification (cf. e.g. Jackendoff 2002:125, 192,259). Note also that the postulation of categorial distinctions, which was originally based on purely linguistic observations, can be argued to be supported by studies of brain function (cf. e.g. Berndt et al. 1997, Caramazza and Hillis 1991, Damasio and Tranel 1993 on the N/V-distinction or Calvin and Bickerton 2000:152f. on the [+/-N, +/-V] feature system).Google Scholar

Copyright information

© Springer Science+Business Media Dordrecht 2002

Authors and Affiliations

  • Eric Haeberli
    • 1
    • 2
  1. 1.University of ReadingUK
  2. 2.University of GenevaSwitzerland

Personalised recommendations