(12) Gweld y ci y mae’r dyn. see the dog wh be-the man ‘‘It is seeing the dog that the man is.’’ (Sproat 1985: 178) McCloskey also notes that the participle and object can be the focus of the ach ‘‘only’’ particle, an honor reserved only for constituents in Irish (McCloskey 1983): (13)Nı ´ raibh me ´ ach ag de ´ anamh grinn. neg be.past I only prog make fun ‘‘I was only making fun.’’ (McCloskey 1983: 20) There thus seem to be numerous examples of VP-like constituents in VSO languages, lending some support to the idea that VSO order is derived from some structure that has a VP constituent. There are, however, some problems with such an analysis. It is not at all clear that these structures are headed by verbs. In the traditional gram- mar of Irish, all of the constructions discussed above involve a ‘‘verbal noun’’ (see Willis 1988,Borsley1993, 1997,DuYeld 1996, Guilfoyle 1997, Borsley and KornWlt 2000, Carnie 2006bforadiscussionofthese‘‘mixed’’ categories in Celtic). Within the LFG framework, I argue in Carnie (2006a) that if these structures are analyzed as NPs then some otherwise baZing properties of Irish copular constructions follow naturally. I claim there that Irish has a full NP structure but a Xat clausal structure. Irish is not only verb initial but more generally predicate initial: (14)Is9 dlı ´ odo ´ ir (e ´ ) Liam decl lawyer (agr) Liam ‘‘Will is a lawyer.’’ What is surprising is that when the predicate is a noun, it may be complex10 (15a, b). This is in contrast to verbal structures where the object may not be adjacent to the tensed verb11 (15 c). 9 I assume here, following O ´ Se ´ (1990), Doherty (1992), and Carnie (1995, 1997) that Is here is not a true verb but a complementizer indicating declarative mood. See the above- mentioned work for evidence in that regard. 10 See Carnie (1995, 2000), Doherty (1996, 1997), Legate (1997), Lee (2000), Massam (2000), Travis and Rackowski (2000), and Adger and Ramchand (2003) for alternative analyses of these kinds of construction. 11 Kroeger (1993) presents remarkably similar data from Tagalog predicate-initial struc- tures and uses this to argue that Tagalog is non-conWgurational, and has a completely Xat structure both in verbal and non-verbal constructions. Kroeger’s analysis cannot be extended to Irish which diVers in some signiWcant ways from Tagalog, including the fact that Irish has strict VSO order. 230 controversies (15) (a) Is [amhra ´ na L bhuailWdh an pı ´ obaire] ‘‘Ma ` iri’s Wedding’’. decl song wh play.fut. the piper ‘‘Ma ` iri’s Wedding’ is a song which the piper is going to play.’’ (b) Is [fear alainn] Liam. decl man handsome Liam ‘‘Liam is a handsome man.’’ (c) *D’ [o ´ l tae] Sea ´ n. past drink tea Sea ´ n ‘‘Sean drank tea.’’ I proposed that a system like LFG allows a straightforward explanation of the diVerences among verbal and non-verbal predicates in these constructions. Following Nordlinger (1998), I proposed that phrase structure categories vary not only over category (as is standard in X-bar theory) but also over phrase-level. These elements vary both in terms of category (N, V, A, P, etc.) and in terms of phrasal level (word/ head, phrase, etc.). I notate this variable as X P .12 This variable will interact with the set of phrase structure rules to produce situations where verbal predicates can only be heads, but nominal predicates can be heads or phrases. In particular, this will occur because Irish has an NP rule, but not a VP rule. Consider the following Irish c-structure rules: (16) (a) S ! X P NP NP "¼# ("SUBJ)¼# ("OBJ)¼# (b) NP ! Det N NP "¼# "¼# ("OBJ)¼# The head ("¼#) of the sentence rule is variable in terms of both phrasality and category: X P , meaning that either a phrase or word can be inserted into this position. There is no VP rule,13 so an N, NP, or V can all feed into this position. Because of the phrasal variable X P , either words or phrases may appear in the predicate position. Nominal predicates are allowed to surface either as simple nouns or as complex NPs. By contrast, with verbal predicates, only the verb with no mod- iWers or complements is allowed in this position. This is because Irish 12 Nordlinger (1998) uses C, Bresnan (2001) uses X (italicized). 13 See Borsley (1989, 1996) for the claim that the subject argument with Wnite predicates in the VSO language Welsh is a second complement, rather than a subject, which provides some support for the lack of a VP proposed here. phrasal categories and cartography 231 has an NP rule (as attested in other positions, such as the subject position), but no VP rule. Returning now to the constituency tests from Irish participle constructions, since these are at least partly nominal in character, the fact that such a constituent exists follows from the NP rule does not necessarily argue for a VP. Other arguments against a Xat, VP-less, structure for VSO lan- guages comes from tests of the relative prominence of subjects and objects. This of course assumes that subject–object asymmetries are deWned over tree structure (using, for example c-command) rather than over argument structure or functional structures as in HPSG, LFG, and RRG. If one does not accept that subject–object asymmet- ries are best expressed by c-command then the following arguments evaporate. In Xat structure, subjects and object are each other’s sisters. Given this, we expect that there will be no structure-dependent subject– object asymmetries in VSO languages. The evidence seems to point away from this. For example, in Irish, a reciprocal14 in subject position cannot be bound15 by an object (16b), but the reverse is grammatical (17a).16 (17) (a) Chonaic Sea ´ n agus Ma ´ ire lena che ´ ile. saw John and Mary with.their other ‘‘John and Mary saw each other.’’ ’(b) *Chonaic lena che ´ ile Sea ´ n agus Ma ´ ire. saw with their other John and Mary ‘‘Each other saw John and Mary.’’ 14 A brief comment about reXexives in Irish is in order here. Strangely, Irish seems to allow completely unbound instances of the reXexive particle in emphatic contexts: (i) Chonaic se ´ fein an re ´ altlong. saw he self the starship ‘‘Himself saw the starship.’’ (referring to a particular person in the discourse setting) Because of this emphatic use of the reXexive morpheme, which in such contexts seems to have little or nothing to do with true anaphora, I avoid using reXexives as examples of anaphora in this book and use reciprocals, which do not have this emphatic reading, instead. See O ´ Baoill (1995) for discussion. 15 Here, we operate under the standard, but not incontrovertible, assumption of Reinhart (1981, 1983) that binding theory makes reference to the relations of c-command, rather than simple linear precedence. 16 DuYeld (1995) presents similar evidence of subject–object asymmetries which are not dependent upon binding theory. He notes that, in Irish, resumptive pronouns are allowed in object position, but are not allowed in subject position. 232 controversies Similar eVects are seen in Niuean (Seiter 1980, Woolford 1991) and Berber (Choe 1987). Sproat (1985) and Hendrick (1988, 1990) show that subject and object arguments in Welsh and Breton diVer with respect to parasitic-gap eVects. Anderson (1984) presents evidence from con- trol in relative clauses in Kwakwala (Kwakiutl) that show similar eVects. On the other hand, Craig (1977) and Woolford (1991) present data from argument prominence in Jacaltec Mayan in favor of a Xat structure. The binding facts in Jacaltec seem to indicate that the object does indeed mutually c-command the subject, as would be predicted in a Xat-structure analysis. An R-expression embedded in the subject NP cannot be co-referent with an object pronoun:17 (18) (a) Xil [smami naj pel] Ø i . saw poss-father cl Peter him ‘‘Peter i ’s father saw him j ’’. *‘‘Peteri’s father saw him i .’’ This data could be analyzed as a condition-C eVect (Chomsky 1981), where the object c-commands the R-expression in the subject NP: ()S c-command V saw NP NP N father NP him D CL N Peter Thus, Jacaltec might well be a candidate for a Xat-structure VSO language, as Woolford claims. The problem with such an analysis, however, is that Jacaltec does show standard subject–object asymmet- ries. For example, just as in English, reXexives are not permitted in subject position (Craig 1977). Similarly, only subjects are available for the rule of Promotion discussed by Craig. This phenomenon, similar to subject-to-subject raising, is seen in the following example: (20) X’iche smunla naj. asp.abs.3.began erg.3.work cl ‘‘He began to work.’’ 17 Due to constraints on disjoint reference, the object pronoun must surface as null in this construction; see Craig (1977: 158). phrasal categories and cartography 233 Perhaps, then, the eVect seen in (18) is due to something other than condition C. The ungrammaticality of (18) with the coreferent reading could be due to a condition-B violation on the object pronoun. The R-expression possessor of the subject NP is functioning like the sec- ondary head of that NP,18 thus its features percolate to the higher NP node and trigger a condition-B violation. The subject NP c-commands the object. Note that this kind of head-like behavior of possessors is found in many languages; for example, Japanese allows passivization of possessor NP (Terada 1991). This kind of analysis is too complex to elaborate on here, can be concluded that it is more consistent with the other evidence from Jacaltec, which suggests that subject–object asym- metries do occur in the language. Aissen (2000) presents a diVerent analysis of these facts; she claims that they are not due to the binding theory at all, but that they are due to eVects of obviation, where the head of the genitive (father) is forced to be marked as an obviative, which in turn is incompatible with a subject position. Under both these analyses, these facts cease to be evidence in favor of the Xat structure approach. Borsley (1989) argues that VP constituents exist in VSO languages at an abstract level. Working in GPSG (although see Borsley 1996 for an update of this analysis in HPSG, using a lexical rule), he proposes there is a metarule, similar to the Subject–Aux-inversion rule, which licenses a verbal projection with two NP complements from rules with a single complement. VPs in this system never exist in the constituent structure, only in the licensing rules; see Dowty (1996) for a discussion of VSO languages using categorial grammar. In this system, VP constituents also exist only at the abstract tectogramma- tical level. 