The subject of defective Tense in Slavic
In human languages, a phrasal unit that is interpreted thematically at Logical Form (LF) as occupying one syntactic position may occur overtly in Phonetic Form (PF) in a different position, a phenomenon referred to as the displacement property As Chomsky has noted (1995c:221–22), this property appears to be unique to human languages, especially in contrast to artificially constructed formal systems such as logic (in its variety of formulations) and so-called computer languages. It has been a constant focus in the study of modern generative grammar from the outset. Within the Principles and Parameters framework of the past two decades and more particularly within recent refinements of the Minimalist Program, attempts to discover what motivates this property have identified three potential factors: Case, agreement, and the requirement that certain functional categories must have syntactically realized specifier phrases. For example, the displacement that occurs in the English passive construction in (1) below could be motivated in terms of: (i) Case: the nominative (NOM) pronoun must move to the specifier position of the clause (headed by the functional category T(ense) and hence Spec-TP) to be licensed; (ii) Agreement: the pronoun must move to Spec-TP to establish agreement with the passive auxiliary was; and (iii) the property of the functional category T, which requires that the pronoun move to create Spec-TP.
With only English data it is difficult to determine to what extent each factor alone might motivate displacement. From another point of view, this becomes a problem of separating the operative factor from epiphenomena that might cooccur.
This article examines unaccusative constructions in Russian and Ukrainian where the predicate assigns accusative Case (ACC) to a nominal expression that shows up at PF in the specifier position of TP. These constructions provide empirical evidence for teasing apart the operative roles of the three potential factors in displacement phenomena. They show that T, in the absence of agreement or Case-licensing properties, has an independent requirement that triggers movement to its specifier. They also demonstrate a correlation between defective T lacking agreement features and a υ below, which can check ACC in the case of all unaccusative predicates. The correlation is not merely possible, but necessary. These unaccusative predicates cannot occur with T because if they did, the result would not yield a convergent derivation: the of υ would be checked but not the of T. Exactly how this works will be discussed below.
In recent refinements of the Minimalist Program (in particular, Chomsky 2000b, 2001), displacement in narrow syntax is driven by an uninterpretable feature of a head of one of the core functional projections: CP, TP, and υP. Non-defective T and υ have a full set of uninterpretable which are checked via matching of corresponding of nominal expressions. Uninterpretable structural Case features of the nominal expression delete under such matching. Each core functional projection may have a specifier position that is not semantically-selected by the functional head. These are designated “EPP-positions”—i.e., positions not forced by the Projection Principle. These positions are motivated by a functional head’s requirement for an overt specifier.
A major innovation in recent minimalist syntax is that feature checking of uninterpretable features need not involve movement. Uninterpretable features enter the derivation (from the lexicon) unvalued. They are valued and deleted (hence checked) via the relation Agree, which matches the unvalued features of the probe to the corresponding valued features of the goal. In the case of the unvalued uninterpretable of a verbal element, matching with the valued interpretable of an NP results in the valuation of the uninterpretable Case feature of the NP (and its subsequent deletion)—e.g., the on the probe T and the Case feature on the goal NP in (1).
Chomsky 2001 assumes that “a goal as well as a probe must be active for Agree to apply” (cf. his (3i)). To be active is to have an unvalued (uninterpretable) feature. It follows that only N with unvalued structural Case can be active. Once structural Case has been valued via Agree, the NP headed by this N is frozen in place. This would account for the impossibility of raising the subject of a finite clause, as in (2).
Given that NOM on Mary is valued under agreement with the head T of the complement TP, Mary would be frozen in the complement TP. The only way for this derivation to converge is with merger of expletive it in the root Spec-TP. An alternative derivation for (2) in which the Case feature of Mary is valued under matching with the root T would result in the of the complement T remaining unvalued—in which case this derivation fails to converge.
As will be discussed below, the facts from accusative unaccusative constructions suggest that the assumption that an NP is unmovable once its Case has been valued is untenable.
The operation Move applies only in those cases in which the probe has an EPP requirement, in addition to its or wh-features. The question of whether the EPP requirement of T can drive movement independently of its is the central concern of this paper. In the standard case of movement (e.g., the passive construction in (1)), the moved constituent also involves the operation Agree to check structural Case. However, Agree can apply without movement, as in the case of the English existential construction in (3), where the unmoved expletive associate a visitor values the features of T:
In these constructions, the EPP requirement of T is satisfied by the merger of the expletive there.6 In this regard, the satisfaction of the EPP is separate from the checking of Case and agreement. However, these constructions are famously controversial with respect to the whether the expletive bears (NOM) Case and, if so, whether the associate bears some other Case (see Belletti 1988 and Lasnik 1995a for discussion). We will show that the Russian and Ukrainian constructions introduced in section 2 are significantly more straightforward in the way the EPP is distinct from properties of Case and agreement, and thus make an important contribution to this discussion.
