Person Agreement with Anaphors: Evidence from Tatar

Abstract

In this paper, I present evidence for variable agreement with anaphors in Tatar. I show that inflected reflexives trigger co-varying person agreement as DP/nominalization subjects and as complements of postpositions, which appears to contradict the generalization on the anaphor agreement effect (AAE). At the same time, inflected reciprocals induce 3p agreement on external targets. These data are puzzling in two aspects. First, it is unclear how to derive co-varying agreement with inflected reflexives because it cannot be handled as a regular exception to AAE predicted to arise by the agreement-based theory if the antecedent of the anaphor is positioned lower than the agreement target. Secondly, the difference between reflexives and reciprocals with respect to external agreement looks enigmatic. I propose that Tatar reflexives and reciprocals, despite their superficial resemblance, have different internal structures, which in turn bring about differences in their feature sets, and external agreement reveals these differences. As to AAE violations, I propose that the Tatar data can be accounted for under the feature sharing approach whereby the features on the anaphor and on the external probe are first identified as instances of the same feature set and then valued by the anaphor’s binder.

1. Introduction

The aim of this paper is to examine agreement with anaphors in Tatar. Tatar possesses several anaphors which are regularly attested in local contexts: the simple reflexive üz-e ‘self-3’, the reduplicated reflexive üz-üz-e ‘self-self-3’ and the (reduplicated) reciprocal ber-ber-se ‘one-one-3’.1 Similarly to their counterparts in other Turkic languages, Tatar anaphors consist of a root (üz ‘self’, ber ‘one’) and a possessive affix. The possessive affix in anaphors co-varies with their binder with respect to a bundle of person and number features, much like the English reflexive pronouns myself, yourself, etc. Unlike English reflexives, however, Tatar anaphors can occur in syntactic positions construed with agreement: as subjects of nominalized clauses, as genitive possessors in DPs and as arguments of postpositions. Therefore, Tatar provides us with an opportunity to study agreement patterns available for anaphors. The two options we might expect are agreement co-varying with the possessive affix (thereafter person agreement pattern, (1a)) and invariable 3^rd person/default agreement (thereafter default agreement pattern, (1b)).2

(1)	a.	person agreement pattern
		Probe	Goal
		X	üz-em
		[uϕ: 1sg]	self-1sg
	b.	default agreement pattern
		Probe	Goal
		X	üz-em
		[uϕ: 3]	self-1sg

Agreement with anaphors is of interest for several reasons. First of all, there is a robust cross-linguistic generalization called Anaphor Agreement Effect (AAE) which states that anaphors tend to avoid agreeing positions or, if licit in syntactic positions construed with agreement, can only trigger a default, non-co-varying agreement (Rizzi 1990; Woolford 1999; Sundaresan 2016). There are two major approaches accounting for AAE: the feature deficiency approach and the structural encapsulation approach. The feature deficiency approach (Kratzer 2009; Rooryck and Vanden Wyngaerd 2011; Murugesan 2019) relies on the idea that referential deficiency of anaphors results from their featural deficiency: possessing unvalued phi-features, anaphors need them to be valued by syntactic binding. Accordingly, anaphors’ phi-features only become valued after binding. This reasoning underlies the timing-based approach to AAE (Murugesan 2019): if the agreeing probe is lower than the binder, agreement with an anaphor fails or yields default values. The encapsulation approach (Preminger 2019) suggests that the reason for agreement failure is the anaphors’ complex internal structure: their phi-features are buried under a functional layer specific to anaphors, which makes them inaccessible for external agreement probes. Tatar data on agreement with anaphors is of high relevance for this line of research, because they allow us to test predictions of both approaches.

The second reason is that Tatar reflexive and reciprocal pronouns belong to a very intricate structural class of nominals centered around partitive constructions. Thus, Tatar anaphors pattern structurally with inflected quantifiers such as (bezneŋ) barı-bız da ‘all of us’, (sezneŋ) kajsı-gız ‘which of you’, (alarnıŋ) eki-se ‘two of them’, etc., cf. (2a–c).

(2)	a.	Bez	üz-ebez-ne	gajeple	sana-bız.
		we	self-1pl-acc	guilty	believe.ipf-1pl
		‘We consider ourselves guilty.’
	b.	Bez	barı-gız-nı	da	gajeple	sana-bız.
		we	all-2pl-acc	ptcl	guilty	believe.ipf-1pl
		‘We consider you all guilty.’
	c.	Bez	ike-gez-ne	gajeple	sana-bız.
		we	two-2pl-acc	guilty	believe.ipf-1pl
		‘We consider two of you guilty.’

Inflected quantifiers are true partitives (Seržant 2021) or canonical partitives (Falco and Zamparelli 2019), where the quantifier identifies the subset and the optional genitive possessor cross-referenced in possessive agreement denotes the superset (von Heusinger and Kornfilt 2017, 2021). Partitives are known for triggering variable agreement patterns both intra- and cross-linguistically (Martí i Girbau 2010; Danon 2013; Leclercq and Depraetere 2016; Pérez-Jiménez and Demonte 2017); in particular, agreement with inflected quantifiers has been reported to be sensitive to semantics, e.g., group reading vs. distributive reading of the partitive (Pérez-Jiménez and Demonte 2017). Consequently, we might expect Tatar anaphors to pattern with inflected quantifiers in their agreement properties; moreover, we should evaluate agreement with anaphors against agreement with inflected quantifiers. Comparing agreement with anaphors and agreement with inflected quantifiers would allow us to distinguish between AAE, which would only affect anaphors, and general agreement constraints in partitive constructions, which would influence equally anaphors and inflected quantifiers. Looking a bit ahead, Tatar data presented in this paper point towards the latter.

Finally, Tatar data are interesting against the background of other Turkic languages. To date, there is detailed information about agreement with inflected anaphors and inflected quantifiers in Turkish (Aydın 2008; Ince 2008; Kornfilt 1988; Paparounas and Akkuş 2020, forthcoming; Satık 2020); for Kyrgyz, Sakha, Altai and Uzbek, there is a more limited set of data concerning possessive agreement with inflected quantifiers coming from Satık 2020. Although data are scarce for generalizing over all Turkic languages, it is evident that there is significant variation in agreement patterns: thus, according to Aydın 2008 and Ince 2008, in possessive configurations, Turkish only allows for default agreement with inflected anaphors (see also Kornfilt 1988) and quantifiers, and Uzbek strongly prefers person agreement with inflected quantifiers (no data on anaphors), whereas Kyrgyz, Sakha and Altai allow for both patterns with inflected quantifiers (again, no data on anaphors). The data on Turkish inflected quantifiers presented in Paparounas and Akkuş (2020) and Satık (2020) suggest that there can also be variation between agreement configurations; thus, predicate agreement with nominative subjects exhibits the person agreement pattern, whereas predicate agreement with genitive subjects in nominalizations and possessive agreement with genitive possessors only allows for the default agreement pattern. However, more recent work (Paparounas and Akkuş, forthcoming) recognizes that Turkish inflected quantifiers allow for both agreement patterns in all agreement configurations; overt agreement with anaphors is not discussed. Given these findings, the complete Tatar dataset on agreement with anaphors and related constructions would contribute significantly to the intragenetic typology of Turkic languages.

Given what we know about agreement with anaphors in other Turkic languages and cross-linguistically, Tatar presents a previously undescribed case. The striking characteristic of Tatar is that agreement patterns attested with inflected anaphors are distributed not among various agreement positions, but among the anaphors themselves. Specifically, inflected reflexives invariably trigger the person agreement pattern, whereas inflected reciprocals strongly prefer the default agreement pattern. In (3a–b), this contrast is shown for the possessive agreement triggered by the nominalization’s genitive subject.

(3)	a.	Bez	üz-üz-ebez-neŋ	awıl-ga	kil-ü-ebez-gä	/
		we	self-self-1pl-gen	village-dat	come-nml-1pl-dat
		*kil-ü-e-nä		šatlan-dı-k.
		come-nml-3-dat		become_glad-pst-1pl
		‘We were pleased with our return to the village.’
	b.	Bez	ber-ber-ebez-neŋ	awıl-ga	*kil-ü-ebez-gä	/
		we	one-one-1pl-gen	village-dat	come-nml-1pl-dat
		kil-ü-e-nä		šatlan-dı-k.
		come-nml-3-dat		become_glad-pst-1pl
		‘We were pleased with each other’s return to the village.’

Moreover, this distribution is maintained in related partitive constructions. The lexical heads üz ‘self’ and ber ‘one’ are not only used in building reflexives and reciprocals, but also give rise to non-anaphoric partitive constructions exemplified in (4): ber ‘one’ produces the inflected quantifier (one of X), whereas üz ‘self’ produces the inflected intensifier (X oneself). These items, unlike inflected anaphors, are licit in the finite subject position construed with finite predicate agreement. Importantly, in this configuration, they show agreement patterns attested elsewhere with their anaphoric counterparts. The fact that the agreement pattern of a partitive is ultimately determined by its subset-denoting element provides us with a cue for capturing the contrasting properties of anaphors with respect to external agreement.

(4)	a.	Üz-ebez	kal-ırga	ujla-dı-k	/	*ujla-dı
		self-1pl	stay-inf	think-pst-1pl		think-pst
		‘We ourselves decided to stay.’
	b.	Ber-ebez	kal-ırga	*ujla-dı-k	/	ujla-dı.
		one-1pl	stay-inf	think-pst-1pl		think-pst
		‘One of us decided to stay.’

The rest of the paper is organized as follows. In Section 2, I discuss major agreement configurations in Tatar and show that anaphors maintain their agreement patterns across all the contexts. Section 3 examines the internal structure of anaphors and provides their characterization with respect to syntactic and semantic binding. The aim of this section is to demonstrate that the mismatch of agreement patterns between reflexives and reciprocals cannot be attributed to their different status with respect to binding. In Section 4, agreement patterns attested with inflected quantifiers are investigated. I show that the choice between the person agreement pattern and the default agreement pattern strongly correlates with the subset–superset relation specified by the quantifier. Section 5 sketches the analysis of the two agreement patterns based on a structural representation of the semantics of partitives. Section 6 concludes.

The data for this study come from several sources. Non-elicited examples are from the two corpora of Tatar—Corpus of written Tatar (620 mln tokens, https://search.corpus.tatar/en; accessed on 7 August 2022, tagged as [CWT] in the examples) and «Tugan Tel» Tatar National Corpus (180 mln tokens, http://tugantel.tatar/?lang=en; accessed on 7 August 2022, tagged as [TT]). Information about acceptability of anaphors and personal pronouns in various syntactic positions and with various agreement patterns was obtained by running a survey on the Yandex Toloka crowdsourcing platform (https://toloka.yandex.ru/en/; accessed on 7 August 2022); 15 native speakers of Tatar were asked to rate 55 sentences presented in a random order on the binary (yes/no) scale. Another survey whereby sentences exemplifying alternative agreement patterns attested with inflected quantifiers and intensifiers were evaluated against a wider context (forced choice task) was run on the Google Forms service (ten native speakers of Tatar, 15 sentences, two contexts for each). Judgments about availability of strict and sloppy readings were provided by my Kazan colleagues Ayrat Gatiatullin, Alfiya Galimova and Bulat Khakimov.

