Turkic default agreement

Baker & Vinokurova (2010) argue for a hybrid account of morphologically observable case, which involves both a configurational theory for accusative and dative, as in Marantz (1991), and case assignment by functional heads for nominative and genitive, as in Chomsky (2001). In the spirit of Levin & Preminger (2015), I argue that a configurational theory is enough to account for the assignment of genitive case. Based on the understudied phenomenon of default agreement with complex possessors in Turkic, I argue that a configurational theory is better equipped at handling the distribution of genitive case. I present a phase-based account of case blocking agreement in order to derive default agreement.


Introduction. Baker & Vinokurova (2010) (B&V) and Baker
present a two-modality approach to case assignment, in which different cases may be assigned in one of two ways: either configurationally, as in Marantz (1991), where case is assigned depending on its location and its relationship with other nominals in its domain, or assigned via agreement with functional heads, as in Chomsky (2000). In particular, they argue that the agreement of nominative and genitive case cannot be assigned configurationally based on agreement patterns in Sakha, a Turkic language, although accusative and dative are assigned in this manner.
In the spirit of Levin & Preminger (2015), I argue that positing two different modalities of case is not necessary. I provide two novel empirical arguments for genitive case being assigned as an unmarked case (and, by extension, nominative) not just in Sakha, but other Turkic languages as well. The first argument is the presence of morphologically unmarked pronouns (nominative) in Turkic partitive subjects, such as (biz) iki-miz 'the two of us,' where the pronoun is optional, but the nominal agreement in bold on the head numeral iki is not. Contrary to fact, agreement would assign genitive case to this pronoun inside the partitive subject under B&V's approach. I argue that this can be accounted for if genitive case is assigned configurationally.
Second, I introduce the problem of Turkic default agreement, which has received very little attention in the literature in the case of Turkish, and not discussed at all for other Turkic languages. 1 In Turkic default agreement, complex possessors such as partitive subjects and adnominal pronoun constructions (APCs, ex. 'we Turks') agree fully in finite clauses with matrix verbs, but do not agree with head nouns in simple possessive structures, relative clauses and elsewhere; in other words, when they receive genitive case in Turkic. 2 In these cases, full agreement, 1PL, is not permitted, and default agreement, 3SG is required, as in (1a)-(1d): (1) a.
(Biz) (1PL) iki-miz-in two-POSS.1PL-GEN kedi-si cat-POSS.3SG 'the cat that belongs to the two of us' b. biz 1PL Türk-ler-in Turk-PL-GEN kitab-ı book-POSS.3SG 'the book that belongs to us Turks' c. * (Biz) (1PL) iki-miz-in two-POSS.1PL-GEN kedi-miz cat-POSS.1PL 'the cat that belongs to the two of us' d. * biz 1PL Türk-ler-in Turk-PL-GEN kitab-ımız book-POSS.1PL 'the book that belongs to us Turks' However, in all other Turkic languages, I present novel data to show that default agreement with complex possessors is merely optional, and banned in Uzbek. To account for this, I provide an account of default agreement based on Chomsky (2001)'s weakened Phase Impenetrability Condition and the assumption that case, blocks agreement. A similar observation has been made in Hungarian and Finnish in the unpublished Holmberg (2017), who also argues that case is responsible for blocking agreement; in Hungarian in particular, dative-case marked complex possessors also force default agreement on the head noun.

Background.
2.1. TWO THEORIES OF CASE. The theory of case that this paper will argue against, at least for the assignment of genitive case, is one in which case assignment is a consequence of agreement. 3 Case is assigned to a noun phrase, NP, that is local to its probe, a functional head F 0 . If F 0 c-commands NP and is able to find an NP in its search domain to assign case to. According to Baker & Vinokurova (2010) (B&V), this case assignment is parasitic on the simultaneous φfeature agreement between F 0 and NP. If there is no agreement, there is no case assignment.
I will argue that the assignment of genitive case is best captured under the configurational case approach, in which different cases are assigned based on whether there are other nominals in the same local domain. Marantz (1991) defines four kinds of case in the following hierarchy: 4 (2) Case realization disjunctive hierarchy (top to bottom) a. Lexical/oblique case: case determined by the lexical properties of an item. Ex. quirky case in Icelandic b.
Dependent case: assigned depending on the relationship between nominals in some domain. Ex. accusative and ergative c.
