2. CASE IS NOT VALUED TO DPs

Traditionally, it has been assumed that Case is valued only to DPs. There is much data that support this hypothesis. For one, in English, overt case is associated only with pronouns, but pronouns have the syntactic status of full DPs. Likewise, the genitive marker (-'s) is merged to a whole DP. A number of important syntactic generalizations can be captured by assuming that since certain elements lack the ability to value Case (e.g., adjectival and nominal heads), they cannot be complemented with DPs unless a Case valuator, such as a preposition, intervenes (John's search *(of/for) the proof). Distributional evidence shows without doubt that Case is associated with DPs rather than, say, with verbs or prepositions.

Yet none of the above speaks directly against the Kaynean model. Surely any model of Case must explain why Case keeps good company with DPs. Instead, the Kaynean model purports to generalize the valuation mechanism by saying that Case is valued to the word-like elements within those DPs, rather than to DPs qua phrases. This does not prevent pronouns from obtaining Case nor does it prevent the use of Case Filter for the purposes of grammatical explanation. Kayne, for instance, reformulates the Case Filter so that it concerns nominal elements (+N) rather than DPs (Kayne Reference Kayne2005:141; see also chomsky Reference Chomsky1981:49).

One type of data which speaks against the assumption that Case is exclusively a DP phenomenon comes from the fact that there are several languages in which grammatical elements inside of DPs value Case to their non-DP complements. These valuators are typically numerals or other quantificational expressions; the phenomenon has been documented for Polish (Przepiórkowski Reference Przepiórkowski, Grover and Vallduví1996, Rappaport Reference Rappaport, Bański and Przepiórkowski2001), Russian (Babby Reference Babby1987, Franks Reference Franks1995) and Serbo-Croatian (Wechsler & Zlatić Reference Wechsler and Zlatić2001). In Finnish, several numerals and other quantificational expressions value partitive Case to their complements (Vainikka Reference Vainikka, Holmberg and Nikanne1993, Reference Vainikka, Nelson and Manninen2003) irrespective of whether the DP occurs in the object or in the subject position. Compare the overt Case features realized in (4a–b) and (4c–d).

1. (4)

The following example from Polish shows that some numerals, such as the numeral sześć ‘six’, value genitive case to the elements they c-command (see Rappaport Reference Rappaport, Bański and Przepiórkowski2001, Franks Reference Franks2002).

1. (5)

Facing this evidence, one way to maintain the idea that Case is valued to DPs is to assume that the numeral/quantifier selects another DP and then values partitive Case to that DP. I will call this the numeral-DP hypothesis. This model is proposed by Vainikka (Reference Vainikka, Holmberg and Nikanne1993), although she assumes in the spirit of the Government and Binding theory that the relevant categories are maximal nominal projections (NPs). Alternatively, one could think that the numeral is complemented with a PP that then contains the DP. Either way, the problem with this hypothesis is that the proposed rule is recursive (i.e., DP → D + NumP / NumP → NUM + DP) and thus it predicts recursive layering of determiners, quantifiers and numerals within the DPs. However, it is generally assumed that one cannot layer determiners/demonstratives, quantifiers or numerals recursively within a single DP. Finnish is not an exception:

1. (6)

Another reason which speaks against this hypothesis is the following. If the numeral would be the nominal head of the DP (taking a PP/DP complement), it should show Case agreement with the matrix element. This prediction is not true; rather, the numeral which values partitive Case always takes a ‘bare form’, that is, a form without any overt Case suffixes. This cannot be explained by relying on the morphological properties of the numeral, since in other grammatical contexts the numeral shows overt Case suffixes (compare (7a–b) with (4c–d)).Footnote ⁴

1. (7)

Hence, if the numeral would be the nominal head of the DP it should be valued the matrix Case. This is not the case:

1. (8)

Third, while the numeral is an optional element within the DP, it cannot appear there alone; rather, it always requires a complement noun.

Fourth, the numeral cannot be complemented with a CP that can be complemented to nominal heads. This is shown in (9a–b).