2. Agreement in Tatar

2.1. Basic Configurations

Tatar exhibits agreement in a wide array of configurations: finite predicate, nominalized predicate, possessive construction, postpositional phrase. An important property of Tatar agreement is that the categories involved in agreement—person and number—are the same for all the agreement configurations. Thus, Tatar differs from, e.g., German or French, which attest verbal predicate agreement for person and number but nominal concord for other categories.

The finite predicate agrees with its nominative subject. There are two sets of person–number agreement markers distributed between TAM forms of verbal predicates, see Table 1.

Table 1. Agreement markers in finite verbal forms (adapted from Zakiev 1995, vol. 2, p. 86).

Subject’s Features	Set I (“Full”): Present, Future 1 and 2, Perfect Indicative	Set II (“Truncated”): Past Indicative, Conditional, Hortative, Imperative
1sg	-mIn3	-m
2sg	-sIn	-ŋ
3sg	—	—
1pl	-bIz	-k
2pl	-sIz	-gIz
3pl	(-lAr)	(-lAr)

Table 1. Agreement markers in finite verbal forms (adapted from Zakiev 1995, vol. 2, p. 86).

Subject’s Features	Set I (“Full”): Present, Future 1 and 2, Perfect Indicative	Set II (“Truncated”): Past Indicative, Conditional, Hortative, Imperative
1sg	-mIn3	-m
2sg	-sIn	-ŋ
3sg	—	—
1pl	-bIz	-k
2pl	-sIz	-gIz
3pl	(-lAr)	(-lAr)

For 1–2p subjects, agreement is obligatory with overt and non-overt (pro) subjects (5a–c). For 3p subjects, there is no special agreement marker for person, and the predicate can optionally bear a plural affix –lAr (5d–e).

(5)	a.	Menä	min	šušı	urın-da	kičermäslek
		but	I	this	place-loc	unforgivable
		zur	xata	jasa-dı-*(m).
		big	mistake	make-pst-1sg
		‘But at that place I made a big, unforgivable mistake.’ [CWT]
	b.	Närsä-gä	öjrän-ep	kajt-tı-*(gız)	sez?
		what-dat	learn-cvb	return-pst-2pl	you
		‘What did you learn?’ [CWT]
	c.	pro1sg	[pro3sg	siz-gän-e]	juk	dip
				notice-pf-3	neg.cop	comp
		ujlıj	i-de-*(m).
		think.ipf	aux-pst-1sg
		‘I thought he did not notice (it).’ [TT]
	d.	Kız-lar	kul-lar-ı-n	jua-lar.
		girl-pl	hand-pl-3-acc	wash.ipf-pl
		‘The girls are washing their hands.’ [CWT]
	e.	Kız-lar	aŋa	borıl-mıjča	tüzä	al-ma-dı.
		girl-pl	this.dat	turn-neg.cvb	resist.ipf	can-neg-pst
		‘The girls could not stand it and turned to him’ [CWT]

The choice between agreeing and non-agreeing predicate with 3pl subjects is influenced by a number of parameters. First of all, number agreement is obligatory if the subject is non-overt, cf. (6). With overt subjects, the use of the plural marker can be semantically motivated, reflecting the collective/distributive distinction (cf. Zakiev 1995, p. 96; Lyutikova 2017, p. 32). However, it cannot be analyzed as a pure “semantic” agreement reflecting semantic plurality of the referent, since collective nouns like police or numeral constructions, which are grammatically singular, never trigger plural agreement (7). Performance factors can influence the use of –lAr as well: the larger the distance between the subject and the predicate, the more likely the plural agreement.

(6)	pro3pl	Jırak-jırak	ǯir-lär-gä	oč-ıp	kitä-*(lär).
		far-far	land-pl-dat	fly-cvb	leave.ipf-pl
	‘They are flying away to distant lands.’ [CWT]

(7)	a.	Policija		aeroplan-nar-dan	gaz	bomba-lar-ı	tašla-dı-(*lar).
		police		airplane-pl-abl	gas	bomb-pl-3	drop-pst-pl
		‘The police dropped gas bombs from airplanes.’ [CWT]
	b.	Kazan-nan		Mäskäü	konservatorija-se-nä		uk-ırga
		Kazan-abl		Moscow	conservatory-3-dat		study-inf
		ike	kız	kil-de-(*lär).
		two	girl	come-pst-pl
		‘Two girls came to study in the Moscow Conservatory from Kazan.’ [CWT]

Possessive agreement is characteristic for the genitive possessive construction (ezafe 3 in traditional grammatical descriptions, e.g., Zakiev 1963). Tatar possessive constructions can feature either a genitive or an unmarked (nominative or caseless) possessor (the latter is characteristic for ezafe 2 constructions); in both cases, the ezafe marker on the head noun is obligatory to license a dependent nominal constituent4. The distribution of genitive vs. unmarked possessors is influenced by a number of structural and interpretational factors (see Pereltsvaig and Lyutikova 2014 for discussion). Importantly, DP possessors (pronominal, definite and possessive noun phrases) cannot be unmarked and require genitive marking. Accordingly, unmarked possessors can only be 3p, whereas genitive possessors are not restricted with respect to the person feature.

The ezafe marker can host an agreement probe responsible for the possessive agreement. Possessive agreement markers are given in

Table 2. Agreement markers in ezafe forms (adapted from Zakiev 1995, vol. 2, p. 32).

Possessor’s Features	Possessive Affix
1sg	-m
2sg	-ŋ
3sg	-I5
1pl	-bIz
2pl	-gIz
3pl	-(lAr)I

.

With unmarked possessors, which can only be 3p, the ezafe marker is invariably 3p (-I/-sI); number agreement is also illicit, cf. (8a–b). I consider this as evidence that the ezafe marker itself does not necessarily contain a phi-probe but can come without it.

(8)	a.	bala-lar	bakča-sı
		child-pl	garden-3
		‘a/the kindergarten’
	b.	bala-lar	bakča-lar-ı
		child-pl	garden-pl-3
		‘(the) kindergartens’ / ‘*a/the kindergarten’

With genitive possessors, the ezafe marker obligatorily displays co-varying possessive agreement. I consider this generalization as a direct outcome of genitive case assignment under phi-agreement. Thus, the ezafe marker licenses the nominal argument whereas the phi-probe is responsible for case assignment.

With 1–2p possessors, the head noun bears a possessive affix which agrees with the genitive possessor for person and number (9). Importantly, the plural marker –lAr on the head noun can only be assessed as an exponent of the interpretable number of the head, cf. (9d). Non-overt (pro) 1–2p possessors are readily available and trigger possessive agreement in the standard manner (9e).

(9)	a.	sineŋ	ukıtučı-ŋ
		you.gen	teacher-2sg
		‘your teacher’
	b.	sineŋ	ukıtučı-lar-ıŋ
		you.gen	teacher-pl-2sg
		‘your teachers’
	c.	bez-neŋ	ukıtučı-bız
		our-gen	teacher-1pl
		‘our teacher’
	d.	bez-neŋ	ukıtučı-lar-ıbız
		our-gen	teacher-pl-1pl
		‘our teachers’ (*‘our teacher’)
	e.	pro2sg	ukıtučı-lar-ıŋ
			teacher-pl-2sg
		‘your teachers’

A deviation from this pattern is a genitive construction lacking a possessive affix, or a possessive-free genitive, PFG, as Satık (2020) dubs its counterpart in Turkish. PFG is only licensed with 1–2p possessors, and for some speakers, with 3p singular pronoun anıŋ (when used for humans) as well (10a–b); however, according to my Tatar consultants, neither personal names nor common nouns participate in PFG constructions (10c–d). The possessive-free genitive construction appears to have a more limited scope of application, but unlike the Turkish PFG (Öztürk and Taylan 2016), it can denote a kinship relation or a part–whole relation with body parts, cf. (11a–b).6

(10)	a.	bez-neŋ	at-ıbız	/	at
		we-gen	horse-1pl	/	horse
		‘our horse’
	b.	a-nıŋ	at-ı	/	at
		this-gen	horse-3	/	horse
		‘his/her horse’
	c.	Marat-nıŋ	at-ı	/	*at
		Marat-gen	horse-3	/	horse
		‘Marat’s horse’
	d.	äti-m-neŋ	at-ı	/	*at
		father-1sg-gen	horse-3	/	horse
		‘my dad’s horse’

(11)	a.	Minem	bala-nı	tatar-ča	ukıt-ma-gız!
		I.gen	child-acc	Tatar-adv	teach-neg-2pl
		‘Do not teach my child in Tatar!’ [CWT]
	b.	Belä-seŋ,	minem	kul	jalgıš-mıj.
		know.ipf-2sg	I.gen	hand	mistake-neg.ipf
		‘My hand never makes mistakes, you know.’ [CWT]

The possessive construction with 3p genitive possessors differs from possessive constructions with 1–2p genitive possessors in that the plural genitive possessor can, but need not, trigger the appearance of the plural marker –lAr on the head noun (12).7 As a result, the plural marker in possessive phrases with a 3p plural possessor is ambiguous between interpretable and agreement-induced (13). Note that these possessive constructions contrast with ezafe 2 constructions with 3p plural unmarked possessors (8), where the plural marker on the head noun can only be interpretable.

(12)	a.	Bondarenko	očučı-lar	mäktäb-en-dä	ukı-gan-da,
		Bondarenko	pilot-pl	school-3-loc	learn-pf-loc
		major	alar-nıŋ	ukıtučı-lar-ı	bul-gan.
		major	they-gen	teacher-pl-3	be-pf
		‘When Bondarenko was in the pilot school, the major was their teacher.’ [CWT]
	b.	Äle	kečkene	čag-ım-da…	min	alar-nıŋ
		already	small	time-1sg-loc	I	they-gen
		ukıtučı-sı	bula	i-de-m
		teacher-3	be.ipf	aux-pst-1sg
		‘Since my childhood, I was their teacher.’ [CWT]

(13)	alar-nıŋ	ukıtučı-lar-ı
	they-gen	teacher-pl-3
	‘their teachers’ / ‘their teacher’

It is important to mention that the nominal possessive structure is also employed in partitives based on quantifiers. The DP denoting the superset is represented by the genitive possessor, whereas the quantifier gets substantivized8 and corresponds to the possessum. Like in the standard possessive construction, the genitive possessor may be non-overt. With 1–2p genitives, possessive agreement is obligatory; with 3p genitives, number agreement is optional.