Unmarked case: case assigned automatically to any NP in a clause (nominative/absolutive) or any NP in an NP/DP (genitive) d.
Default case: assigned to any NP left unmarked for case Some authors, such as B&V and Baker (2015), have argued that both of these methods of assigning case exist crosslinguistically, and in some cases even in the same language, such as Sakha. Both argue that nominative and genitive case assignment cannot be accounted for in the Marantz (1991) approach, because nominative and genitive case appear only when a verb or a determiner respectively agrees with them. 5 Some of their evidence is given in (3a)-(3b), which are ruled out because in the absence of nominal agreement, genitive case cannot be assigned.
Masha Masha terilte-ni company-ACC salaj-yy-*(ta) manage-EV.NOML-*(POSS.3SG) 'Masha's managing the company.' (Baker & Vinokurova 2010;p. 634) But as Levin & Preminger (2015) point out, we could assume that agreement itself is parasitic on case and derive the same results, following Bobaljik (2008). According to Bobaljik, case is assigned configurationally. Agreement looks for case-marked nominals and the appropriateness of the target obeys the Revised Moravcsik Hierarchy, which is as follows: unmarked case ≫ dependent case ≫ oblique case. 6 Only the unmarked cases, nominative and genitive, are accessible to agreement in Turkish and Sakha. Rather than saying case is parasitic on agreement, we could provide Levin & Preminger (2015)'s account of agreement in Sakha under the configurational case theory, in which the probes T 0 , D 0 or Poss 0 will search their c-command domains for a nominal bearing unmarked case and agree with it. A final assumption I want to make is that nominative case is just caselessness, following Bittner & Hale (1996) and Kornfilt & Preminger (2015) among others, who argue that unmarked case is best represented as having no proper grammatical representation.
2.2. DEFAULT AGREEMENT WITH COMPLEX POSSESSORS. This phenomenon is not unique to Turkish or the Turkic languages. In an unpublished manuscript, Holmberg (2017) attempts to come up with an account of default agreement in Finnish and Hungarian, although he does not refer to it as default agreement. He notes that the pronoun in the adnominal pronoun construction (APC) does not reflect the case assigned to the entire pronoun; for example, in the possessive structures below, the possessive clitic must attach to the entire APC, and not the determiner which heads the possessor: (4) a. * your children opinions b. ? you children's opinions He notes that this possible in Finnish, however. First, he points out that Finnish has APs: Importantly for our purposes, he notes that in a possessive construction in which the possessor is an adnominal pronoun, there cannot be agreement on the head noun; in Finnish the null form of the noun is the 3rd person singular form, as seen in (6a)-(6b). This is despite normal possessive structures in Finnish having obligatory or optional agreement, depending on the dialect, as in (6c). In addition, the pronoun of the AP has genitive case in (6a)-(6b): How do we account for there being default agreement only with genitive case-marked APCs? An intuitive answer is to rely on the KP layer projected by the genitive case blocking agreement, and indeed, this is Holmberg's answer. The nominative does not project a KP layer according to this account. In the tree below, KP, which blocks agreement, is circled in gray: In a nutshell, Holmberg proposes that KP blocks access to the φ-features of the pronoun, but the derivation does not crash and default agreement arises on the head noun. But this has the obvious problem of deriving default agreement even with genitive case-marked regular pronouns; Holmberg does not address this problem. Both regular and adnominal pronouns must be casemarked, or in other words project KPs; Holmberg seems to suggest only APCs project KPs.
Even so, evidence for the KP-account is seen in Holmberg's evidence from Hungarian, which has dative-marked possessors. Hungarian allows two kinds of possessive constructions, one in which the possessor is morphologically unmarked and the possessor follows the definite article, and another in which the possessor is dative-marked and precedes it: nektek you.DAT a the vélemén-ye-tek opinion-POSS-2PL 'your opinion' Dative case-marking blocks agreement with adnominal pronouns, as seen in (9) This provides us reasons to assume that agreement itself is parasitic on case. 8 3. Turkic default agreement and partitive subjects. 3.1. DATA. Turkic default agreement occurs on head nouns with complex possessors like partitive subjects, for example (biz) iki-miz 'the two of us,' and adnominal pronouns, for example biz Türkler 'we Turks.' 9 In the case of adnominal pronouns, the pronoun cannot be dropped, as in (10a). This is perhaps due to competition with the generic plural Türkler 'Turks,' which cannot mean we Turks; unagreement is not possible.