1. (9)

Fifth, adjectives which occur within the DP always modify the nominal head and never the numeral. Here is one way to see this. In Finnish, the adjective parillinen is ambiguous: it can mean either ‘even numbered’ or ‘something which has a pair’. If this adjective modified an odd-numbered numeral, rather than the head which follows it, the expression below would be contradictory. However, as (10) shows, the reading in which it modifies the numeral is impossible.

1. (10)

Adjectives never modify the numeral, only the (true) nominal head.

Sixth, the numeral which values partitive Case is in singular, whereas the elements below the numeral can be either in singular or in plural; (11a–b) illustrate.

1. (11)

There is no grammatical number agreement between the numeral and the demonstrative; rather, there is singular agreement between the nominal head and the numeral. While this fact is a puzzle in its own terms,Footnote ⁵ it argues against the assumption that the numeral constitutes the nominal head of the NP. Instead, all the data can be fitted into a model in which the numeral constitutes its own head, projecting a numeral phrase, above the nominal head and the NP (Ritter Reference Ritter and Rothstein1991). Under this analysis, the numeral values Case to something other than a full DP.

Another argument derives from the so-called kasa constructions (the label is mine).Footnote ⁶ These are constructions which on the surface look much like the numeral constructions discussed in the previous section: they are constituted by a quantificational expression, such as kasa ‘stack’, valuing partitive Case to its complement. It turns out that these quantificational expressions do constitute the nominal head of the DP, taking a full DP complement, but that they also contrast with the numeral constructions in every relevant syntactic test.

The kasa construction under discussion is illustrated in (12).

1. (12)

The quantificational expression kasa ‘stack’ values partitive Case to its complement, much like the numerals. However, on closer inspection, it turns out that the syntax of this construction is markedly different from the syntax of the numerals discussed above.

First, unlike the numerals, the quantifier kasa shows matrix Case even if it values partitive Case to its complement; see (12a–b). Second, kasa allows both singular and plural complements, as shown in examples (13a–b). This contrasts with the Case of the numerals, (14), which do not allow the plural complement, (13a).

1. (13)

2. (14)

Third, this expression does not belong to the same grammatical category as quantifiers and numerals, since it may coexist with them within the same DP, and is then situated below the numeral:

1. (15)

Fourth, this element may occur in the verbal complement position by itself, while the numeral is awkward, marginal and requires contextual information to be accetable.

1. (16)

Fifth, this expression shares the number feature with all other nominal elements above the kasa:

1. (17)

Sixth, it is possible to insert quantifiers, numerals and adjectives before the partitive complement of kasa:

1. (18)

Thus, DP recursion is possible with kasa but not with the numerals.

The seventh reason to analyze kasa differently from the numerals concerns adjective data. I showed earlier that adjectives cannot modify numerals (see example (10) above). In the case of kasa, this pattern is reversed. An adjective modifies kasa and never its partitive complement:

1. (19)

Quantifier kasa behaves analogously to nominal heads, and exactly opposite with respect to the numerals. If so, then kasa ‘stack’ must be the head of the noun phrase, taking some form of prepositional complement which contains an additional DP. The prepositional element, rather than kasa ‘stack’, then values the partitive Case to the complement. This hypothesis is depicted in (20).

1. (20)

The numeral construction contrasts with this construction in all the relevant tests, showing beyond any doubt that unlike kasa ‘stack’, the numeral is not the head of the DP. Assuming the analysis provided in (20), this proposal can be verified by considering extraction data. In Finnish, the prepositional complement or prepositional adjunct of a noun can be extracted and moved higher within the DP. The resulting expressions e.g (21b) share their distribution with ordinary DPs.

1. (21)

The same grammatical operation applies to the kasa construction irrespective of whether the complement is in elative or partitive case:

1. (22)

2. (23)

However, the result with the numeral is ungrammatical:

1. (24)

The partitive PP in the kasa construction can be extracted beyond its DP host, while this is impossible with numerals. In the following minimal pair, the interrogative clitic -ko has been added to the partitive phrase and the resulting phrase topicalized to the left periphery.

