2.2 Case and Agree
2.2.1 Case and Agree in syntax
It is an ongoing debate where case and Agree take place: in syntax or morphology (at the PF branch). Discussing English, Vergnaud (1977/2008) proposed three types of case, Subject Case, Genitive Case and Governed Case, based on syntactic configurations: “the Subject Case, which is the Case of subjects in tensed clauses; the Genitive Case (cf. Mary’s book, hers, yours, mine, etc.); the Governed Case, which is the case of complements of verbs and prepositions, among others (cf. Mary saw him,Mary gave him a book,Mary talked to him,a book by him, etc.)” (Vergnaud 1977/2008:3). Vergnaud proposes these case types despite the fact that morphology does usually not distinguish between, e.g., Subject Case and Governed Case; John
is no less of a Subject Case than he in John / He ate and Mary is no less of a Governed Case than her in John saw Mary / her, even though only the pronouns show morphological distinction.
Ideas of case along these lines are worked out further in Chomsky 1980, 1981, where abstract Case is assigned in syntax before it is morphologically realized.
(1) a. NP is nominative if governed by AGR
b. NP is objective if governed by V with the subcategorization feature: –NP (i.e., transitive)
d. NP is genitive in [NP–X0]
e. NP is inherently Case-marked as determined by properties of its [–N]
governer (Chomsky 1981:170)
More recently, Chomsky (2000, 2001) has argued for syntactic case as being parasitic on ϕ-agreement, where unvalued case features are checked in syntax via Agree.
There are also various proposals, going back to at least Yip et al. (1987), H.Á. Sigurðsson (1989) and Marantz (1991) that argue for a less responsibility of syntax in case manipulation.
In his works on Icelandic, H.Á. Sigurðsson has emphasized that A-licensing is unrelated to morphological case. On Sigurðsson’s (2012a, 2012b) case star augmen- tation account for case, syntax does not operate with case features (see also the No Case Generalization in H.Á. Sigurðsson 2009). A-licensing relates to Case, but is unrelated to morphological case.
Yip et al. (1987) propose that syntax contains a linearly ordered case tier, nom- inative to the left of accusative. Case is “associated” to arguments from left to right in nominative-accusative languages, such that in a transitive clause with two structural case arguments, nominative is first associated with the argument on the left (the higher argument) and then accusative with the argument on the right. Intransitive clauses have the same case tier but only nominative is associated with an argument as there is only one argument.
(2) Icelandic Ég I DP lamdi hit hana her DP nom acc Ég I DP lamdi hit hana her DP | nom acc Ég I DP lamdi hit hana her DP | | nom acc → →
Quirky, or lexical, case does not interfere with nominative-accusative case associa- tion, it is on a separate case tier. Under Yip et al.’s (1987) approach, nominative and accusative are associated with syntactic positions but quirky case with certain arguments of a verb.
However we want to formulate it, with a case tier or not, structural accusative case assignment or realization requires nominative case in the same dependency or cycle. H.Á. Sigurðsson (2003, 2006c) dubs this the Sibling Correlation, Woolford (2003) attempts to replace Burzio’s (1986:178) generalization with the generaliza- tion that “[t]he object gets nominative Case when there is no (nominative) subject” (Woolford 2003), and Marantz (1991) proposes a disjunctive case hierarchy, where accusative case DP, governed by V+I, is dependent on another, higher argument, governed also by V+I.
As in Yip et al. 1987, even though syntactic configurations matter, Marantz (1991) does not make a division into abstract Case licensing in syntax and case realization in the morphology. However, government plays an important role in Marantz’s approach (and that is the case also for the classical GB approach; see,
e.g., Chomsky 1981). In MC (on the PF branch), the syntactic output is interpreted for the realization of case, on a disjunctive case hierarchy.
(3) Case realization disjunctive hierarchy a. lexically governed case
b. dependent case (accusative and ergative) c. unmarked case (environment-sensitive)
d. default case (Marantz 1991:24)
In Marantz’s hierarchy, shown in (3), more specific requirements are selected over more general requirements. On that hierarchy lexically governed case is the most specific, followed by dependent case (accusative in a nominative/accusative sys- tem), unmarked case and default case, respectively. Accusative (dependent case) is assigned by “V+I to one argument in opposition to another argument position”; the argument’s case is dependent on properties of another DP position, also gov- erned by V+I (Marantz 1991:24). Even though this is not abstract Case (licensing) in the sense of, e.g., Chomsky 1981, this is an extra abstract layer in the deriva- tion. Syntax manipulates arguments and their relations with other elements in the tree. At Spell-out, the output is sent to the Morphological Component which gives arguments a case label (unmarked, dependent, etc.) according to the argument’s position in the tree.
Marantz stresses the fact that there is not a one-to-one realationship between syntactic licensing of arguments and their eventual case realization. This becomes
very clear in Icelandic, as he points out, where nominative does not seem to be dependent on T licensing it. In addition, dative case arguments, for example, can originate as the complement of the verb or higher in the structure, such as in SpecApplP; oblique case arguments can move to a derived subject position but they are never generated in the external argument position of the verb phrase, which is here taken to be SpecVoiceP. It is therefore a legitimate question to ask, how much work T or v carry out in case assignment.