(14)	a.	bez-neŋ	barı-bız	da	/	pro1pl	barı-bız	da
		we-gen	all-1pl	ptcl			all-1pl	ptcl
		‘all of us’
	b.	alar-nıŋ	ike-se	/	alar-nıŋ	ike-lär-e	/	pro3pl	ike-se	/
		they-gen	two-3		they-gen	two-pl-3			two-3
		pro3pl	ike-lär-e
			two-pl-3
		‘two of them’9
	c.	alar-nıŋ	kajsı-sı	/	alar-nıŋ	kajsı-lar-ı	/	pro3pl	kajsı-sı	/
		they-gen	which-3		they-gen	which-pl-3			which-3
		pro3pl	kajsı-lar-ı
			which-pl-3
		‘which of them’

With most quantifiers, the plural affix –lAr can only be agreement-induced, since universal quantifiers and numerals exclude plural marking on the noun: ike kitap-(*lar) ‘two books’. However, the numeral ber ‘one’, the interrogative modifier kajsi ‘which’ and indefinite modifiers (nindider ‘any’, berkadär ‘some’, etc) are compatible with a plural-marked nominal. In the partitive construction, they can have an interpretable plural marker –lAr. With a 3p plural genitive, the plural marker –lAr is then ambiguous between interpretable and uninterpretable.

(15)	a.	bez-neŋ	kajsı-lar-ıbız
		we-gen	which-pl-1pl
		‘which ones among us’
	b.	alar-nıŋ	kajsı-lar-ı
		they-gen	which-pl-3
		‘which one among them’/‘which ones among them’

Nominalized clauses exhibit two patterns depending on their status in the argument vs. adjunct dichotomy (see Kornfilt 2003, 2007; Aygen 2007 for the same distinction in Turkish). Argumental nominalizations are predominantly headed by regular deverbal nouns (-U) or perfective participles (-gAn); they take case affixes corresponding to the clause’s grammatical function within the main clause. In argumental nominalized clauses, the subject is genitive, and the nominalized predicate bears a possessive affix and agrees with this subject. Agreement with 1–2p subjects is obligatory (16a–b); number agreement with 3p subjects is optional (16c–d).

(16)	a.	Bez-neŋ	žurnal-lar-nı	ǯiber-ü-*(ebez)-ne	sora-dı	ul.
		we-gen	magazine-pl-acc	send-nml-1pl-acc	ask-pst	this
		‘He asked that we send him the magazines.’ [TT]
	b.	Ul	minem	toz-la-gan	it	jarat-ma-gan-*(ım)-nı
		this	I.gen	salt-vbl-pf	meat	like-neg-pf-1sg-acc
		gel	onıta.
		always	forget.ipf
		‘He always forgets that I do not like salty meat.’ [CWT]
	c.	Bügen	min	ukučı-lar-ım-nıŋ	mine
		today	I	reader-pl-1sg-gen	I.acc
		aŋla-w-ı-n		sorıj-m.
		understand-nml-3-acc		ask.ipf-1sg
		‘Today, I ask my readers to understand me.’ [TT]
	d.	Keše-lär-neŋ	üz-e-nnän	kurk-u-lar-ı-n	telä-gän	ul.
		man-pl-gen	self-3-abl	fear-nml-pl-3-acc	want-pf	this
		‘He wanted people to be afraid of him.’ [TT]

In adjoined nominalized clauses with adverbial functions, we observe the same nominalizing morphology; specific semantic relations the adverbial clause bears with respect to the main clause are expressed by case markers and/or postpositions. Thus, dative and locative forms (-U-gA, -gAn-gA, -gAn-dA) introduce temporal adverbial clauses, ablative forms and postpositional phrases (-gAn-nAn, -gAn öčen, -gAn/-U arkasında) introduce causal adverbial clauses, etc.

As a general rule, in adverbial nominalized clauses the subject is nominative, and the nominalized predicate has no ezafe marker and exhibits no agreement (17). However, with several types of adverbial clauses (specifically, with temporal –U-gA clauses and causal clauses introduced by postpositions like –U arkasında) two other patterns are also licit—nominative subject plus agreeing ezafe marker and genitive subject plus agreeing ezafe marker. Tatar grammars describe the variation as free (Zakiev 1995, vol. 3, p. 344; Pazel’skaya and Shluinsky 2007). Adverbial nominalized clauses based on the –gAn participle only attest the general pattern—nominative subject plus ezafe-less predicate (Zakiev 1995, vol. 3, pp. 351–52). In the rest of this paper, I only consider argumental nominalizations, leaving the more puzzling patterns of agreement available in some adverbial nominalizations for future research.

(17)	a.	Sin	kit-ü-gä,	a-nı	tiz	arada
		you	leave-nml-dat	this-acc	quickly	meanwhile
		juk	it-ärgä	mömkin-när.
		neg.cop	do-inf	possibility-pl
		‘Once you leave, it becomes possible to destroy10 it quickly.’ [CWT]
	b.	Min	awır-gan-da,	Kimov	ta	ker-ep	čık-kan
		I	be_sick-pf-loc	Kimov	ptcl	enter-cvb	exit-pf
		bul-sa	kiräk.
		be-cnd	necessary
		‘Kimov must have come in when I was sick.’ [CWT]

Finally, let us consider agreement in postpositional phrases. Among Tatar postpositions, only denominal postpositions exhibit agreement with their arguments. In what follows, we only discuss this subtype of postposition.

Denominal postpositions come in two forms, plain and agreeing. Plain postpositions consist of the relational root and one of the case markers—dat, loc, abl. In agreeing postpositions, there is a possessive affix between the root and the case marker.11 Example (18) illustrates these options.

(18)	a. plain form
	minem	jan-da
	I.gen	near-loc
	‘near me’
	b. agreeing form
	minem	jan-ım-da
	I.gen	near-1sg-loc
	‘near me’

Arguments of denominal postpositions appear in the genitive or nominative (caseless) form. Personal pronouns and 3p sg human pronoun are genitive; all other nominals, including other pronouns, proper names, etc., are caseless.

Case assignment and agreement in postpositional phrases are interrelated. With 1–2p and 3p sg human pronouns, which are always genitive-marked, both plain and agreeing forms of postpositions are licit (19a–b); in the agreeing form, 1–2p agreement is obligatory (19a). With the rest of the nominals, plain postpositions are illicit (19c–e); agreeing postpositions can optionally attach a plural affix –lAr between the root and the possessive affix signaling agreement with a 3p plural argument (19c,e). This option is predominantly used when the postposition’s argument is expressed by 3p plural pro (20).

(19)	a.	minem	jan-da	/	minem	jan-ım-da	/	*minem	jan-ı-nda
		I.gen	near-loc		I.gen	near-1sg-loc		I.gen	near-3-loc
		‘near me’
	b.	a-nıŋ	jan-da	/	a-nıŋ	jan-ı-nda
		this-gen	near-loc		this-gen	near-3-loc
		‘near her/him’
	c.	*a-lar	jan-da	/	a-lar	jan-ı-nda	/	a-lar	jan-nar-ı-nda
		this-pl	near-loc		this-pl	near-3-loc		this-pl	near-pl-3-loc
		‘near them’
	d.	*Marat	jan-da	/	Marat	jan-ı-nda
		Marat	near-loc		Marat	near-3-loc
		‘near Marat’
	e.	*kız-lar	jan-da	/	kız-lar	jan-ı-nda	/	kız-lar	jan-nar-ı-nda
		girl-pl	near-loc		girl-pl	near-3-loc		girl-pl	near-pl-3-loc
		‘near the girls’

(20)	pro3pl	Jan-nar-ı-nda	min	bul-ma-sa-m,
		near-pl-3-loc	I	be-neg-cnd-1sg
	jä	pro3pl	berär	küŋelsez	xäl-gä	tor-ır-lar.
	then		some	unpleasant	situation-dat	stay-fut-pl
	‘If I won’t be with them, they will get to some unpleasant situation.’ [CWT]

Properties of agreement configurations in Tatar are summarized in

Table 3. Agreement configurations in Tatar.

Context	Controller	Case	Target	Features	Status
Finite clause	1–2p pronouns	nom	finite predicate	person+number	obligatory
	3p nominals	nom	finite predicate	number	optional
Possessive construction	1–2p pronouns	gen	ezafe-marked noun	person+number	obligatory
	1–2p pronouns, 3p sg human pronoun	gen	ezafe-less noun	—	—
	3p nominals	gen	ezafe-marked noun	number	optional
Nominalization (argumental)	1–2p pronouns	gen	ezafe-marked nml	person+number	obligatory
	3p nominals	gen	ezafe-marked nml	number	optional
Postpositional phrase	1–2p pronouns	gen	agreeing p	person+number	obligatory
	1–2p pronouns, 3p sg human pronoun	gen	plain p	—	—
	3p nominals	nom	agreeing p	number	optional

.

Summarizing the discussion, 1–2p pronouns trigger obligatory person+number agreement in all contexts construed with agreement. Agreement with 3p nominals is for number exclusively; it is optional unless the controller is non-overt. Additional properties that distinguish 1–2p pronouns (and 3p sg human pronoun anıŋ ‘this.gen’) from other nominals are that they allow for PFG in nominal possessive constructions, are marked with genitive as postpositions’ arguments and can combine with plain denominal postpositions.

2.2. Agreement with Anaphors

In this section, I present data on agreement with anaphors: the simple reflexive üz-e ‘self-3’, the reduplicated reflexive üz-üz-e ‘self-self-3’ and the (reduplicated) reciprocal ber-ber-se ‘one-one-3’. I show that reflexives and reciprocals differ systematically in all agreement configurations: 1–2p reflexives pattern with 1–2p pronouns, and 1–2p reciprocals pattern with 3p nominals.

Tatar anaphors are excluded from the finite clause’s subject position for binding reasons (I address this issue in Section 3); consequently, there are only three types of positions construed with agreement available for anaphors: possessors in nominal possessive construction, subjects of nominalizations and arguments of agreeing postpositions.

Both simple and reduplicated 1–2p reflexives trigger 1–2p agreement in all licit agreement configurations, cf. elicited examples in (21), as well as corpus examples (22) and (23). Thus, they exhibit the person agreement pattern (1a).

(21)	a. possessive construction
	Min	(üz)-üz-em-neŋ	bala-m-nı	/	*bala-sı-n		jarata-m.
	I	self-self-1sg-gen	child-1sg-acc		child-3-acc		love.ipf-1sg
	‘I love my own child.’
	b. nominalization
	Min	(üz)-üz-em-neŋ	sine	üpkälät-ü-em-ne		/	*üpkälät-ü-e-n
	I	self-self-1sg-gen	you.acc	hurt-nml-1sg-acc			hurt-nml-3-acc
	bel-mi	i-de-m.
	know-neg.ipf	aux-pst-1sg
	‘I didn’t know that I was hurting you.’
	c. postpositional phrase
	Min	jılan-nı	(üz)-üz-em-neŋ	jan-ım-da		/	*jan-ı-nda	kür-de-m.
	I	snake-acc	self-self-1sg-gen	near-1sg-loc			near-3-loc	see-pst-1sg
	‘I saw a snake near myself.’