On the other hand, the partitive subject construction (partitive) does have the possessive suffix on the numeral; this could indicate that there is agreement between the pronoun in the partitive and the numeral. Furthermore, only numerals can be the head noun in the partitive. An example of the partitive is given in (10b). These examples also show that full agreement is required in simple finite clauses, because the removal of the 1st person verbal agreement -k would lead to there being 3rd person agreement on it instead as it is null: The two of us won.' However, the pronoun in the partitive is usually dropped, ex. iki-miz 'the two of us.' I assume that it is always present but optionally null for two reasons: the φ-features on the possessive suffix have to come from somewhere-likely from a null pronoun in the partitive, as Turkish is a famous pro-drop language-and because in any context with a partitive, a pronoun can optionally be pronounced overtly, perhaps for emphasis purposes.
Nominal agreement on a bare pronoun in a partitive is unpredicted by Baker's analysis of genitive case assignment in Sakha, given that it would predict the partitive pronoun to have genitive case rather than nominative. A Poss 0 under Baker's analysis should always assign only genitive case, but we see that it is bare; *biz-im iki-miz is ungrammatical as a partitive subject.
However, full agreement is never possible with complex possessors; this leads to default agreement, perhaps due to the presence of genitive case. This is shown in (11) below.  (12) below; full agreement is impossible in each of these: want-PRES 'Zeynep wants the two of us to come.' Inflected infinitival clauses c.
Zeynep Zeynep said-PST 'Zeynep said the two of us will go.' Nominalized clauses With this background in mind, we can now consider an independent argument for the opacity of genitive case-marked adnominal and partitive pronouns being opaque for agreement. Kornfilt (2003) points out strong evidence in favor of this, based on an asymmetry between argument and adjunct nominalized clauses. She points out that the subjects of factive nominalized clauses must be nominative if the clause is an adjunct, as shown in (13b): know-PROG-PST-1SG 'I knew the truth when Ali broke the glass.' In these factive nominalized adjuncts, the subject must be nominative; it cannot be genitivemarked, and it triggers full nominal agreement, as shown in (14a). The partitives and APCs show the same pattern in (14b), but they never trigger default agreement, as seen in (14c). This indicates the opacity of these elements, when marked with the genitive, to agreement. gid-e-me-di-m. go-ABIL-NEG-PST-1SG 'Because we cooked, I was unable to go to the concert.' We now move on to discuss the different properties of Turkic default agreement. Aydın (2008) argues that default agreement is always optional in finite clauses, based on his sentence below where the partitive is paired with sadece 'only': Default agreement is not optional in the much simpler counterpart (10b), so there must be another factor at play. Let us take for granted that case does block agreement. Rather, the reason default agreement is optional is perhaps because sadece projects focus, which is syntactically represented and can also block agreement, similar to case. Indeed, there is evidence that sadece can block full agreement even in simple clauses, as shown by the contrast in (16) 'Ten of us applied to Harvard, but (only) two of us were accepted.' Finally, if focus and case can both separately block agreement, one prediction of this account would be that there could be blocking effects, causing default agreement, even with regular pronouns if they had both focus and case. This prediction may be borne out, as shown in (19), but it was not acceptable to most of the Turkish native speakers consulted. I will propose an analysis of default agreement which could explain these facts is Focus is a phase head: 11 ... self i , which Despic argues shows that only/even projects a focus-phrase, which then blocks c-command out. 11 The optionality of the agreement, however, is something I leave open to future research. It is possible that sadece or bile can optionally adjoin as an adjunct, or take a DP as a complement, the latter of which would make DP a Spell-Out domain, but the former would not. The reason it is unacceptable to most could be because it may only adjoin as an adjunct, and not as a phase head, for these people.  (22a) (24). I also propose that the pronoun in APCs is located in Spec,NumP given the plurality of the lexical NP in APCs, as in (25) I concur with Höhn (2019) that the structure of APCs in head-final languages such as Turkish with prenominal APCs differs from that of the structure given in Höhn (2017); the pronoun in APCs is a specifier in Turkish rather than the head, unlike in English where it is the head. 13 But I must make one explicit stipulation. As noted prior, the partitive subject construction biz ikimiz has a bare pronoun with possessive agreement. In an unmarked theory of genitive case, how can a pronoun be left bare if it is in some Spec position of the nominal? Here I must slightly change the definition of unmarked case in Marantz's hierarchy; it cannot be assigned in Spec position of the extended nominal domain; it can only be assigned in Spec,NP (nP to be more precise). This definition is given as follows: (26) Unmarked genitive case: Assigned automatically to any NP in a Spec,nP position.
