(h) (if) Mike had have played, then . . . (If Mike would have played . . . ) (i) Mike had been beaten. (j) Mike had been playing bridge. (k) Mike must have played bridge. (l) TP Mike TЈ T must vP v have vP v been vP v being VP V beaten The origins of the little v approach lie in the proper treatment of double-object constructions like (33a). X-bar theory allows only one complement position in the tree, so the two obligatory complement NPs in (33a) are mysterious. (33) (a) John gave Mary a book. (b) John gave a book to Mary. Further, Barss and Lasnik (1986) show that in goal–theme orders (33a), goals c-command themes (34a, a’), and in theme–goal orders (33b), themes c-command goals (33b, b’): (34) (a) Mary showed John himself (in the mirror). (a’) *Mary showed himself John (in the mirror). (b) Mary showed John to himself. (b’) ??Mary showed himself to John. This means that the leftmost argument is higher in the tree than the one on the right. Larson (1988) argues that a split-VP approach solves this problem. There are two verbs give, one with two NPs and the other with an NP and a PP. With the addition of the vP to host the agent, we have a place for both the theme and the goal such that the Wrst c-commands the second. Surface order is derived by movement of the V into the v. 240 controversies () (a) vP (b) vP NP agent vЈ NP agent vЈ vVP v VP NP goal VЈ NP theme VЈ VNP theme V PP goal Having a split-VP structure then accounts for a wide variety of data, while capturing both the locality of the VPISH and the distance of the traditional VP. Other little v-like categories have been proposed in the literature, including ApplP which, is tied to applicative constructions (McGinnis 2001; Pylkka ¨ nen 2001, 2002). See also Harley and Noyer (1998) for a discussion of the role of little v in nominalizations. There is one further aspect of the split VP that remains to be mentioned. Travis (forthcoming)26 argues using morphological evi- dence from a variety of Austronesian languages and syntactic evidence from languages with object shift that there is a functional category that lies between the vP and VP which represents situation aspect (roughly, aktionsart). (36)[ vP v[ AspP Asp [ VP .V .]]] This projection also serves to case-mark objects. There is some seman- tic motivation behind this projection: certain kinds of objects lead to various situational aspect distinctions. For example, if you ate an apple, the eating event ends when the apple is Wnished, by contrast, if you just eat, the eating event has no clear end point. An object- licensing category between vP and VP, in the guise of AgrO (object agreement) is also found in Koizumi (1994, 1995). Carnie (1995) and Carnie and Harley (1997) argue that you need both Aspect and AgrO to account for the order of elements in Irish and Scots Gaelic non-Wnite clauses; see also Noonan (1992, 1993, 1994) and Adger (1996). The claim that Aspect is tied to the VP (or the domain of predication) is also consistent with (under very diVerent assumptions) the views of Dik (1989) and Role and Reference Grammar. 26 See also Ramchand (1993) and MacDonald (2006). phrasal categories and cartography 241 11.4 The clausal layer Much of the discussion in the following two sections concerns what is known in the P&P framework as ‘‘functional categories’’. The literature on functional categories is vast; so vast that I cannot hope to even expose the tip of the iceberg. I give here some of the leading questions and some of the matter that has the most consensus behind it, but the reader should not consider these sections even remotely authoritative or complete about the topic. In this section, we consider the nature of the clause, that is, the structure traditionally labeled S in early gen- erative grammar (and is still labeled as such in many approaches, including LFG, GPSG, and HPSG). There are three major classes of treatments for the clausal layer. There are those views that have S as an unheaded category, whose function is simply to license the predication between the subject and the VP. This was the view of early transformational grammar and survives to a lesser degree in some versions LFG but only in clauses where there is no auxiliary to head an IP (InXectional Phrase).27 Far more approaches adopt one of the other two approaches. In HPSG and GPSG, the clause is a projection of the predicate of the clause, usually the V. In GPSG this is because InXection is usually part of the verb, and since the theory is meant to be ‘‘surface-true’’ it cannot express a more abstract category like InX. In HPSG, the S category is simply the feature structure that has all of its argument features fully resolved. A sign- iWcant portion of the literature in the Principles and Parameters tradition (both GB and MP) focuses on the third treatment: the clausal layer is made up by one or more projections of functional categories, which are headed by grammatical properties. Abney (1987) and Grim- shaw (1992) suggest hybrids of the second and third approaches. They have functional categories, but these are viewed as extended projec- tions of the V. See Bury (2003) for a more up-to-date version of this claim. For the most part, we will focus on the functional category approach and make reference to the others as necessary. I will largely assume that the whole premise of functional categories (or something like them, such as the operators in RRG) is well motivated; see Hudson (1996) for a criticisms of this assumption and of the entire functional-projection 27 This is part of an larger programmatic restriction of ‘‘economy of expression’’, where forms appear only in the c-structure of the sentence if they have overt expression as words. See Falk (1983) for more discussion. 