(22)	a. possessive construction
	Min	üz-em-neŋ	xatın-ım-nı	häm	ike	ul-ım-nı	üter-de-m.
	I	self-1sg-gen	wife-1sg-acc	and	two	son-1sg-acc	kill-pst-1sg
	‘I killed my wife and my two sons.’ [CWT]
	b. nominalization
	Üz-em-neŋ	matur	i-kän-em-ne	dä	ǯir-dä	bel-mičä
	self-1sg-gen	beautiful	aux-pf-1sg-acc	ptcl	earth-loc	know-neg.cvb
	jör-gän-men.
	walk-pf-1sg
	‘I walked the Earth without knowing that I was beautiful.’ [CWT]
	c. postpositional phrase
	Kal-dır-gan	bul-sa,	üz-em-neŋ	jan-ım-a	al-ıp	kajt-ır
	stay-caus-pf	be-cnd	self-1sg-gen	near-1sg-dat	take-cvb	return-fut
	i-de-m	dä	ber	iptäš	bul-ır	i-de,	ič-ma-sa.
	aux-pst-1sg	ptcl	one	comrade	be-fut	aux-pst	drink-neg-cnd
	‘If he quitted (drinking), I would take him with me, he would be my only friend, if he did not drink.’ [CWT]

(23)	a. possessive construction
	Min	üz-üz-em-neŋ	uj-fiker-lär-em-ne		šušı	bloknot-ka
	I	self-self-1sg-gen	thought-thought-pl-1sg-acc		that	notebook-dat
	töšer-ep	bara-m.
	put_in-cvb	go.ipf-1sg
	‘I keep writing down my thoughts into that notebook.’ [CWT]
	b. nominalization
	Čönki	bel	ütkän	zaman-da	üz-üz-ebez-neŋ	kem
	because	know.imp	past	time-loc	self-self-1pl-gen	who
	bul-gan-ıbız-nı	tirännän	aŋla-p		bel-ergä	tiješ.
	be-pf-1pl-acc	deeply	understand-cvb		know-inf	need
	‘Because we should know it having understood deeply who we were in past times.’ [CWT]
	c. postpositional phrase
	Äni-eŋ	ölkän	jaš-tä …	al-ıp	kil	üz-üz-eŋ-neŋ	jan-ıŋ-a.
	mom-2sg	old	age-loc	take-cvb	come.imp	self-self-2sg-gen	near-2sg-dat
	‘Your mom is old … come and take (her) to your place.’ [CWT]

Note also that 1–2p reflexives pattern with 1–2p pronouns, in that they are genitive-marked in postpositional constructions (24a), can appear in the possessive-free genitive construction (24b–c) and combine with plain denominal postpositions (24a,d–e)12.

(24)	a.	Min	a-nı	üz-em-*(neŋ)	jan-ım-da	/	jan-da		kür-de-m.
		I	this-acc	self-1sg-gen	near-1sg-loc		near-loc		see-pst-1sg
		‘I saw it near myself.’
	b.	Bez	Azija	belän	Jewropa	ara-sı-nda		üz-ebez-neŋ
		we	Asia	with	Europe	between-3-loc		self-1pl-gen
		kunak-lar-nı	karšı		ala-bız.
		guest-pl-acc	towards		take.ipf-1pl
		‘We meet our guests between Asia and Europe.’ [TT]
	c.	Üz-üz-egez-neŋ	bala	belän	arlaš-u
		self-self-2pl-gen	child	with	communicate-nml
		alım-nar-ı-n	üzgert-egez.
		manner-pl-3-acc	change.imp-2pl
		‘Change your manner of communication with your child.’ [CWT]
	d.	Ber	zaman	min	bul-ma-m,	üz-em-neŋ		urın-ga
		one	time	I	be-neg.fut-1sg	self-1sg-gen		instead-dat
		Rädif-ne	kal-dır-a-m.
		Radif-acc	stay-caus-ipf-1sg
		‘Once I am gone, I leave Radif in my place.’ [CWT]
	e.	Üz-ebez-neŋ	ara-da	šulaj	gına	atıj-bız		sez-ne.
		self-1pl-gen	between-loc	so	only	call.ipf-1pl		you-acc
		‘Among us, we only call you like that.’ [TT]

With 1–2p reciprocals, on the other hand, we generally find the default agreement pattern, cf. elicited examples in (25) and corpus examples in (26); only a couple of corpus examples shown in (27) attest 1–2p agreement, which was judged as marginal by my consultants. In postpositional phrases, 1–2p reciprocals appear in the nominative (caseless) form, which is also characteristic for 3p nominals; moreover, they do not occur in the PFG construction and do not combine with plain denominal postpositions (28).

(25)	a. possessive construction
	Bez	ber-ber-ebez-neŋ	bala-lar-ı-n	/	*bala-lar-ıbız-nı		jarata-bız.
	we	one-one-1pl-gen	child-pl-3-acc		child-pl-1pl-acc		love.ipf-1pl
	‘We love each other’s children.’
	b. nominalization
	Bez	ber-ber-ebez-neŋ	sine	xatın		it-ep
	we	one-one-1pl-gen	you.acc	wife		do-cvb
	sajla-gan-ı-n	/	*sajla-gan-ıbız-nı	bel-mi		i-de-k.
	choose-pf-3-acc		choose-pf-1pl-acc	know-neg.ipf		aux-pst-1pl
	‘We didn’t know that each of us had chosen you (as wife).’
	c. postpositional phrase
	Bez	jılan-nar-nı	ber-ber-ebez-(*neŋ)	jan-ı-nda		/	*jan-ıbız-da
	we	snake-pl-acc	one-one-1pl-gen	near-3-loc			near-1pl-loc
	kür-de-k.
	see-pst-1pl
	‘We saw snakes near each other.’

(26)	a. possessive construction
	Barıber	ber-ber-ebez-neŋ	isänleg-e-n	beleš-ep	tora	i-de-k.
	still	one-one-1pl-gen	health-3-acc	ask-cvb	stay.ipf	aux-pst-1pl
	‘Still, we kept asking about each other’s health.’ [CWT]
	b. nominalization
	Ber-ber-ebez-neŋ		sulıš	al-u-ı-n		išetä-bez.
	one-one-1pl-gen		breath	take-nml-3-acc		hear.ipf-1pl
	‘We feel each other’s breath (lit. We hear each other’s taking breath)’ [CWT]
	c. postpositional phrase
	Bez	küp	millet-le	töbäk-tä	jäši-bez,	ber-ber-ebez	tur-ı-nda
	we	many	people-atr	state-loc	live.ipf-1pl	one-one-1pl	about-3-loc
	kečkenä-dän	bel-ep	üs-ü	bik	kiräk.
	childhood-abl	know-cvb	grow-nml	very	necessary
	‘We live in a multinational country; we shall grow up knowing about each other from our childhood.’ [CWT]

(27)	reciprocal, agreeing (possessive construction)
	a.	Kitap-ta	da	“ber-ber-egez-neŋ	gajeb-egez-ne	ezlä-mä-gez”
		book-loc	ptcl	one-one-2pl-gen	fault-2pl-acc	search-neg-2pl
		dip	jaz-ıl-gan.
		comp	write-pass-pf
		‘It is also written in the Qur’an, “do not look for each other’s fault”.’ [CWT]
	b.	Uŋıšlı	xezmättäšlek	öčen	ber-ber-ebez-neŋ
		beneficial	cooperation	for	one-one-1pl-gen
		mömkinlek-lär-ebez-ne		häm	ixtıjaǯ-lar-ıbız-nı
		capacity-pl-1pl-acc		and	interest-pl-1pl-acc
		öjrän-ergä	kiräk.
		study-inf	necessary
		‘For a mutually beneficial cooperation, we have to study capacities and interests of each other.’ [CWT]

(28)	a. possessive construction: no PFG
	*Bez	ber-ber-ebez-neŋ	bala-lar-nı	jarata-bız.
	we	one-one-1pl-gen	child-pl-acc	love.ipf-1pl
	Int.: ‘We love each other’s children.’
	b. postpositional phrase: no plain postpositions
	*Bez	jılan-nar-nı	ber-ber-ebez-(neŋ)	jan-da	kür-de-k.
	we	snake-pl-acc	one-one-1pl-gen	near-loc	see-pst-1pl
	Int.: ‘We saw snakes near each other.’

To sum up, 1–2p reflexives induce the person agreement pattern and behave like 1–2p pronouns in other grammatical respects (form a PFG construction, combine with plain denominal postpositions, require genitive marking with postpositions). On the other hand, 1–2p reciprocals generally trigger the default agreement pattern and behave like 3p nominals in other grammatical respects (do not form a PFG construction, do not combine with plain denominal postpositions, lack genitive marking with postpositions).

2.3. Tatar Agreement in Theoretical Perspective

In this section, I address theoretical accounts of Tatar agreement and show its relevance for all theories aiming to tackle agreement with anaphors.

I consider all Tatar configurations discussed in Section 2.1 as syntactic agreement configurations. They have all the hallmarks of agreement: they involve different extended projections, they are based on a one-to-one correspondence between probe and goal, they involve person features and they are rigidly connected to case licensing.

In fact, possessive agreement is the exact counterpart of the predicate agreement in the nominal extended projection. This parallelism in the structure of Turkic clauses and noun phrases has been generally acknowledged at least since Abney’s (1987) account of nominalizations. Updating Abney’s hypothesis with Chomsky’s (2000) theory of Agree, we can characterize predicate and possessive agreement in Tatar in the following way (see Pereltsvaig and Lyutikova 2014; Lyutikova 2017 for details). The relevant functional head (finite T or possessive D) bearing unvalued phi-features functions as a probe and finds a caseless nominal goal in its c-command domain. The emerging Agree relation yields valuation of the probe’s phi-features and case-licenses the goal nominal. T and D differ as to the structural case licensed: T assigns nominative and D assigns genitive. Finally, the goal is attracted to the specifier of the probe. The same holds for argumental nominalizations, where embedded clausal projections cannot case-license the subject and it enters the Agree relation with D. For denominal postpositions, we can assume a complex internal structure whereby the projection of a lexical head P selects for an argument which is case-licensed (and agreed with) by a functional head p similar to D.13 Agreement configurations are schematically represented in (29). It is important to emphasize that in all the agreement configurations in (29), the agreement target is higher than the controller, which complies with the standardly assumed structural relation between the probe and the goal.