We can now move onto default agreement. I take for granted that possessive structures have the same basic structure as in Alexiadou et al. (2007) and Holmberg (2017), among others, who argue that Poss is present as a functional projection containing the possessor in its specifier, as in (27). In line with Alexiadou et al. (2007), we can also assume that the possessor is base-generated in a Spec,nP position and moves up to PossP, as I schematize in (27). Genitive case has been assigned to the possessor in Spec,nP, in line with the configurational theory of case, since there is no competitor in the domain: Let us derive agreement in a simple PS such as bizim kedimiz 'our cat.' K itself does not inherit the φ-features on DP, in line with Ackema & Neeleman (2018) who make the same point based on independent reasons. As Holmberg (2017) also points out, this assumption is required to block agreement with quirky case-marked nominals in Icelandic. Finally, note that there are two Spell-Out domains, which I have circled. For Bošković (2011) and Despić (2015), among others, the phase head is merely the highest head in the nominal domain; K if there is no DP and D if there is no K. However, for this account to go through, I must assume that K and D are both phases, rather than just one being the phase, as is commonly assumed. 14 As a result, the two Spell-Out domains are DP and PossP. However, only one of these Spell-Out domains is impenetrable, in line with Chomsky (2001)'s weakened PIC, defined below: 15 (28) Phase Impenetrability Condition (weak): In phase A with head H, the domain of H is accessible to operations outside A only until the next (strong) phase head is merged.
Only NumP becomes inaccessible to further operations when K is merged. The head of the regular pronoun, D, has already inherited the number features from NumP via agreement. The Poss probe agrees with its goal, DP; though DP is the Spell-Out domain of K, Poss itself is not a phase head, and it is therefore in its search domain. This structure is given in (29) The derivation of default agreement is slightly different, but it follows immediately if we assume a weakened version of the PIC. If the AP and the pronoun in the partitive are both unable to move to Spec,DP, then they would be outside of the search space of the outer Poss probe. 16 Agreement with a regular pronoun was possible because the DP itself had φ-features to agree with. However, in both the partitive and AP construction, the bearer of φ-features has been Spelled-Out; the features do not pass on to DP, or whatever the maximal projection is. As a result, agreement is attempted, but it fails, triggering default agreement in the Preminger sense rather than the derivation crashing. This tree is shown in (30); the Spelled-Out domain is in gray: One potential problem could be that my account could accidentally derive full agreement in (31b), via agreement with the embedded 1SG pronoun because the PS is nominative, which does not project a KP layer to help block agreement. But the pronoun itself has genitive case, and benim kedim 'my cat' is itself a DP, so agreement with the embedded pronoun is blocked.
other Turkic languages. This might simply due to a lack of a phase-edge feature on D in Turkish, but it is optionally present in the Turkic languages where default agreement is optional, and obligatory in Uzbek.
Full agreement in finite clauses can be derived with the weakened version of the PIC. The phase head K is not present if nominative case is caselessness, and it does not project a KP layer; T 0 itself is not a phase head, as Citko (2014) shows, and it can therefore agree with the caseless matrix subject. In addition, the data in the subsection prior with default agreement with only can also be derived if Focus 0 is a phase head, as we have already discussed. As noted above, I have left open the question of why this agreement is optional; it is possible that there are two different ways for sadece to adjoin, one of which is with a phase head and the other is an adjunct, and default agreement reflects the former while full agreement reflects the latter.
These derivations rely on the assumption that DP is the maximal projection of the nominal phrase in Turkish; it would be preferable to eliminate it. If there were another phase head in the nominal layer, this would be straightforward. One way of doing so might be to assume that the pronoun bearing φ-features is base-generated deeper inside the nominal and stays there. Since Embick & Marantz (2007) the categorizing head n 0 has usually been considered to be a phase, and this would get the same results as the DP layer. However, one must assume that the pronoun is not at the edge of the nP phase, and deeper inside, so that it is not accessible to probing.