242 controversies endeavor. I will also assume that functional categories and their relative order to each other is universal. This too is controversial. For example, Fukui (1995) claims that Japanese lacks functional categories, which explains why wh-phrases in the language are in situ. Lebeaux (1996) claims that functional categories are only present if they are required for licensing. Ouhalla (1991, 1994) argues that SVO languages diVer from VSO languages in precisely the order in which the TP and AgrSP functional projections are found. He claims that VSO languages have TP over AgrSP (which licenses nominative case) and SVO languages have AgrSP dominating TP. These claims aside, a universal hierarchy of functional categories—if it can be found—is to be preferred on economy grounds. Within transformational grammar and related approaches in the 1960s and 1970s (see for example Chomsky 1965), the S category was unheaded, but often included an AUX category to host auxiliaries. This category was thought to contain two elements, tense and agreement. With the widespread adoption of X-bar theory for other categories, the existence of an unheaded S category seemed suspicious. JackendoV (1977) proposed that S was a verbal projection (V’’’ or a higher bar level). This view was (and is) still widely adopted in the GPSG and HPSG approaches to syntax. ()VЈЈЈ=S NP VЈЈ VЈ V Auxiliaries and temporal adverbs were in the speciWer of V’’. Emonds (1978) proposed instead that there is a category InX (for inXection) that hosts auxiliaries, and, in their absence, the features associated with tense and inXection (37). Ken Hale suggested that InX was the head of the S category in class lectures and an unpublished paper in 1979. Stowell (1981) proposes that this category be assimilated to the X-bar approach giving us InX,I’, and IP. Pesetsky (1982), Huang (1982), and Falk (1983) all pursue and argue for this kind of approach. When an auxiliary appeared, it occupied this head (38a). When the inXection appeared on the verb, the InX category lowered to the V (38b): phrasal categories and cartography 243 () (a) IP (b) IP NP IЈ NP IЈ Infl was VP Infl -ed VP VV Koopman (1984) (see also Emonds 1978) argued that variation in the position of the verb relative to various elements, such as negation or adverbs between languages like French and English (and other lan- guages including Vata) could be attributed to whether the moved to InX (41a) or the InX lowered to the V (41b) (for the LFG equivalent to this—head mobility—see Kroeger 1993): (39) (a) I often eat apples. Adv V Obj (b) Je mange souvent des pommes. V Adv Obj I eat often of.the apples ‘‘I often eat apples.’’ (40) (a) I do not eat apples. not V object (b) Je ne mange pas des pommes. V not object I neg eat not of.the apples ‘‘I do not eat apples.’’ () (a) TP (b) TP NP TЈ NP TЈ Je T [pres] VP I T [pres] VP VЈ VЈ AdvP VЈ AdvP VЈ souvent V NP often VNP mange …… eat With the addition of the VP internal subject hypothesis V ! InX movement can also explain VSO order (see section 11.4). Positing an InX head, even in contexts where there is no overt auxiliary, thus provides a mechanism for explaining variation in word order across languages. 244 controversies Pollock (1989) argues that a single functional category like IP is not enough to solve various problems of word order. He suggests splitting InX into the two subcomponents of the old Aux category T and Agr. The evidence comes from French non-Wnite clauses. Consider Wrst a Wnite clause with both negation and an adverb (42a), where the adverb is adjoined to the projection of the verb, and pas ‘‘not’’ is in some functional category lower than InX. Next consider the inWnite in (43a) the verb appears in some position between the negation and the adverb. There is no obvious head position in (42b) for the verb to land in. (42) (a) Je n’ai pas souvent arrive ´ en retard. I neg-have not often arrived in late ‘‘I have not often arrived late.’’ (b) [ IP InX [ NegP not [ VP [ V’ often [ V’ V ]]]]] (43) Ne pas arriver souvent en retard c’est triste. neg not arrive-inf often in late it’s sad ‘‘It’s sad not to arrive late often’’. To account for this, Pollock proposes that the InX category is more correctly identiWed as T (explaining why the verb does not move to it in tenseless non-Wnite clauses), and that the position the verb lands in in (43)isAgr(44). (44)[ TP T[ NegP not [ AgrP Agr [ VP [ V’ often [ V’ V ]]]]]]. Chomsky (1991, 1993) extends this approach. Considering the nature of the category Agr, he proposes that this category is fundamentally part of the mechanism for assigning Case (for a contrasting view see Massam 1994). The intuition behind this approach is that both case and agreement are mechanisms for expressing grammatical relations. As such he proposes that there are really two Agr categories,28 one for assigning nominative case (AgrS) and one for accusative case (AgrO). These straddle the TP: (45)[ AgrSP Nominative AgrS [ TP T[ AgrOP Accusative AgrO [ VP . Case assignment is not solely the domain of AgrPs. Since nominative case is often tied to tense, T and AgrS jointly assign the case when T has head-moved to AgrS. Since accusative case is typically tied to the 28 See Belletti (2001) and Rizzi (2004) for illuminating surveys of the literature on AgrPs. phrasal categories and cartography 245 argument structure of the verb, AgrO serves as a host for case assign- ment only when the V has head moved into it. Obviously, this architecture is challenged by the wide variety of word orders expressed in language. However, Chomsky (1993) provides a unique explanation for word-order variation given this clausal structure. He suggests that all movement of the subject to the speciWer of AgrS, the object to the speciWer of AgrO and the movement of the V head through all of the head positions are universal. However, the universality of the movement holds only at the abstract level of logical form, where a condition requiring that all features be checked must hold (the principle of full interpretation).29 Cross-linguistic variation follows from variation in timing of these operations relative to a rule of Spell-out which creates the PF (or phenogrammatical structure) of the sentence. The position- ing of elements in the surface string is due to movements that occur overtly, or before Spell-out. Those that are not seen in the language occur covertly, or after Spellout. For example, since English verbs and objects appear to be low in the tree movement to these positions is covert. As strange as the idea of a ‘‘covert’’ movement operation might be, this provides a very constrained model of acquisition: children need only learn the timings of operations to learn the grammar of their language. It also predicts a limited number of permutations of order, which should correspond to the overt expression of the tense and agreement morphology of the language. Let us consider some of the evidence that has been proposed in this line of argumentation. Bobaljik and Jonas (1996) argue that three argument positions are needed to account for each of the arguments in the transitive expletive construction: (46) Það hafa margir jo ´ lasveinar borðað bu ´ ðing. there have many christmas.trolls eaten pudding ‘‘Many Christmas trolls have eaten pudding.’’ Evidence from languages with object-positioning alternations also points to the existence of an AgrO position below TP. As a typical example, consider the facts in (46) from Irish (See Bobaljik and Carnie 1996 among many other sources): 29 The ideas here Wnd correlates in other approaches. The Principle of Full Interpret- ation is similar to the requirement in LFG and GSPG that the structure represents a uniWcation of features, and the requirement in HPSG that all features are resolved. The idea of an abstract LF is quite tightly related to the notion of a tectogrammatical structure (see Ch. 10). 246 controversies (47) (a) Ta ´ Sea ´ n ag scrı ´ obh na habairte. VO be John prog write the sentence.gen ‘‘I want John to write the sentence.’’ (b) Ba mhaith liom [Sea ´ n an abairt a L scrı ´ obh].OV c good with.1.s John.acc the sentence.acc prt write ‘‘I want John to write the sentence.’’ In some circumstances (in the progressive aspect, or in non-Wnite clauses in the literary Munster dialect), the object follows non-Wnite verbs and takes an inherent genitive case (47a); in other circumstances (such as non-Wnite clauses in Northern dialects, or in other periphras- tic aspects such as the perfective) the object precedes the non-Wnite verb and takes accusative case. In such cases, a particle homophonous with the third-person possessive pronoun appears between the object and the verb. Bobaljik and Carnie (1996) identify this pronoun as an overt instantiation of AgrO. In Irish, when an overt subject is present verbal agreement takes the default third-person form; when the subject is null, by contrast, the agreement expresses the person relations. The same pattern seems to hold of the particle in (46b), when the object is null, the form of the particle is identical to the possessive pronoun that agrees in person and number with the object (48). (48) Ba mhaith liom pro mo L /do L /a L /a/ar N /bhu ´ r N /a N (m)b(h)ualadh. c good with.me 1s/2s/3ms/3fs/1pl/2pl/3pl strike ‘‘I would like to strike me/you/him/her/us/you/them.’’ It thus seems not unreasonable to claim that in Irish, VSO order involves overt movement of the V to AgrS; overt movement of the subject to the speciWer of the TP to check the EPP requirements— the formalization of the idea that all clauses must have subjects—of the clause; Objects appear in the speciWer of the AgrO category. The speciWer of the AgrS category is used only covertly, distinguishing Irish from languages like English or French (DuYeld 1995; Bobaljik and Carnie 1996). (49)[ AgrSP AgrS-T-AgrO-V [ TP Subject t T [ AgrOP Object t AgrO [ VP t Subj t V t Obj ]]]]. Splitting the InX category into three thus provides three positions for the arguments to appear in. Nevertheless, there is evidence that the AgrS-T-AgrO-V cartography of the clausal layer is not entirely correct. First, again from Irish phrasal categories and cartography 247 consider the position of the VP adverb ariamh ‘‘always’’ relative to the subject (50), taken from McCloskey (1996b). If this adverb is VP adjoined, then the subject must be outside the VP (cf. the discussion in section 11.4 where we showed that McCloskey 1991 uses evidence from ellipsis to argue that both the subject and object in Irish are VP- internal). The presence of the TP category provides a simple analysis for this subject placement. However, the position of the object relative to the adverb is more surprising. It follows an element we are propos- ing is at the left edge of the VP. (50)Nı ´ or shaothraigh Eoghan ariamh pingin. V S adv O neg earned Owen ever penny ‘‘Eogan never earned a penny.’’ This suggests that the accusative-case position is not in an AgrOP that dominates VP. Recall however, the discussion of inner aspect and AgrPs, in section 11.4, that appear in a layered VP. Carnie (1995), following Koizumi (1995) and Harley (1995), argues that the AgrO category is between the vP and the VP, allowing it to follow the vP adjoined temporal adverb. (51)[ vP agent v [ AgrOP Accusative AgrO [ VP V theme]]]. Adding in Travis’s inner aspect, a VP internal AgrOP provides a position for each element in the sentence (51), which—although prag- matically odd—is grammatical. (52)Nı ´ l[ TP Aindriu ´ [ vP ariamh [ AspP tar-e ´ is [ AgrOP a thrachtas a L Neg.be Andrew ever ASP his thesis agro [ VP chrı ´ ochnu ´ ]]]]] finish ‘‘Andrew has never just finished his thesis’’. The order of elements at the top end of the clausal layer is also suspect. AgrS is linked to case, T to the EPP. In a language like Irish, where the subject is supposed to appear in the speciWer of the TP30 we expect to 30 I assume here that the verb in is not in C following DuYeld (1995) and McCloskey (1996a), based on co-occurrence with overt complementizers and the behavior of TP adjoined adverbs. This contrasts with the older analysis of VSO as movement of the V to C Wrst proposed by Emonds (1980), and also found in De ´ prez and Hale (1986), Hale (1989) and Stowell (1989). Watanabe (1993), Clack (1994), Carnie, Pyatt, and Harley (1994) and Carnie, Harley and Pyatt (2000). 248 controversies Wnd EPP eVects, and not Case eVects. The opposite however, is true. Irish has movement for case in the perfective passive (53). (53) Beidh an trachtas crı ´ ochnaithe agam ama ´ rach. be.fut the thesis Wnished at.me tomorrow ‘‘The thesis will have been Wnished by me tomorrow.’’ By contrast, McCloskey (1996b) shows that Irish does not display EPP eVects. Not only does it not have any overt expletives in subject position, but in one kind of unaccusative in the language, where the theme is quirkily marked with a preposition so does not require case, the sole argument behaves as if it is a complement with respect to positioning in non-Wnite contexts, and with respect to various tests for constituency. If the EPP holds in Irish, then we would expect these unaccusative subjects to appear in subject position, but they do not. McCloskey (1996b, 1997) argues that the higher of the two functional categories is the EPP licensor,31 and the lower one is associated with nominative Case. (54)[ F1P EPP F1 [ F2P Nom F2P . Carnie (1995) identiWes these two positions as TP and AgrSP, respect- ively. Taking this together with the layered VP we get the following interleaved clausal and VP layers: (55)[ TP EPP T[ AgrSP Nom AgrS [ vP agent v[ AspP Asp [ AgrOP Acc AgrO [ VP .]]]]]]. See Carnie and Harley (1997), who examine some of the consequences of this approach. For an alternative view of the architecture of the clausal layer, see Dooley (1990), Guilfoyle (1990, 1993, 1994), Rouveret (1991), Fassi-Fehri (1993), Roberts (1994), DuYeld (1995), Roberts and Shlonsky (1996) The whole AgrP endeavor was questioned on conceptual grounds by Chomsky (1995a), who argued that AgrPs have a diVerent status from other functional projections such as TP (see also Iatridou 1991 for arguments against AgrO). For Chomsky (1995a), an element in the syntactic tree must be a legitimate object of interpretation (i.e. have an interpretation at LF). He claims that while TP has a clear semantic function, AgrPs do not, so should not be part of the clausal cartography. 31 For a very diVerent view of the EPP, where the eVects are not due to a particular head but follow from general principles of structure formation, see Mohr (2005). phrasal categories and cartography 249 . entire functional-projection 27 This is part of an larger programmatic restriction of ‘‘economy of expression’’, where forms appear only in the c -structure of. projec- tions of the V. See Bury (2003) for a more up-to-date version of this claim. For the most part, we will focus on the functional category approach