Finally, note that the notion “defective category” plays a central role in the analysis of accusative unaccusative constructions. A defective category is defined in (4):
In English, infinitival T (as in (5)) and passive υ (as in (1)) are and hence defective categories.
The independent status of the EPP requirement of T with respect to the Case/agreement system is tested precisely when T is defective. If movement to Spec-TP occurs in the absence of on T, the EPP alone must be sufficient to cause displacement.
In (6–8) we provide examples of accusative unaccusatives in Russian and Ukrainian:
In the case of the Russian construction the indirect internal argument can also appear pre-verbally in discourse-neutral speech. The basic requirement for discourse-neutral word order entails (in descriptive terms) a simple prohibition on V-initial structures. Examples with preverbal instrumental (INST) NPs are given in (7):
The Ukrainian accusative unaccusative construction is a nonagreeing passive. The example in (8c) contains a passive wh-phrase:
The basic facts of these constructions are given in (9):
These predicate types, referred to in Babby (2000) as “un-Burzio” verbs, contradict Burzio’s Generalization, the correlation between a predicate’s thematic and Case properties, given in (10):
In what follows, we will suggest that Case never drives the A-movement attributed to the correlation in (10) (cf. Marantz 1991). Instead, displacement may reduce to the requirement of T for a specifier (cf. Rothstein’s (1983) “syntactic predication”), i.e., a requirement of core functional projections as in Chomsky 2000b.
The central task regarding (9) is to show how these properties are connected in a principled way. We will argue that the underlying ACC- (or INST-) Case-marked complements of these unaccusative predicates move to the specifier position of a defective T, thus providing evidence for separating T’s EPP requirement from its other, purely morphological, uninterpretable Thus the structure of accusative unaccusative constructions is (11) and (12) on the opposite page.
The dotted lines indicate that either the NP:ACC or the NP:OBL can move to Spec-TP. υP lacks a specifier. This indicates that Russian and Ukrainian, unlike Scandinavian, are not object shift languages. ACC Case on the direct internal argument is valued “long distance” via Agree with υ. The oblique semantic Case on the indirect internal argument is assigned by lexical V when the two merge. Because T is (i.e., defective) it cannot value NOM Case.
Note the correlation between Tdef (where Tense lacks agreement morphology, indicated as [–AGR] in (6–8)) and the ability of an unaccusative predicate to converge with an ACC complement. When T is the sole argument of an unaccusative (or one of the internal arguments of a di-unaccusative) would be forced to enter into a relation with T’s uninterpretable a relation that is precluded in Slavic for NPs bearing non-NOM Case. While any unaccusative υ can potentially assign ACC to its complement, this operation will survive only those derivations that do not subsequently require that this ACC NP enter into a relation with T’s Consider in (13) the four logical possibilities for the of T and υ that could occur with unaccusatives:
The configuration in (13a) gives the case of standard unaccusatives, which conform to Burzio’s Generalization. Unaccusative υ is defective and consequently fails to value ACC on its complement. Defective υ creates a weak phase (cf. Chomsky 2001) in which the unvalued complement is still active and, thus, can be valued by the higher T. The result is an unaccusative predicate with a NOM subject (e.g., English arrive and disappear, and their Russian equivalents pryti and isčeznut’, respectively).
When both T and υ are (as in (13b)), a clause whose verb is unaccusative will converge only in the case of di-unaccusatives, where there are two NPs available: one to value the of υ and another to value the of T. The examples in (6a–b) will converge as follows with Tcomp:
In the case of unaccusatives with a single internal argument, the configuration in (13b) will necessarily fail to converge. Only one of the two probes will be able to value and then delete its uninterpretable features.
The configuration of (13c) yields a convergent derivation for the accusative unaccusatives in (6–8). Because unaccusative υ is it values ACC on its complement. The moved ACC NP satisfies the EPP of Tdef.
The accusative unaccusative construction provides some motivation for not considering the EPP to be a species of feature checking. If it were, then Tdef would enter a one-way checking relation with the ACC NP in which the NP checks the EPP-feature of Tdef but Tdef checks no feature of the NP. This would contradict the otherwise well grounded assumption that both probe and goal must be active (i.e. have an unvalued feature) to be related by Agree (see Chomsky 2001).
Turning briefly to (13d), note that this configuration is hopelessly deviant. Assuming that all unaccusatives select at least one internal argument, if both T and υ are defective (as in (13d)), this argument will not be valued for Case. The resulting uninterpretable Case-feature will cause the derivation to crash.
The unaccusative types available with respect to the of T and υ are identified in (15).
Note that the configuration in (15c) is our main concern. The two central questions posed by this configuration are: (i) Can a defective category impose an EPP requirement?15 and (ii) Is a structurally Case-marked NP obligatorily “frozen” for subsequent narrow syntactic operations once its Case is valued?.