(29)	a. finite clause
	[TP T[uϕ:__] [AspP … DP[iϕ:Val], [uCase:__] …]
	b. possessive noun phrase
	[DP D[uϕ:__] [NumP … DP[iϕ:Val], [uCase:__] …]
	c. argumental nominalization
	[DP D[uϕ:__] [vP … DP[iϕ:Val], [uCase:__] …]
	d. denominal postposition phrase
	[pP p[uϕ:__] [PP … DP[iϕ:Val], [uCase:__] …]

The two constructions which apparently do not comply with this analysis are the PFG construction and the plain postposition construction. They seem to lack possessive agreement (and consequently, the ezafe marker in nominals and its counterpart in denominal postpositions) but still assign genitive case to personal pronouns.

Satık (2020) considers the Turkish PFG as an instance of genitive-marked adjuncts. Indeed, in Turkish, the PFG seems to be restricted to non-argumental uses. However, this is not the case in Tatar, see above. Moreover, the PFG construction can be employed to accommodate the nominalization’s subject, as in (30a). In any case, PFG nominals behave like DP arguments with respect to differential object marking; exactly like standard possessed nominals, they cannot remain caseless and require overt accusative marking, cf. (30b).

(30)	a.	Tik	ul	minem	bügen	monda	kil-ü-ne
		but	this	I.gen	today	here	come-nml-acc
		bel-ergä		tiješ	tügel.
		know-inf		need	neg.cop
		‘But he need not know that I come here today.’ [CWT]
	b.	A-nıŋ		üz-e-neŋ	bala-lar-ı-na	äti	kiräk,
		this-gen		self-3-gen	child-pl-3-dat	father	need
		šuŋa	da	minem	äti-*(ne)	al-dı.
		hence	ptcl	I.gen	father-acc	take-pst
		‘Her own children need (a) father, that is why she took mine.’ [CWT]

Therefore, I conclude that the PFG is rather a specific phonological realization of the standard possessive construction than a separate syntactic construction. I assume that the same logic applies to plain forms of denominal postpositions, though we lack similar diagnostics for PPs. I refrain from formulating a specific PF rule responsible for these phenomena and only subsume them under the generalization that possessive and postpositional constructions can receive a special spell-out when their argument bears a marked person feature.

Though agreement configurations in (29) are structurally parallel, possessive noun phrases are unique featurally. Indeed, possessive DPs have two phi-feature sets: one is its own interpretable phi-feature set which is inherited by DP from the lower nominal projection and the other one is the uninterpretable phi-feature set which is valued via Agree and spelt out on the possessive affix. DPs cannot have two complete phi-feature sets, since an interpretable person feature is only present in indexicals, and they are DP proforms themselves and cannot combine with a possessive D. However, it is possible for a possessive DP to have an interpretable number feature inherited from the Num head.14 In this case, DP will possess two instances of number: interpretable number and agreement-induced uninterpretable number.15

Importantly, it is the interpretable phi-feature set that is employed in the external agreement with a possessive DP, cf. (31). In other agreement configurations, the relevant functional head only has an uninterpretable phi-feature set valued by agreement, which is never used as a source of phi-features by a higher probe.16

(31)	a.	Bez-neŋ	ukıtučı-lar-ıbız	kil-de	/	kil-de-lär	/	*kil-de-k.
		we-gen	teacher-pl-1pl	come-pst		come-pst-pl		come-pst-1pl
		‘Our teachers came.’
	b.	bez-neŋ	ukıtučı-lar-ıbız-nıŋ	kil-ü-e	/	kil-ü-lär-e	/	*kil-ü-ebez
		we-gen	teacher-pl-1pl-gen	come-nml-3		come-nml-pl-3		come-nml-1pl
		‘our teachers’ coming’
	c.	bez-neŋ	ukıtučı-lar-ıbız-nıŋ	kitab-ı	/	kitap-lar-ı	/	*kitab-ıbız
		we-gen	teacher-pl-1pl	book-3		book-pl-3		book-1pl
		‘our teachers’ book’

If the analysis presented above is essentially correct and all the agreement configurations in Tatar are construed in a similar fashion, as (29) depicts, we expect consistent behavior of anaphors across all the agreement configurations; our data show that this is indeed the case. Moreover, the fact that anaphors are excluded from finite subject positions in Tatar cannot be attributed to the AAE but requires an alternative explanation (which will be presented in Section 3).

Furthermore, the properties of agreement in Tatar allow us to exclude analytical options proposed in Satık (2020) for deriving the possible agreement with partitives and the lack of agreement with anaphors in Turkish. The author assumes that when the partitive construction triggers person agreement, which is the case in finite predicate agreement configurations, it is the 1–2p pronoun in the highest specifier of the possessive DP that is the controller of the agreement, and that phi-features of this pronoun are transferred directly to the probe, without any intermediate agreement process. The default agreement pattern with partitives (and anaphors as their subtype) in possessive agreement configurations results from the blocking effect produced by genitive marking. Genitive is assumed to increase the structural complexity of the partitive construction, which, in its turn, makes the possessor’s phi-features inaccessible for external probes. However, the generalization on Tatar agreement runs counter to this hypothesis: in possessive constructions and postpositional phrases, genitive goals trigger agreement whereas nominative/caseless goals do not.

The uniformity of agreement configurations in Tatar and the consistent behavior of anaphors across these configurations suggest that the difference between reflexives and reciprocals with respect to agreement patterns can only be accounted for by drawing on their own characteristics, e.g., their different internal structures, their different feature sets or their different statuses with respect to binding theory. In the next section, we examine the binding-theoretical properties of reflexives and reciprocals and investigate the relation between their anaphoric nature and their internal structure.

3. Tatar Anaphors and Their Binding

In this section, I discuss binding-theoretical properties of Tatar anaphors. To my knowledge, there are no detailed descriptions of the Tatar anaphoric system, let alone its characterization in terms of syntactic and semantic binding. The few relevant works include Shluinsky (2007) on anaphoric dependencies between the matrix and embedded clauses, and Podobryaev (2014) on indexical shift and alternative anaphoric strategies in finite dependent clauses, both based on the Mishar dialect of Tatar. For this reason, I have to present my own findings rather than build on previous literature, though exact and complete characterization of literary Tatar anaphora goes far beyond the purpose of this paper.

Reduplicated reflexives and reciprocals pattern together with respect to a number of properties. Both require a local binder; both are obligatorily bound semantically; both disallow overt expression of the possessor. Let us start with syntactic binding.

First of all, reduplicated reflexives and reciprocals are anaphors; they require a c-commanding antecedent (32a)–(33a). Importantly, the c-command requirement cannot be dispensed with and replaced by linear precedence, cf. (32b)–(33b).

(32)	a.	Kızi	üz-üz-e-ni	fotoräsem-dä	kür-ep
		girl	self-self-3-acc	photograph-loc	see-cvb
		tan-dı.
		recognize-pst
		‘The girl recognized herself on the picture.’
	b.	*Kız-nıŋi	ukıtučı-sı	üz-üz-e-ni	fotoräsem-dä
		girl-gen	teacher-3	self-self-3-acc	photograph-loc
		kür-ep	tan-dı.
		see-cvb	recognize-pst
		Int.: ‘The girl’s teacher recognized her on the picture.’

(33)	a.	Kız-lari	ber-ber-(lär)-e-ni	fotoräsem-dä	kür-ep
		girl-pl	one-one-pl-3-acc	photograph-loc	see-cvb
		tan-dı-(lar).
		recognize-pst-pl
		‘The girls recognized each other on the picture.’
	b.	*Kız-lar-nıŋi	ukıtučı-lar-ı	ber-ber-se-ni	fotoräsem-dä
		girl-pl-gen	teacher-pl-3	one-one-3-acc	photograph-loc
		kür-ep	tan-dı-(lar).
		see-cvb	recognize-pst-pl
		Int.: ‘The girlsi’ teachers recognized themi on the picture.’

The next thing to note is that Tatar reduplicated reflexives and reciprocals are not subject-oriented, i.e., they allow for a non-subject c-commanding antecedent.17 This is illustrated with corpus examples in (34a–b).

(34)	a.	Bez	keše-nei	üz-üz-ei	belän	genä	kal-sa-k, …
		we	man-acc	self-self-3	with	only	leave-cnd-1pl
		‘If we leave a man alone with himself, …’ [CWT]
	b.	Isem-när…		keše-lär-nei		ber-ber-se-nnäni	ajıra-lar.
		name-pl		man-pl-acc		one-one-3-abl	distinguish.ipf-pl
		‘Names … distinguish people from each other.’ [CWT]

Finally, we have to determine the binding domain for reduplicated reflexives and reciprocals. Examples (32)–(34) suggest that it is at least as large as the clause containing the anaphor. To proceed further, we have to determine major types of clause embedding available in Tatar. In what follows, I delimit my study to complement clauses.

There are three major complementation strategies, which employ non-finite nominalized clauses (-U and –gAn), infinitival clauses (-rgA) and finite clauses introduced by the complementizer (dip, digän). Argumental nominalizations do not license nominative subjects; instead, they make use of nominal functional projections hosting possessive agreement and licensing a genitive subject (see Section 2.1 and Section 2.3 above). Infinitival clauses are used in control configurations, with desiderative, implicative and causative verbs, as well as with non-verbal modal predicates (e.g., kiräk ‘need’, tiješ ‘need’); their subject is the controlled PRO.18 Finally, a large class of matrix verbs including verbs of saying, thinking and emotions make use of the finite embedding strategy with the complementizer dip (digän). Finite embedded clauses license their own nominative subject which controls predicate agreement. A peculiar property of many Turkic languages including Tatar is the availability of accusative-marked subjects in finite embedded clauses (Baker and Vinokurova 2010; Baker 2015; Kornfilt and Preminger 2015; Lyutikova and Ibatullina 2015). Accusative subjects, like nominative subjects, control embedded predicate agreement; the only difference is that accusative subjects are only licit at the left edge of the embedded clause, whereas nominative subjects can appear clause-internally.19

The binding domain of reduplicated anaphors can be roughly defined as a minimal clause (finite or non-finite) or a DP containing a subject. This is shown in examples (35) for reduplicated reflexives (for reasons of space, I skip parallel examples for reciprocals); additional corpus examples of both reciprocals and reduplicated reflexives are provided in (36).