1. (25)

One nominal construction type deserves to be discussed before we can conclude this section. It is possible to complement a full DP to a numeral if the complement DP shows elative case. The nominal head within the upper DP layer is then optional. Because the head is optional, I will call these head drop constructions. Footnote ⁷

1. (26)

When the head drop construction occurs without a nominal head, one could – at least theoretically – pursue the hypothesis that it exemplifies a numeral-DP or numeral-PP construction. According to this hypothesis, the two forms underlying (26), one with the nominal head and the other without it, should be analyzed as follows:

1. (27)

According to an anonymous NJL reviewer, this hypothesis would in turn show that contrary to what has been claimed so far, numerals could be complemented directly with DPs or PPs (see example (27b)). Several reasons exist to doubt this hypothesis, to my mind all quite compelling. The strongest reason is that even if the head drop construction is formed by somehow suppressing the nominal head, the head does not seem to go away in the semantic sense. Thus, without any contextual help, expression (26) cannot be interpreted in any other way besides referring to cars (compare (28a–b)).

1. (28)

Another way to see this is via ellipsis. If the head drop construction (26) is paired elliptically with a similar expression that contains an overt nominal head, then (26) behaves exactly as if it contained a nominal head identical in meaning to the first conjunct.

1. (29)

Both conjuncts should therefore share their syntactic structure, hence the latter conjunct should be analyzed as NUM–N⁰–PP. Furthermore, like the kasa construction the head drop construction allows recursive stacking of DPs.

1. (30)

As can be seen from these examples, the elative complement, unlike the partitive construction, allows both singular and plural complements.

The elative phrase can be extracted from the material that precedes it:

1. (31)

As I pointed out earlier, this extraction is impossible with numerals valuing partitive Case; to recapitulate:

1. (32)

2. (33)

The head drop construction is not related to the presence of the numeral per se, as all following expressions, none of which containing the numeral, are grammatical:

1. (34)

Likewise, the head drop is not restricted to elative complements nor to nominal heads. Several nominal constituents may be dropped, as is shown in (35a–b), as long as at least one nominal element remains, shown in (35c).

1. (35)

In sum, the head drop construction differs from the partitive numeral construction in several ways. The simplest assumption seems to be that (27a–b) are syntactically identical, while in (27b) the head has been silenced phonologically by some general mechanism that is not restricted to the DPs containing numerals or elative complements. Furthermore, these properties suggest that the elative complement is like the partitive complement of kasa ‘stack’ and other similar nouns: it constitutes a complement of the head noun, not material above the head noun. Why and when several constituents within the noun phrase can be silenced (provided sufficient contextual information) remains to be studied.

Based on the above evidence, I conclude that the numeral values partitive Case to its complement that is not a DP. This breaks the connection between Case and DPs. If Case can be valued to other phrases besides DPs, the question of course arises to which phrases it may be valued. In the next section I will argue that Case is not valued to phrases at all (at least not in Finnish); rather, it is valued to the lexical elements. This data constitutes yet another piece of evidence against the hypothesis that Case is exclusively a DP-level phenomenon.

3. CASE IS VALUED TO INDIVIDUAL LEXICAL ELEMENTS

I begin with examples (36a–b), which contain a numeral, a nominal head (in partitive) and a demonstrative pronoun and a quantifier above the numeral. Notice that the demonstrative, the quantifier and the numeral all occur in their bare forms, that is, without any visible case suffixes. The same form occurs both in the context in which the whole DP is valued nominative and in the context in which it is valued accusative.

1. (36)

The reason these elements do not show case suffixes (hence they are glossed as Ø) turns out to be more complex than what one would first conjecture. The first point to note is that the plural demonstrative and the quantifier do not show, as a morphological property, nominative or accusative case irrespective of whether the numeral is part of the DP. These forms cannot therefore be used as a diagnostic of which Case feature is assigned to the nominal elements above the numeral. Fortunately, singular demonstrative, such as se ‘that’, shows overt contrast between nominative and accusative Case:

1. (37)

The second point to note about these examples is that the numeral behaves differently: it can inflect for accusative Case, but when it does, it cannot value partitive Case to its complement. Thus if the numeral is in its bare form, the nominal head shows partitive Case. If the numeral inflects for matrix Case, then so does the nominal head. The tentative generalization is that if the DP contains a bare numeral, matrix Case valuation reaches only the elements above the numeral, whereas the numeral values the elements below. A striking confirmation of this generalization can be found by examining the behavior of Finnish adjectives. In Finnish, adjectives may be situated below or above the numeral within the DP. When the adjective is located above the numeral, it is valued matrix Case, as in (38a, c). When it is located below the numeral, it is valued partitive Case, as in (39b, d).