This account successfully derives default agreement in Turkish, along with Finnish and Hungarian. What remains is how to derive the optionality of default agreement in most other Turkic languages, and outright ban of default agreement in Uzbek. One can assume that the pronoun bearing φ-features obligatorily moves to a Spec,DP position in Uzbek, to a phase edge position so that it is able to agree with the outer Poss probe. This process may be optional in the other Turkic languages. However, I cannot eliminate the stipulatory nature of this proposal, and I leave this to future research.
3.3. THEORETICAL DISCUSSION. Default agreement is difficult to account for without further assumptions in the B&V account of genitive case assignment. This analysis relies on the presence of a KP layer, put on the possessor via configurational case assignment, blocking agreement in possessive structures; it happens prior to agreement, and is not parasitic on agreement. By contrast, there is nothing to block agreement under B&V's account of case assignment.
One potential solution under B&V's account, which I will argue against, can be found by changing the nature of the operation Agree, which is responsible for agreement. Arregi & Nevins (2012), among others, splits the operation of Agree into two sub-operations, Agree-Link and Agree-Copy. Agree-Link is responsible for matching the probe with its goal, and Agree-Copy may copy features between the two Agree-Linked objects. Agree-Copy can happen either within the narrow syntax or postsyntactically, but Agree-Link can only happen in the former.
Suppose that difference between default and full agreement is syntactic vs. semantic agreement; one can argue that the kind of agreement in Turkic finite clauses is semantic and not syntactic, but it must be syntactic in possessive structures. This might be due to the presence of the KP layer on the possessor. One such account is provided by Smith (2015), who attempts to derive the difference in plural agreement between American and British English; semantic agreement is possible in British English but not American (ex. the committee are here). This account could also provide an account of default agreement based on this framework, if we stipulate that Agree-Link is responsible for assigning genitive case, but Agree-Copy, which happens later in the derivation, does not, and KP blocks Agree-Copy from copying semantic features; it resorts to copying syntactic features instead.
This account makes two incorrect predictions: first, the lack of genitive case on the pronoun in partitives such as biz iki-miz 'the two of us' remains problematic, given that Agree-Link has not assigned genitive case to the pronoun despite agreement. More importantly, it makes an incorrect prediction with polite pronouns in Turkish, which are syntactically plural but semantically singular. If agreement in Turkish finite clauses was semantic, we would expect agreement with polite pronouns to be semantic, which is contrary to fact; it is obligatorily syntactic in both finite clauses and possessive structures, as shown below: (32) a. Siz 2PL geldi-niz. came-POSS.2PL 'You sg came.' (polite) b.
Siz-in 2PL-GEN kedi-niz cat-POSS.2PL 'Your sg cat' (polite) One further prediction this account would have to make is that agreement in Uzbek is always semantic; since as noted prior in this section, full agreement is always required in Uzbek. But similarly to Turkish, Uzbek always has syntactic agreement with polite pronouns: To recap, the argument that case blocks agreement with complex possessors in Turkish seems to be correct; it is not clear how the B&V approach would derive the same results.

Conclusion.
Prior to concluding, I would like to point out some potential issues to this approach pointed out by Paparounas & Akkuş (2020), who provide independent arguments for genitive case blocking agreement in Turkish, coming to the same conclusion presented here. But they also point out potential issues with the approach; for example, that the account provided here makes very specific assumptions on the nature of case assignment and its timing. I would like to point out that, as they also assume case assignment must take place before agreement, this is not such an issue after all; that is the only assumption that is needed for the argument in this paper to go through, and it is precisely the same assumption that they make as well. 17 A more important issue is that they point out that the approach here is not capable for accounting for the full range of facts, noting that, for example, inflected pronouns such as bizler "we-PL" and complex reflexives also trigger default agreement, yet the approach here only discusses APCs and partitives. If we assume, following Kornfilt (2003) that such complex reflexives have a null pronoun subject with a complex structure-precisely as APCs and partitives dothat would provide a straightforward derivation of default agreement. Second, it might be plausible to think of inflected pronouns as very similar in structure to APCs, but with a null noun that agrees with the pronoun.
I have argued that deriving the existence of Turkic default agreement is straightforward if, as the empirical evidence suggests, case is responsible for blocking agreement; as in languages such as Hungarian where dative case-marked complex possessors also trigger default agreement. At the very least, this paper provides a novel empirical and theoretical argument to an ongoing debate concerning the relationship between case assignment and agreement.