(35)	a.	Alsui	Räfik-neŋj	üz-üz-e-nj,*i	kür-gän-e-n	belä.
		Alsu	Rafik-gen	self-self-3-acc	see-pf-3-acc	know.ipf
		‘Alsu knows that Rafik saw himself/*her.’
	b.	Alsui	Räfik-nej	PROj	üz-üz-e-nj,*i	kürsät-ergä	ǯiber-de.
		Alsu	Rafik-acc		self-self-3-acc	show-inf	send-pst
		‘Alsu sent Rafik to show himself/*her.’
	c.	Alsui	Räfik-(ne)j	üz-üz-e-nj,*i	jarata	dip	ujlıj.
		Alsu	Rafik-acc	self-self-3-acc	love.ipf	comp	think.ipf
		‘Alsu thinks that Rafik loves himself/*her.’
	d.	Alsui	Räfik-neŋj	üz-üz-ej,*i	tur-ı-nda-gı	xikejä-se-n	išet-te.
		Alsu	Rafik-gen	self-self-3	about-3-loc-atr	story-3-acc	hear-pst
		‘Alsu heard Rafik’s story about himself/*her.’

(36)	a.	Sini	üz-üz-eŋ-nei	alda-p	jör-gän-eŋ-ä
		you	self-self-2sg-acc	deceive-cvb	go-pf-2sg-dat
		min	gajeple	tügel.
		I	guilty	neg.cop
		‘It is not my fault if you were deceiving yourself.’ [CWT]
	b.	Bezi	ber-ber-ebez-nei	jaxšı	belä-bez	dip
		we	one-one-1pl-acc	well	know-1pl	comp
		ujlıj	i-de-m.
		think.ipf	aux-pst-1sg
		‘I thought that we knew each other well.’ [CWT]
	c.	pro1sg	Alar-nıŋi	ber-ber-sei	tur-ı-nda-gı
			they-gen	one-one-3	about-3-loc-atr
		fiker-lär-e-n		bel-de-m.
		thought-pl-3-acc		know-pst-1sg
		‘I knew their opinion about each other.’

However, if a reduplicated anaphor is itself in the possessor/subject position, its binding domain is extended to the inclusion of another nominal which is a potential binder. Accordingly, the binding domain of the reduplicated anaphor is a minimal clause or a DP containing the anaphor itself and another DP which could serve as a binder.20 Extension of the binding domain can be observed in elicited examples (37) and in corpus examples (38a–c) where the reduplicated anaphor is in the subject/possessor position.

(37)	a.	Alsui	[Räfik-neŋj	[üz-üz-e-neŋj,*i	Kazan-ga	kit-ü-e-n]
		Alsu	Rafik-gen	self-self-3-gen	Kazan-dat	leave-nml-3-acc
		bel-gän-e-n]		sizen-de.
		know-pf-3-acc		feel-pst
		‘Alsu felt that Rafik knew that he/*she was going to Kazan.’
	b.	Kız-lari	jeget-lär-nej	[PROj	[ber-ber-se-neŋj, *i	xikejä-lär-e-n]
		girl-pl	boy-pl-acc		one-one-3-gen	story-pl-3-acc
		tıŋla-rga]	mäǯbür	it-te.
		listen-inf	obliged	do-pst
		‘The girlsi made the boysj listen to each other’sj, *i stories.’

(38)	a.	Bez-neŋ	härkajsı-bızi	üz-üz-e-neŋi	adwokat-ı.
		we-gen	each-1pl	self-self-3-gen	lawyer-3
		‘Each of us is his own lawyer.’ [CWT]
	b.	Šušı	portatiw	fotokamera	belän	keše-läri
		that	handy	camera	with	man-pl
		üz-üz-lär-e-neŋi		közge-dä-ge		čagılıš-ı-n
		self-self-pl-3-gen		mirror-loc-atr		reflection-3-acc
		töšer-ä		i-de-lär.
		take_down.ipf		aux-pst-pl
		‘With that handy camera, people take pictures of their reflection in the mirror.’ [CWT]
	c.	Alari	monda	ber-ber-se-neŋi	ni	belän	jäšä-gän-e-n	belä-lär.
		they	here	one-one-3-gen	what	with	live-pf-3-acc	know.ipf-pl
		‘Here they find out with what each of them lives.’ [CWT]

At the same time, the binding domain of reduplicated anaphors cannot be larger than a minimal finite clause containing the anaphor. Thus, reduplicated anaphors are ungrammatical as finite subjects, either nominative or accusative:21

(39)	a.	*Alsu	[üz-üz-e	/	üz-üz-e-n		Räfik-ne	jaxšı
		Alsu	self-self-3		self-self-3-acc		Rafik-acc	well
		belä	dip]	ujlıj.
		know.ipf	comp	think.ipf
		Int.: ‘Alsu thinks that she knows Rafik well.’
	b.	*Kız-lar	[ber-ber-se	/	ber-ber-se-n		Räfik-ne
		girl-pl	one-one-3		one-one-3-acc		Rafik-acc
		jarata	dip]	aŋla.
		love.ipf	comp	understand.ipf
		Int.: ‘The girls understand that each of them loves Rafik.’
	c.	*Bez	[üz-üz-ebez	/	üz-üz-ebez-ne		ber-ber-ebez-gä
		we	self-self-1pl		self-self-1pl-acc		one-one-1pl-dat
		bulıš-ırga	tiješ	dip]		ujlıj-bız
		support-inf	need	comp		think.ipf-1pl
		Int.: ‘We think that we have to lend support to each other.’22

Therefore, I conclude that reduplicated reflexives and reciprocals pattern together in that they are local syntactic anaphors. The next important property that they share is that they are obligatorily bound semantically in all the positions where they are licit. Examples in (40) show that they do not support a strict interpretation in focused contexts; in (41), the strict reading is excluded under ellipsis:

(40)	a.	Sini	genä	üz-üz-eŋ-nei	kür-ä-seŋ.
		you	only	self-self-2sg-acc	see-ipf-2sg
		‘Only you see yourself.’ (OKsloppy reading, *strict reading)
	b.	Bezi	genä	ber-ber-ebez-neŋi	bala-lar-ı-n	äjt-te-k.
		we	only	one-one-1pl-gen	child-pl-3-acc	invite-pst-1pl
		‘Only we invited each other’s children.’ (OKsloppy reading, *strict reading)

(41)	Alsui	üz-üz-e-neŋi	matur	i-kän-e-n	sanıj,	min	dä.
	Alsu	self-self-3-gen	beautiful	aux-pf-3-acc	consider.ipf	I	ptcl
	‘Alsu considers herself beautiful, and so do I.’ (OKsloppy reading, *strict reading)

The last thing to note is that reduplicated anaphors disallow overt possessors, either nominal or pronominal. Thus, all the combinations listed in (42) are ungrammatical:

(42)	a. nominal possessors:
	*Alsu-nıŋ	üz-üz-e	/	*kız-lar-nıŋ	ber-ber-se	/	ber-ber-lär-e
	Alsu-gen	self-self-3		girl-pl-gen	one-one-3		one-one-pl-3
	b. 3p pronominal possessors:
	*a-nıŋ	üz-üz-e	/	*a-lar-nıŋ	ber-ber-se	/	ber-ber-lär-e
	this-gen	self-self-3		this-pl-gen	one-one-3		one-one-pl-3
	c. 1–2p pronominal possessors:
	*bez-neŋ	üz-üz-ebez	/	*sez-neŋ	ber-ber-egez
	we-gen	self-self-1pl		you-gen	one-one-2pl

The simple reflexive üz-e ‘self-3’ differs from reduplicated anaphors in many respects. First of all, it allows for an overt genitive possessor (minem üz-em ‘I.gen self-1sg’, a-nıŋ üz-e ‘this-gen self-3’, kız-lar-nıŋ üz-(lär)-e ‘girl-pl-gen self-(pl)-3’ etc).23 In this case, it functions as an intensifier (43) and avoids syntactic binding (44).

(43)	a.	At-lar-nı	tap-ma-sa-k,	minem	üz-em-ne
		horse-pl-acc	find-neg-cnd-1pl	I.gen	self-1sg-acc
		al-ıp	kitä-lär	bit.
		take-cvb	leave.ipf-pl	ptcl
		‘If we don’t find horses, they will take me away as well.’ [CWT]
	b.	Sineŋ	üz-eŋ-neŋ	tormoz-ıŋ	ešlä-mä-gän	di-m	min,
		you.gen	self-2g-gen	brakes-2sg	work-neg-pf	say-1sg	I
		belä-seŋ	kil-sä.
		know.ipf-2sg	come-cnd
		‘I say that your own brakes didn’t work properly, if you ask.’ [CWT]
	c.	Ǯir-neŋ	üz-e-nä	dä	köčle	ximikat	daru
		ground-gen	self-3-dat	ptcl	strong	chemical	drug
		sipter-ep	tora-lar.
		pour-cvb	stay.ipf-pl
		‘They pour strong chemical drugs into the soil itself.’ [TT]
	d.	Läkin	min	säbäb-e-n	soraš-ma-dı-m,
		but	I	reason-3-acc	ask-neg-pst-1sg
		Azat-nıŋ	üz-e-neŋ	äjt-kän-e-n	köt-te-m.
		Azat-gen	self-3-gen	tell-pf-3-acc	wait-pst-1sg
		‘But I didn’t ask for an explanation, I waited that Azat would tell (it) himself.’ [TT]

(44)	a.	*Sin/pro2sg	sineŋ	üz-eŋ-ne	kürä-seŋ.
		you	you.gen	self-self-2sg-acc	see.ipf-2sg
		Int.: ‘You see yourself.’
	b.	*Bez/pro1pl	bez-neŋ	üz-ebez-neŋ	bala-bız-nı	äjt-te-k.
		we	we-gen	self-1pl-gen	child-1pl-acc	invite-pst-1pl
		Int.: ‘We invited the child of ours.’

With a non-overt possessor, the simple reflexive üz-e ‘self-3’ has a peculiar behavior. In configurations where the reduplicated reflexive is bound, the simple reflexive can have a c-commanding antecedent, too. In non-subject positions, the antecedent is found within its own clause (45a–b); in non-finite subject position, the binding domain extends up to the next clause, exactly like with reduplicated anaphors (45c). Importantly, in these cases, the simple reflexive can (or, in most local cases, is even strongly preferred to) be semantically bound (46).24 On the other hand, it can be coindexed with a non-local c-commanding antecedent (47) without being semantically bound by it (48). Finally, it can have no antecedent at all (49).