1. (38)

The following data show that the distribution of the accusative and partitive Case correlates strictly with word order:

1. (39)

It turns out that Case distribution is regulated by a locality principle: each element within the DP searches for the closest possible valuator (v, numeral, finite C/T) and values its Case accordingly. The resulting Case valuation relations are illustrated in (40). Valuation takes place under a local c-command relation.

1. (40)

This is exactly what the Kaynean model predicts: Case valuators are valuing Case features to their complements, while the effects of more local valuators override the non-local ones. Finally, Case concord is an automatic consequence of this model, and the discontinuous Case concord pattern (40) is automatically deduced.

4. THE KAYNEAN MODEL OF CASE

In this section, I will take a few steps towards the Kaynean model of Case without trying to be cross-linguistically exhaustive; instead, given the focus of the present paper, the discussion will be limited to Finnish.

I will take as my point of departure in this section what Vainikka (Reference Vainikka1989) and Clahsen, Eisenbeiss & Vainikka (Reference Vainikka, Holmberg and Nikanne1993) call the ‘Structural Correspondence Principle’, namely, the principle that there is a one-to-one correspondence between structural positions and overt structural Case. This provides a good starting point because Vainikka has already argued that this model fits well with the Finnish facts. According to this idea, adopted in many guises in several models of grammar (e.g., Chomsky Reference Chomsky2006:12), Case encodes structure. One way to make sense of the Kaynean approach to Case is to think of the Kaynean model as a particular implementation of this idea. Since we would perhaps like to follow Minimalist tradition and abandon the X-bar theory, correspondence should not be defined with reference to absolute grammatical positions, but instead relativized structural positions which are defined in terms of local relations between functional heads and lexical heads. Chomsky's (Reference Chomsky2006) idea that Case features are valued under the probe–goal relations and Salo's (Reference Salo2003:93–122) hypothesis that Case and lexical categories are valued under ‘feature vectors’, i.e. sequences of c-commanding elements, represent other formulations of this basic insight.

I have adopted a similar hypothesis in my earlier work (Brattico & Huhmarniemi Reference Brattico and Huhmarniemi2006, Brattico & Leinonen in press). The background for these models was in Kayne's work on linearization. In influential work, Kayne (Reference Kayne1994) argued that syntax must establish a certain constellation of asymmetric c-command relations for the purpose of linearization. Our proposal in earlier work was to extend this idea to the theory of Case. Thus, our hypothesis was that in the same way that linearization can be based on asymmetric c-command relations established in syntax, so Case valuation relations should be too. The discontinuous and layered Case valuation patterns reported in this paper and elsewhere are thus the overt manifestation of these asymmetric c-command relations. Let us assume the following tentative formulation of this idea:

1. (41) Each lexical element with an unvalued Case feature is valued this feature by the closest asymmetrically c-commanding Case valuator.

The hypothesis that Case is always valued by the closest c-commanding valuator is problematic on at least two grounds. There are cases where the non-local valuator overrules the local one, and cases where Case valuation does not seem to be based on c-command relation at all. Let us next look at these problems in turn.

The first case emerges when the DP in Finnish is valued some case other than nominative or accusative: in that situation, the numeral declines with the rest of the nominal elements and does not value partitive Case to its complement. This is shown in examples (42a–c).

1. (42)

Babby (Reference Babby1987), who notes the same phenomenon in Russian, has explained this phenomenon by relying on a Case Hierarchy, which states that non-structural Case features overrule the structural Case features (hence the inessive overrides partitive). While the explanation of the Case Hierarchy in itself would constitute a worthwhile inquiry,Footnote ⁸ the interesting thing about this phenomenon is the fact that the Case feature that should be valued to the nominal head cannot be determined locally. Suppose, for instance, that the derivation has reached a point where the numeral is merged to the NP. The nominal head bears an unvalued Case feature. Should we value this Case feature by the numeral? The problem is that we cannot decide this before we know what occurs later in the derivation.