McFadden (2004) argues in a similar fashion that morphological case is deter- mined without reference to Case licensing. As for Marantz, the syntactic output is important for case realization at the PF branch, although “[w]hatever syntactic Case/DP-licensing is, it has no empirical connection to case morphology” (McFad- den 2004:10).
A more recent version of the case realization on the disjunctive hierarchy, with an emphasis on unmarked and dependent case, is presented below, as formulated by Wood (2011:8):
(4) If a DPαhas no case feature at spellout, it is assigned accusative iff there is some other DP α0 which is visible to α and where (a) α0 has no case feature and (b) α0 c-commands α. Otherwise, α will be nominative.
(Wood 2011:8)
That is, if two DPs are in the same dependency, or domain, and one of them does not get inherent or quirky case, one of these DPs, the highest structural case
argument, will get unmarked case (nominative in a nom/acc system) and the other one will receive dependent case (accusative in a nom/acc system). This is a somewhat radical approach, given the original formulation by Marantz which defined case realization with respect to syntactic relations. Here, however, the focus is on whether or not a DP has a case feature at Spell-out, without any reference to syntactic position.
Approaches to post-syntactic case may differ on where to place Agree (ϕ- agreement) in the derivation. H.Á. Sigurðsson (2006a), for example, places Agree in syntax and post-syntactic agreement realization in (deep) PF. Bobaljik (2008), on the other hand, argues that Agree also takes place post-syntactically.
Timing of agreement with respect to case assignment is crucial to Bobaljik’s (2008) account: “[I]f agreement is dependent on the outcome of postsyntactic oper- ation (m-case), then agreement must also be postsyntactic” (Bobaljik 2008:297). He argues that Icelandic gives examples of agreement being dependent on the outcome of case calculations. In Icelandic, finite T shows overt agreement with subjects but also objects (never both at the same time) but only if they are in the nominative case (e.g., Thráinsson 1979, Zaenen et al. 1985, H.Á. Sigurðsson 1990–1991).
(5) Icelandic a. Þessir these bílar cars.nom hafa have.3pl aldrei never hentað suited mér. me.dat ‘These cars have never suited me.’
b. Mér me.dat hafa have.3pl aldrei never hentað suited þessir these bílar. cars.nom
That is, it looks like T must know whether its target DP bears nominative case or not. That would mean that if case were post-syntactic, Agree would also have to take place post-syntactically.
There have also been more syntactic approaches to case which use a disjunctive case hierarchy, most notably Legate (2008) and Preminger (2011, 2014). Legate (2008) argues for syntactic licensing of case. She discusses case in absolutive-ergative languages in which absolutive case is the elsewhere case (default case). In these specific languages, which have rich case morphology, there is not a morphological distinction made between nominative and accusative. Legate argues, however, that there is an abstract nominative Case as well as an abstract accusative Case. A syntactic [case] feature is translated to nominative Case when it forms an Agree relation with finite T but accusative Case translation of [case] is not dependent on finite T. Intransitive subjects are therefore in the nominative Case and transitive objects in the accusative Case, even though both are realized in the absolutive. The prediction, which is borne out in three out of the four languages (Legate notes that it cannot be tested for Niuean), is that absolutive on intransitive subjects should be unavailable.
Preminger adopts Marantz’s disjunctive case hierarchy and applies it in syn- tax. Unlike Legate, however, he does not argue for syntactic licensing of case. For him, unmarked case is “the morphological form given to a noun phrase whose case features have never been valued—just as “3rd person singular agreement” is the
morphological form given to ϕ-probe whose features have never been valued” (Pre- minger 2014:207). If there is another noun phrase unvalued for case in the same case dependency, then it will have dependent case. That is, Preminger does not take structural case to be assigned by functional heads but moves Marantz’s disjunctive hierarchy into syntax.
He furthermore argues that Bobaljik’s argumentation, discussed above, is based on the false premise that morphological case is post-syntactic rather than syntactic. Crucial examples for him come from languages that do not have quirky subjects; in such languages an XP is targeted by a findϕ(f) function (see (27) below) and moves subsequently. If XP is not targeted, then it does not move. That is, in these languages, ϕ-agreement feeds movement and since movement takes place in syntax,
ϕ-agreement must take place in syntax.
I argue that Agree and case assignment take place in syntax (pace Bobaljik 2008) but Agree relations and case are interpreted at the Morphological Component and then finally realized at Vocabulary Insertion. Following a large line of work, I take A-movement to take place in syntax as it can have an effect on semantic interpretation (see, however, Sauerland and Elbourne 2002). If relative timing of A-movement can affect Agree, then that suggests that Agree takes place in the same module, namely syntax. We will see an example of exactly that from Icelandic in §2.2.2.1.