(45)	a.	Sini	eš-tä	üz-eŋ-nei	kürsät-sä-ŋ,	aklana
		you	work-loc	self-2sg-acc	show-cnd-2sg	redeem.ipf
		ala-sıŋ.
		can.ipf-2sg
		‘If you prove yourself in work, you will be able to redeem yourself.’ [CWT]
	b.	Mini	üz-em-neŋi	xatın-ım-nı	häm	ike
		I	self-1sg-gen	wife-1sg-acc	and	two
		ul-ım-nı	üter-de-m.
		son-1sg-acc	kill-pst-1sg
		‘I killed my wife and my two sons.’ [CWT]
	c.	Mini	alar-ga	üz-em-neŋi	ike	operacija	jasat-u-ım-nı
		I	they-dat	self-1sg-gen	two	surgery	perform-nml-1sg-acc
		äjt-ep	karıj-m.
		tell-cvb	look.ipf-1sg
		‘I look at them and tell that I have performed two surgeries.’ [TT]

(46)	a.	Alsui	genä	üz-e-ni	sekcijä-gä	jaz-dır-dı.
		Alsu	only	self-3-acc	section-dat	write-caus-pst
		‘Only Alsu enrolled herself in the sports section.’ (OKsloppy reading, ?*strict reading)
	b.	Bezi	genä	üz-ebez-neŋi	süz-ebez-ne	wlast’-ka
		we	only	self-1pl-gen	word-1pl-acc	authorities-dat
		ǯitker-ergä	tiješ-bez.
		inform-inf	must-1pl
		‘Only we have to communicate our statement to the authorities.’ (OKsloppy reading, ?strict reading)
	c.	mini	üz-em-neŋi	matur	i-kän-em-ne	sanıj-m,
		I	self-1sg-gen	beautiful	aux-pf-1sg-acc	consider.ipf-1sg
		Räfik	dä.
		Rafik	ptcl
		‘I consider myself beautiful, and so does Rafik.’ (OKsloppy reading, okstrict reading)

(47)	a.	[pro3pl	Üz-em-nei	jarat-u-lar-ı]	belän		bäxetle	mini.
			self-1sg-acc	love-nml-pl-3	with		happy	I
		‘I am happy to be loved.’ [TT]
	b.	Mini	berenče	tapkır	[[üz-em-neŋi		öst-em-ä	kil-gän]
		I	first	time	self-1sg-gen		over-1sg-dat	come-pf
		fašist-nıŋ	tilergän	küz-lär-e-n]	kür-de-m.
		fascist-gen	crazy	eye-pl-3-acc	see-pst-1sg
		‘For the first time I saw the crazy eyes of the fascist who stood over me.’ [TT]
	c.	Uli,	[[[üz-ei	jaxšı	dip]		ujla-gan]	berničä	šigır-e-n]
		this	self-3	good	comp		think-pf	several	poetry-3-acc
		bik	tırıš-ıp	ak-ka	küčer-ep,		ber	gazeta-ga
		very	care-cvb	white-dat	copy-cvb		one	newspaper-dat
		bir-ü	öčen	idaräxanä-gä		kit-te.
		give-nml	for	administration_office-dat		leave-pst
		‘He rewrote diligently fair copies of several poetries which he believed to be good and went to the administration office to send (them) to a newspaper.’ [CWT]

(48)	a.	[Äti-m-neŋ	üz-em-nei	Kazan-ga	üz-e	belän	al-gan-ı-n]
		father-1sg-gen	self-1sg-acc	Kazan-dat	self-3	with	take-pf-3-acc
		mini	genä	xäterli-m.
		I	only	remember.ipf-1sg
		‘Only I remember that my father took me to Kazan with him.’ (*sloppy reading, OKstrict reading)
	b.	Mini	genä	[[üz-em-nei	üpkälät-kän]	jeget-tän]	üč	al-dı-m.
		I	only	self-1sg-acc	offend-pf	boy-abl	revenge	take-pst-1sg
		‘Only I took revenge on the guy who offended me.’ (*sloppy reading, OKstrict reading)

(49)	a.	pro3sg		Watan-nı		sakla-rga	bar-ma-sa,
				motherland-acc		defend-inf	go-neg-cnd
		üz-ebez-neŋ		jan-ıbız-da		järdämče	bul-ır.
		self-1pl-gen		near-1pl-loc		assistant	be-fut
		‘If they are not going to defend the motherland, they will be our aide near us.’ [CWT]
	b.	Tatar		jäš-lär-e		üz-ebez-neŋ	matur	jaŋgırašlı
		Tatar		joung-pl-3		self-1pl-gen	beautiful	sonorous
		isem-när-gä		kajta	bašla-dı.
		name-pl-dat		return.ipf	begin-pst
		‘Tatar youth started getting back to our beautiful sonorous names.’ [CWT]

The two opposite patterns—the bound anaphor and semantically free pronominal—suggests that in case of üz-e ‘self-3’, we are dealing with exempt anaphora (Charnavel and Sportiche 2016; Charnavel 2019), whereby the anaphor covers non-reflexive functions, e.g., is used as a logophoric pronoun. Indeed, the logophoric analysis has been proposed for Turkish reflexive kendi-si ‘self-3’ (Kornfilt 2001), which is much like Tatar üz-e in allowing non-local antecedents or antecedent-less configurations. Therefore, it is important to distinguish between purely reflexive and possibly logophoric uses of üz-e.

The standard assumption about logophoricity is that logophors mark reference to the logophoric center of the utterance, which different languages associate with “the source of the report, the person with respect to whose consciousness (or “self”) the report is made, and the person from whose point of view the report is made” (Sells 1987, p. 445). That is, to distinguish between logophoric and reflexive uses, we should consider contexts with non-human antecedents, as suggested in Charnavel and Sportiche (2016); Charnavel (2019); a.m.o.

First of all, both reduplicated and simple reflexive, as well as the reciprocal, allow for (local) non-human antecedents.

(50)	a.	Xäjer,	ul	jarai	bügen	dä	üz-ei	tur-ı-nda
		though	this	wound	today	ptcl	self-3	about-3-loc
		onıt-tır-mıj.
		forget-caus-neg.ipf
		‘Though, this wound still reminds about itself.’ [CWT]
	b.	Bu	ısuli	eš-tä		üz-üz-e-ni	jaxšı	kür-sät-te.
		this	method	work-loc		self-self-3-acc	well	see-caus-pst
		‘This method has proven itself in work.’
	c.	Tarix	bit-lär-e	wakıjga-lar-nıi		ber-ber-se-näi		bäjlä-de.
		history	page-pl-3	event-pl-acc		one-one-3-dat		bind-pst
		‘The pages of history linked the events together.’ [CWT]

As expected, in these configurations simple reflexives are semantically bound:

(51)	Eši	mine	üz-ei	tur-ı-nda	onıt-tır-mıj,	sälamätlek	tä.
	work	I.acc	self-3	about-3-loc	forget-caus-neg.ipf	health	ptcl
	‘Work does not let me forget it, and so does health.’ (oksloppy reading, *?strict reading)

Importantly, in non-local contexts, i.e., in contexts where reduplicated anaphors are disallowed and simple reflexives are not semantically bound, non-human antecedents of simple reflexives are ungrammatical. Compare (52a) with a locally bound reflexive and (52b) with an intended non-local antecedent.

(52)	a.	Bu	problemai		üz-e-neŋi	karaš-ı-n	taläp	itä.
		this	problem		self-3-gen	approach-3-acc	requirement	do.ipf
		‘This problem requires its own approach.’
	b.	*Bu	problemai		bez-neŋ	üz-e-neŋi	karaš-ı-n
		this	problem		we-gen	self-3-gen	approach-3-acc
		kullan-u-ıbız-nı	taläp	itä.
		adopt-nml-1pl-acc	requirement	do.ipf
		Int.: ‘This problem requires that we adopt its (specific) approach.’

Charnavel (2019) argues that apparent antecedent-less uses of logophors can be accounted for under the same lines as long-distance logophors by introducing a logophoric operator in the syntactically represented pragmatic shell of the clause; this operator binds “exempt anaphors”, which derives their logophoric reading. It seems that non-bound (long-distance and antecedent-less) uses of the simple reflexive üz-e can be subsumed under the logophoric pattern too. Indeed, in antecedent-less contexts, we often find 1–2p reflexives, which is expected, since the speech act participants are natural logophoric centers. Moreover, 3p antecedent-less reflexives are attested in free indirect speech contexts like (53).

(53)	Ilšäti	kurka	bašla-dı.
	Ilshat	fear.ipf	start-pst
	Zöläjxa-apa	uz-e-ni	internat-ta	kal-dır-ırga	teli	kebek?
	Zulejxa-aunt	self-3-acc	orphanage-loc	stay-caus-inf	want.ipf	maybe
	‘Ilshat was scared. Maybe aunt Zulejxa will put him into the orphanage?’

Though both anaphors and logophors are bound pronouns under Charnavel’s (2019) approach, we can still distinguish between binding by an antecedent DP and binding by a logophoric operator. In what follows, I consider the exempt anaphors as syntactically free, much like Kornfilt (2001) suggests. Thus, the Tatar simple reflexive allows for both types of uses—syntactically bound and syntactically free.

I believe that this peculiar behavior of the simple reflexive receives a principled explanation under the hypothesis about the internal structure of reflexives put forward in Kornfilt 2001 for Turkish. Kornfilt argues that the Turkish reflexive kendi-si ‘self-3’ “is actually a phrase in disguise” and this phrase, AgrP, hosts the pronominal pro in its specifier (Kornfilt 2001, p. 199). AgrP being a binding domain for pro, pro is trivially free in its binding domain irrespective of its referential index. This allows kendi-si ‘self-3’ to be coindexed with whatever local or non-local antecedent or lack a syntactic antecedent altogether.

Though this elegant hypothesis accounts for the insensitivity of the simple reflexive to syntactic binding, it cannot account for its preferences with respect to semantic binding. Additionally, it does not predict any difference between the behavior of null and overt anaphoric pronouns; however, the former support semantic binding whereas the latter disallow it, cf. (54).

(54)	a. null pro: semantic binding
	Bez	genä	pro1pl	üz-ebez-neŋ	süz-ebez-ne	wlast’-ka
	we	only		self-1pl-gen	word-1pl-acc	authorities-dat
	ǯitker-ergä	tiješ-bez.
	inform-inf	must-1pl
	‘Only we have to communicate our statement to the authorities.’ (OKsloppy reading, ?strict reading)
	b. overt pronoun: no semantic binding
	Bez	genä	bez-neŋ	üz-ebez-neŋ	süz-ebez-ne	wlast’-ka
	we	only	we-gen	self-1pl-gen	word-1pl-acc	authorities-dat
	ǯitker-ergä	tiješ-bez.
	inform-inf	must-1pl
	‘Only we have to communicate our statement to the authorities.’ (*sloppy reading, OKstrict reading)

(55)	a. null pro: semantic binding
	Räfik	kenä	jılan-nı	pro3sg	üz-e	jan-ı-nda	kür-de.
	Rafik	only	snake-acc		self-3	near-3-loc	see-pst
	‘Only Rafik saw a snake near himself.’ (OKsloppy reading, ?strict reading)
	b. overt pronoun: no semantic binding
	Räfik	kenä	jılan-nı	a-nıŋ	üz-e	jan-ı-nda	kür-de.
	Rafik	only	snake-acc	this-gen	self-3	near-3-loc	see-pst
	‘Only Rafik saw a snake near him.’ (*sloppy reading, OKstrict reading)

Therefore, I propose that Tatar pro comes in two binding-theoretical varieties: as an anaphor and as a pronominal. The idea that the possessor of the self-reflexive is an actual anaphor has been successfully exploited by Iatridou (1988) in accounting for the agreement properties of Greek reflexives revealed in clitic doubling, cf. (56). The clitic pronoun shows agreement with the direct object, allegedly violating AAE, but in fact, Iatridou argues, it is the possessive pronoun which is an anaphor. It co-varies with its binder for phi-features, whereas the reflexive phrase is invariably 3p singular masculine.