One way out of this problem is to give up local Case valuation and assume the derivation-by-phase approach (Chomsky Reference Chomsky and Kenstowicz2001). If all Case features are valued when the phase (CP, vP) is sealed, then the required information is available and no competing functional heads are present (e.g., elements above v do not compete with v in valuing accusative Case to the object DP). To incorporate this idea into the Case valuation mechanism, the modified version in (43) will be proposed. In this proposal, the notion of ‘highest ranking valuator’ refers to a valuator which is the most local but highest ranking member in the Case Hierarchy. The Case Hierarchy is tentatively defined here so that non-structural Case features rank higher than structural Case features.

1. (43) Executed separately for each phase (CP, vP), each lexical element with an unvalued Case feature is valued Case feature by a Case valuator that is the most local highest ranking member in the Case Hierarchy.

Suppose that the derivation has reached the stage ‘v–VP[. . .DP. . .]’ and attempts to spell-out the vP in (44a–c). The unvalued Case feature of the nominal head now needs to be valued. If there are no valuators (prepositions, numerals) between the nominal head and v, then this functional head counts as the most local and highest ranking member in the Case Hierarchy, and the nominal head will be valued accusative Case (44a). If the numeral head intervenes between the v and N, and we assume the numeral and the v rank to be equal in the Case Hierarchy (being structural Case valuators), then the more local valuator, the numeral, determines the Case of the nominal head (44b). Finally, if the DP is merged with the preposition, then the preposition ranks the highest in the Case Hierarchy and values non-structural Case to the elements inside of the DP; see (44c).

1. (44)

Another problem for (43) concerns situations where Case valuation does not seem to be based on a c-command relation at all. Vainikka (e.g., Reference Vainikka1989, Reference Vainikka, Holmberg and Nikanne1993) has argued that the genitive in Finnish is valued to the specifier position of the (e.g., nominal) head. The following minimal pair (with virtually identical meanings) illustrates this generalization. In (45a), the DP minua ‘me’ appears after the preposition in the complement position (Comp, PP) and takes partitive Case, whereas in (45b) it appears before the preposition in the specifier position (Spec, PP) and takes genitive Case. Furthermore, in (45b) the element lähellä ‘near’ agrees with the precedeing DP. No such agreement is possible in (45a).

1. (45)

In my opinion, Vainikka's arguments and the data are entirely persuasive. But some additional facts raise concerns about how they should be explained.

First of all, if the nominal head has several arguments, as in the case of a nominalized transitive sentence, both prenominal arguments are valued genitive case. We should therefore not pursue this data by assuming that there is one X-bar-theoretic specifier position which is linked with the genitive case.

1. (46)

Second, many specifier positions are not valued genitive case at all (e.g., Spec, TP in finite sentences). Third, the genitive arguments constitute a ‘case distribution island’ in the sense that whatever Case is valued to the DP, the genitive Case features are not affected (e.g., Vilkuna Reference Vilkuna2000:188–193, Brattico & Leinonen, in press).

1. (47)

To explain these additional facts, we assumed (Brattico Reference Brattico2005, Brattico & Leinonen in press) that the underlying structure behind these constructions is as in (48). Here n is a nominalizer head, V is the verbal root and DPs are the two arguments (Marantz Reference Marantz, Dimitriadis and Siegel1997, Chomsky Reference Chomsky2006). Under this configuration, n values genitive Case according to rule (43) to both DPs (and to the nominal elements inside of these DPs) which it c-commands.

1. (48)

The verbal root is nominalized when it raises to n. The n bears an EPP feature much like v and thus it probes the DPs to its specifier position after they have been valued genitive Case. After these transformations, the resulting nP looks as is shown in (49).

