ИСТИНА

In Tatar, the finite predicate exhibits person/number agreement with its subject. (Marked) person (i.e. 1/2p) agreement is obligatory. The puzzle arises with inflected quantifiers (1), which allow for both non-agreeing (i.e. 3p) and agreeing (1/2p) pattern. The variation encompasses universal and existential quantifiers and adjectival interrogative pronouns. Agreeing inflected quantifiers (AIQs) are attested in Quechua (Muysken 2013), Kinyarwanda (Jerro&Wechsler 2015) and Turkish (Ince 2007). Here I consider and reject two analyses proposed in the literature for similar phenomena — namely, agreement with the prominent possessor (Bárány et al. 2019), (Deal 2017) and agreement with the subject pro (Ince 2007) and claim that AIQs can acquire a marked person feature via feature sharing (Pesetsky&Torrego 2007). AIQs are built as a possessive construction and allow for the restrictor pronoun to surface in the possessor position (2). However, the pattern in (1) cannot be accounted for as a prominent possessor control, since predicate agreement with true possessors is ungrammatical (3). Ince (2007) analyzes a similar phenomenon in Turkish relying on the hypothesis that AIQs are merged with the silent pro ([DP DP pro<1/2p>]). When the DP splits, pro produces person agreement, whereas AIQ remains caseless. Crucially, pro-doubling is only licit in finite subjects, and this is why Turkish disallows AIQs as embedded subjects and possessors. In Tatar, however, person agreement with AIQs is licit in all agreement contexts (possessive, embedded nominalized and postpositional constructions), and AIQs are genitive, not caseless. Since in these constructions only one genitive case is licensed, they cannot contain a silent pro as an agreement controller. Analysis. I propose that person agreement with AIQs is an instance of the standard AGREE and that AIQs bear the ϕ-features this agreement reveals. I build on the idea that AIQs contain a minimal pronoun equipped with a set of unvalued interpretable features (hence, no correlation between interpretability and valuation, along the lines of Pesetsky&Torrego 2007). I adopt Déchaine&Wiltschko’s (2002, 2010) proposal that pronouns come in different size; specifically, I distinguish between DP-pronouns (which are indexical, cannot shift and be bound) and φP-pronouns (which are non-indexical, can shift and be bound). Overt 1/2p pronouns in Tatar never shift and resist binding, hence are DPs, whereas pro<1/2> can shift and be locally bound (Podobryaev 2014), hence ambiguous between DP and φP construals. I propose that AIQs differ from the non-agreeing ones in that they contain an additional layer between D and (substantivized) nominal structure, namely, the φP layer. The interpretable unvalued features on φ are identified with the φ-set of the restrictor in Spec, DP (e.g. via reverse AGREE, Wurmbrand 2017). Interpretable valued features on φP are further inherited by D, producing 1-2p AIQs which are definite R-expressions. Non-agreeing inflected quantifiers are non-pronominal, that is, they lack the φP layer and are structurally identical to possessive noun phrases.