11.3.3 The VP-Internal Subject Hypothesis (VPISH) On the basis of the discussion above in sections 11.4.1 and 11.4.2, let us adopt the idea that there is at least a constituent that corresponds to the verb and its complement. The traditional view was that this was the 18 See Napoli (1989) for a related discussion of how the embedded PP in NPs like that Xower of a girl is the semantic head of the NP. This is shown by the fact that verbs selecting [þhuman] complements can select for such NPs, despite the fact that the syntactic head of the NP is [Àhuman]. For example, the verb marry can only take [þhuman] complements, yet the sentence I want to marry that Xower of a girl is (sexism aside) grammatical. 234 controversies VP. Building on Kitagawa (1986), Koopman and Sportiche (1991)19 claim that the verb þ complement constituent is actually V’. Further, they claim that the speciWer of the VP category is the underlying position of the agent, building on Stowell’s claim that subjecthood is a property of speciWers. Surface order in languages like English (where the subject precedes tensed auxiliaries) is accomplished by movement of the agent to the speciWer of TP (IP): () TP TЈ TVP agent VЈ V … The evidence for this is both conceptual and empirical. On a concep- tual level it allows us to unify (or at least more tightly constrain) the thematic properties of agents and themes. Under the VPISH, these roles are both assigned locally within the VP.20 Empirically, this kind of structure provides a straightforward account of both post-auxiliary subjects in English and for the position of subjects in VSO languages. In existential impersonal constructions21 in English such as (22), the subject NP four men follows the tense auxiliary. If the agent of the verb sit is VP-internal, then this order follows straightforwardly: () (a) There were four men sitting on the floor. (b) TP There TЈ T were VP NP VЈ four men V sitting … 19 See also Guilfoyle, Hung, and Travis (1992). 20 In GB theory, the V governs (under m-command) both the arguments in the clause and theta role assignment happens under government. 21 See Carnie and Harley (2005) for a fuller treatment of existential impersonal constructions. phrasal categories and cartography 235 Assuming a rule of verb movement, where tensed verbs raise to T to check their inXection (or are inserted there because of head mobility as in LFG), then VSO order is also falls out from the VPISH.22 () TP TЈ T+V VP NP subj VЈ t V NP obj McCloskey (1991), building on research by Chung and McCloskey (1987), provides strong evidence for this kind of approach, using data from Modern Irish. One net eVect of this kind of analysis is that there is a putative subject þ object surface constituent in VSO languages, the VP less the moved verb. We thus expect to Wnd constituency tests that target this constituent. The element that appears in the rightmost position in right node raising constructions must be a constituent in Irish. The S–O sequence obeys this test: (24)Nı ´ or thug, no ´ is beag ma ´ thug, [an neg gave, or C small if gave, the pobal aon aird ar an bhean bhocht]. people any attention on the woman poor ‘‘The community paid no attention or almost no attention to the poor woman.’’ (McCloskey 1991) Other evidence for such a constituent comes from ellipsis phenom- ena in Irish. Irish has a process of VP ellipsis which parallels English VP ellipsis in many ways. It applies under identity to a linguistic (i.e. non-pragmatically deWned) antecedent. It is immune to island con- straints. It may apply ‘‘backwards’’ (with the antecedent following the elided material). It tolerates antecedent contained deletion. Finally, it shows strict/sloppy pronominal interpretations. McCloskey thus claims that this phenomenon is the Irish equivalent of English VP 22 We also require some principle to explain why the subject does not raise to the speciWer of the TP as in English. This could be rightward case assignment (Koopman and Sportiche 1991), or covert movement (Bobaljik and Carnie 1996) or some other system. 236 controversies ellipsis. It diVers from English VP ellipsis, however, in what is deleted. In English, the subject obligatorily remains, but the verb and the object (and any other VP internal material) is elided and replaced with did (too). In Irish on the other hand, the verb is the one element which is not elided, rather, it is the VP constituent which is elided: (25) English: S V O and S V O Irish: V S O and V S O (26) Duirt me ´ go gceanno ´ dh sı ´ e ´ agus cheannaigh subj object. said I that would.buy she it and bought ‘‘I said that she would buy it and she did.’’ The VPISH coupled with a process that moves the verb higher in the clause provides us with an elegant account of these facts. The verb has raised outside of the domain of the ellipsis process, whereas the subject and object remain within the VP constituent, which is elided.23 11.3.4 Stacked VPs, Split VPs, vP While the evidence is fairly convincing that subject arguments start lower in the constituent structure than their Wnal position, there are reasons to doubt that they are as low in the tree as the VP. In fact, some converging evidence has led to the idea (now widely accepted in the P&P literature) that agent and other subject arguments are not directly introduced by predication with the verb. Instead the VP structure is 23 One possible objection to this is the lack of other types of constituency eVects such as focus under ach ‘‘only’’ (i) or clefting (ii) (i) *Nı ´ fhaca ach [beirt an duine]. neg saw but two-people the man ‘‘Only two people saw the man.’’ (ii) *[Sea ´ n teach i nDoire] a cheannaigh. John house in Derry C bought Lit.: ‘‘It was John a house in Derry that bought.’’ McCloskey claims, however, that these violations should not be taken as evidence against the constituency of the remnant VP. Instead, he argues that the ungrammaticality of (i) and (ii) follows a violation of the Empty Category Principle. Recall that the VP has the trace of the verb movement in it. If the VP is fronted to the beginning of a clause in a cleft (higher than the verb), or is right-adjoined to the clause in an ach focus, then this trace is not antecedent-governed by the verb, accounting for the ungrammaticality of the forms. phrasal categories and cartography 237 composed of two distinct structures,24 one which represents the lexical verb, its modiWers and a single complement, and a higher structure variously called ìP (Johnson 1991), PredP (Bowers 1993), vP (Chomsky 1995a) and VoiceP (Kratzer 1995). I will use the vP category here since it is by far the most common abbreviation. The head of the vP is the category v, which is called either ‘‘light verb’’ or ‘‘little v’’. ()TP TЈ T vP NP agent vЈ v VP VNP theme With this ‘‘split’’ or ‘‘stacked’’ verbal projection system we have a hybrid between the traditional VP and the VPISH. The agent argument is generated low, thus giving us all the beneWt of the VPISH but also gives a privileged status to the subject. There are a surprising number of arguments in favor of splitting the agent oV of the VP this way. Marantz (1984) observes that external arguments do have several distinct properties. While we Wnd sentential idioms and VþO idioms, there are no VþS idioms. The meaning of the verb can be more directly aVected by the object than by the subject. The examples Marantz gives are in (28)and(29). The (a) examples represent the canonical meaning of the verb. Metaphorical meanings are forced on the verb by the objects in the other examples. So throwing a party does not actually involve swinging the arm and releasing an object. The particular meaning of the verb here is determined by the object. This can be contrasted to (30) where the subject does not seem to alter the meaning of the verb in the same way.25 For a contrasting view of these facts, see Bresnan (1982) and Grimshaw (1992). 24 The two predicatesare linked to one another semantically, not through predicate argument composition, but through a rule of Davidsonian event identiWcation; see Kratzer (1995). 25 There may be a couple of apparent exceptions to this. For example, The death of his father threw John, where throw is taken to mean ‘‘emotional destabilize.’’ However, notice that such interpretations are only available with an animate theme, so the metaphorical content of this construction might be tied to just the object, or be fully sentential in nature. 238 controversies (28) (a) throw a baseball (b) throw support behind a candidate (c) throw a boxing match (d) throw a party (e) throw a Wt (29) (a) kill a cockroach (b) kill a conversation (c) kill an evening watching TV (d) kill a bottle (e) kill an audience (wow them) (30) (a) The policeman threw the ball (b) The boxer threw the ball (c) The social director threw the ball If Marantz is correct about this, then we do not want to tie the agent too closely to the verb. Separating the agent argument out using the v category is one way to do this. A variety of little v heads have been proposed, which vary in their semantic content (a causative, a form that means become, and various forms associated with voice). There is morphological evidence to support this. Many of the Austronesian languages have overt morph- ology on the verb corresponding to these forms. Sentence (31)isan example from Malagasy. (31)Mþanþsasa ny lamba (amin ny savony) Rasoa. Tþvþwash the clothes with the soap Rasoa ‘‘Rasoa washes the clothes with the soap.’’ Stacked light verbs also can explain the fact that English allows multiple auxiliaries. Multiply stacked vPs provide a head for each of these (32). A similar analysis of auxiliaries is found in HPSG (Sag, Wasow, and Bender 2003). (32) (a) Mike is playing bridge. (b) Mike had played bridge. (c) Mike must play bridge. (d) Mike was beaten. (e) Mike has been playing bridge. (f) Mike was being beaten. (g) Mike must be playing bridge. phrasal categories and cartography 239 . predicate-initial struc- tures and uses this to argue that Tagalog is non-conWgurational, and has a completely Xat structure both in verbal and non-verbal. examples of VP-like constituents in VSO languages, lending some support to the idea that VSO order is derived from some structure that has a VP constituent.