1. (49)

The presence of n explains why genitive is associated with certain words (namely, those which contain n) and why several DPs are valued the same genitive Case. The reason why genitive DPs constitute a Case distribution island follows from (43), namely, because they are locally c-commanded by n that is closer to those DPs than, for example, the matrix valuators. Since the genitive arguments constitute a Case distribution island, whatever the higher valuator, we have to assume that no higher Case feature can override the effects of more local n.

Vilkuna (Reference Vilkuna2000) argues that the genitive prenominal modifiers differ in whether they belong to the upper quantificational layer or to the lower attributive layer (see secton 2 above for the explanation of these concepts). This model disagrees with the hypothesis that both genitive DPs are base-generated below the nP (hence to the lower layer) where they are protected from matrix Case valuation. However, one reason to think that both genitive prenominal DPs can be situated in the lower layer is that the adjectives, which are situated in the lower layer without any doubt, may nevertheless precede them:

1. (50)

What appears to be true is that the genitive DP may occur higher in the structure. Huhmarniemi (in preparation) has observed that such displacement has relevance to matters relating to logical scope. She observes that when the genitive DP is situated below the quantifiers/numerals, the DP strongly prefers the narrow scope reading, as in (51b).Footnote ⁹ When it occurs above the quantifiers/numerals, it can take either narrow or wide scope, as in (51a).

1. (51)

This phenomenon is well-known from other grammatical environments. For instance, in certain situations, grammatical elements take narrow scope if left at their base position and wide scope only marginally. The wide scope reading and reconstructed narrow reading occur naturally after leftward movement. In the following example from Finnish, the same phenomenon is observed in the case of topicalization.Footnote ¹⁰

1. (52)

The most common explanation for this phenomenon is to assume that the quantifier raises via A'-movement and is reconstructed to its base position when the sentence is associated with the narrow scope reading. We are therefore led to postulate the same explanation for the variation in the position of the genitive DPs. The genitive DPs may move to a higher position in the DP via A'-movement. The genitive DP situated in the upper layer has therefore been moved into that position.

If the genitive moves, it should follow constraints typical of grammatical movement. One such constraint is ‘Attract Closest’ (Chomsky Reference Chomsky1995:297), which requires that the closest potential goal is moved. That this constraint applies to the DP domain can be shown by using the observation made by Saara Huhmarniemi (in preparation) that in non-echo questions whatever element within the DP is replaced by the corresponding wh-element, that element must occupy the leftmost D position within the DP, as is shown in (53a–b). In agreement with the Attract Closest, the lower genitive DP cannot be moved over the higher one in such a case; it can only be moved if there is no higher DP; see (53c).

1. (53)

The same observation applies to genitive DPs that are not wh-elements. In the example (54a–b), I have tried to move the lower genitive DP. This is impossible, however.

1. (54)

Genitive displacement is thus constrained by Attract Closest, one hallmark of movement. A further argument for the hypothesis that genitive DPs are first generated below the nP and moved to (Spec, nP) or higher is provided by agreement facts. Recall that the nominal head agrees with the genitive DP if the DP is pronominal. Agreement is marked by an agreeing possessive suffix in the nominal head, as shown in (55).

1. (55)

That this is a form of agreement can be demonstrated by comparing it with standard agreement in a finite context. Like a finite verbal context, the pronoun can be dropped in a nominal context if the agreement marker is present. Thus, (55) is grammatical and identical in meaning even if the pronoun is dropped. This mirrors the fact that Finnish is a pro-drop language. Moreover, as in a finite verbal context, the pronoun cannot be dropped if the construction is in the third person singular (see Vainikka & Levy Reference Vainikka and Levy1999). Finally, in MP verbal agreement is triggered when the EPP feature of T probes the subject DP into (Spec, TP). Thus, reasoning analogously, agreement should be triggered by the EPP feature of n after the DP is probed into (Spec, nP). We have just seen that this is exactly what happens.

In agreement with Vainikka (Reference Vainikka1989), I conclude that genitive DPs do appear in the specifier position (Spec, nP) within the noun phrase. However, the valuation of genitive Case is a more complex matter. This Case seems to be valued by n in a downward fashion, after which the DPs raise obligatorily to (Spec, nP) and optionally to a higher position.