(56)	a.	I	Maria	ton	thavmazi	ton
		the.nom.f.sg	Maria	cl.acc.m.sg	admire.prs.3sg	det.acc.m.sg
		eafton	tis.
		self	her(gen.f.sg)
		‘Maria admires herself.’ (Iatridou 1988:(9a))
	b.	Egho	ton	xero	ton	eafton	mu.
		I	cl.acc.m.sg	know.prs.1sg	det.acc.m.sg	self	my(gen.1sg)
		‘I know myself.’ (Iatridou 1988:(9b))

Importantly, pro as a pronominal and pro as an anaphor have different binding domains. The pronominal pro’s binding domain is a minimal clause or DP with its own subject which contains pro. As suggested by Kornfilt (2001), this is the reflexive phrase itself. When pro is an anaphor, its binding domain extends as to the inclusion of a potential binder, but this extension cannot go beyond a minimal finite clause. Consequently, in non-local domains, the pro-anaphor is excluded, whereas the pronominal pro is available, and these uses are responsible for the exempt anaphora.25 In local configurations, both varieties of pro are available.26

The twofold characterization of pro as anaphor or pronominal is a descriptive generalization allowing us to capture properties of simple reflexives with respect to semantic binding. However, in view of minimalist premises, it is highly desirable to eliminate binding-theoretical notions such as anaphor or pronominal from the list of primitives and to explain their specific distribution and interpretation by using mechanisms independently required in the grammar. Accordingly, I am going to make the next step and assume a valuation-based difference between anaphors and pronominals: anaphors possess unvalued phi-feature sets whereas pronominals have valued phi-feature sets. In doing so I, follow the appealing approach in the minimalist research seeking to derive binding from a general Agree operation (Reuland 2005; Heinat 2008; Kratzer 2009; Rooryck and Vanden Wyngaerd 2011; Wurmbrand 2017; Murphy and Meyase 2022; Paparounas and Akkuş, forthcoming, a.m.o.). The basic idea is that referential deficiency of anaphors follows from their featural deficiency. The anaphor enters the derivation with unvalued phi-features, which are then valued under agreement (immediate or mediated) with its antecedent, and the relation between the anaphor and the source of phi-features is interpreted as binding at LF. Semantic binding is then a hallmark of Agree-based valuation of the pronoun’s phi-features; therefore, wherever we observe a bound interpretation of the pronoun, we are dealing with agreement. A free interpretation of the pronoun signals that it entered the derivation with valued phi-features.

Thus, I assume that in Tatar, two varieties of pro are available—pro with valued phi-features and pro with unvalued phi-features (57a–b). Unlike pro, overt pronouns only have valued phi-feature sets (57c–d).

(57)	a. pro [ϕ:Val]: null pronouns
	b. pro [ϕ:__]: null anaphor
	c. min [ϕ:Val]: overt 1sg pronoun
	d. *min [ϕ:__]: overt anaphor

This assumption is supported not only by the distribution and interpretation of reflexives, but also by the interpretation of pronouns outside the reflexive context. Let us consider the configuration where the pronoun is the direct object’s possessor.

(58)	a.	Min	genä	minem	bala-m-nı	jarata-m.
		I	only	I.gen	child-1sg-acc	love.ipf-1sg
		‘Only I love my child.’ (*sloppy reading, OKstrict reading)
	b.	Min	genä	pro1sg	bala-m-nı	jarata-m.
		I	only		child-1sg-acc	love.ipf-1sg
		‘Only I love my child.’ (OKsloppy reading, OKstrict reading)

We observe that the overt possessor only supports a strict reading under coindexing, whereas pro is compatible with both interpretations.

Another configuration where overt and non-overt pronouns differ is the indirect speech context introduced by the complementizer dip. Literary Tatar, as well as its Mishar dialect (see Podobryaev 2014) exhibit optional indexical shift. Importantly, only non-overt indexicals (i.e., pro) can shift, cf. (59a–b).27

(59)	a.	Alsu	[min	kaja	kit-te-m	dip]	äjt-te?
		Alsu	I	where	leave-pst-1sg	comp	tell-pst
		‘Which place did Alsu say I went?’					<non-shifted>
		*‘Which place did Alsu say she went?’					<shifted>
	b.	Alsu	[pro1sg	kaja	kit-te-m	dip]	äjt-te?
		Alsu		where	leave-pst-1sg	comp	tell-pst
		‘Which place did Alsu say I went?’					<non-shifted>
		‘Which place did Alsu say she went?’					<shifted>

Though theoretical accounts of indexical shift do not assume that shifted indexicals have unvalued phi-features, they rely on the idea that shiftable indexicals can (or even must, if the shift is obligatory) be bound by a monster operator, whereas non-shiftable indexicals cannot (Schlenker 1999, 2003; Anand and Nevins 2004; Anand 2006; Deal 2020). In principle, unifying indexical shift and variable binding in Tatar as instances of the same process of agreement resulting in valuation of pro’s features does not seem a priori infeasible (of course, this approach cannot be easily extended to languages like Turkish where overt pronouns cannot be bound but can shift); however, I am not going to delve into this issue any further and only emphasize that in Tatar, overt and non-overt pronouns retain their asymmetry with respect to binding in shifting-licensing contexts.

Thus, I conclude that the properties of simple reflexives are determined by the properties of the DP in their highest specifier: with R-expressions, they function as intensifiers; with overt pronouns, they can be coindexed with c-commanding antecedents but not be bound by them; with non-overt pronouns, they function as bound anaphors in local contexts and as non-bound pronouns elsewhere (60a).

Now let us get back to reduplicated reflexives and reciprocals. Recall that they (i) cannot have an overt genitive possessor (*minem üz-üz-em, *alarnıŋ ber-ber-se) and (ii) should be syntactically and semantically bound within their binding domain. I believe that their properties can be accounted for if we assume that reduplicated reflexives and reciprocals lexically select for a featurally unspecified pro.28 If so, reduplicated reflexives should pattern with simple reflexives containing pro [ϕ:__] in their specifier (60b). Our data suggest that this is indeed the case.

(60)	a. simple reflexive
	[RE [ϕ:Val] üz-e]	intensifier only (syntactically free)
	[minem [ϕ:Val] üz-em]	syntactically free or bound, semantically non-bound
	[pro [ϕ:Val] üz-e]	syntactically free or bound, semantically non-bound
	[pro [ϕ:__] üz-e]	syntactically and semantically bound
	b. reduplicated reflexive
	*[RE [ϕ:Val] üz-üz-e]
	*[minem [ϕ:Val] üz-üz-em]
	*[pro [ϕ:Val] üz-üz-e]
	[pro [ϕ:__] üz-üz-e]	syntactically and semantically bound

The discussion above was intended to clarify the internal structure of anaphors and its contribution to their binding-theoretical status. The rationale behind this research program was the search for differences between reflexives and reciprocals which would account for their different properties with respect to external agreement: reflexives exhibit the person agreement pattern whereas reciprocals exhibit the default agreement pattern. In view of the AAE, a possible solution would be that reciprocals are anaphors whereas reflexives are not.

The study revealed a completely different state of affairs. Reduplicated reflexives and reciprocals pattern together with respect to their structure, their syntactic distribution and their (obligatorily bound) interpretation. Moreover, the simple reflexive behaves uniformly as to the external agreement, i.e., exhibits the person agreement pattern, irrespective of its syntactic and semantic properties. This leads me to conclude that in Tatar, presence/absence of the full agreement on the external probe has nothing to do with the anaphor/non-anaphor status of the goal.

Another important conclusion is that intensifiers and syntactically free reflexives, being structurally identical to bound reflexives but licit in the finite subject position, exhibit the person agreement pattern in all the configurations construed with agreement, including finite predicate agreement with its nominative subject. Therefore, there is nothing special in the finite subject position as to the external agreement—it disallows anaphors because their own phi-features cannot be valued in this position.

Thus, the only reliable difference between reflexives and intensifiers, on the one hand, and reciprocals, on the other hand, is their lexical base: reflexives and intensifiers are built on the basis of the noun üz ‘self’, reciprocals make use of the numeral/indefinite ber ‘one’. In the next section, I examine agreement patterns available with partitives employing various quantifiers as their lexical base and try to draw a generalization relating these parameters.

4. Agreement with Inflected Quantifiers

As indicated earlier, Tatar reflexive and reciprocal pronouns form a single structural class with inflected quantifiers. Turkic languages in general make use of the nominal possessive construction to build partitives where the subset is denoted by the NP-internal material, to the exclusion of the lexical noun (numerals, quantifiers, adjectives), and the superset is expressed by the optional genitive possessor cross-referenced in possessive agreement. In Tatar, all these types of bases are licit in partitives, cf. (61). In what follows, I focus on partitives based on numerals, interrogatives, universal and existential quantifiers, which constitute closed classes of elements. For the sake of space, I dub all of them (inflected) quantifiers.

(61)	a. numeral
	Al	inde	sin	bez-neŋ	ike-bez-ne.
	take.imp	ptcl	you	we-gen	two-1pl-acc
	‘Take two of us with you.’ [CWT]
	b. quantifier
	Bügen	barı-gız-nı	da	su	buj-ı-na	čakıra-m.
	today	all-2pl-acc	ptcl	water	along-3-dat	invite.ipf-1sg
	‘Today I invite all of you to the quayside.’ [TT]
	c. adjective
	Kart	ujlıj	tor-gač, …	säbäp-lär-neŋ	iŋ	akıllı-sı-n
	old_man	think.ipf	stay-tmp	reason-pl-gen	most	smart-3-acc
	ujla-p	tap-tı.
	think-cvb	find-pst
	‘The old man pondered and came up with the smartest reason.’ [CWT]

Let us examine agreement patterns available with inflected quantifiers. In Lyutikova and Grashchenkov (2019); Lyutikova (2022), it is argued that both the person agreement pattern and the default agreement pattern are attested in all the agreement configurations. Thus, Tatar seems to be like Kyrgyz, Sakha and Altai in that it allows for both patterns with inflected quantifiers.

However, a more fine-grained study reveals that specific quantifiers tend to favor specific agreement patterns. A corpus study suggests that non-distributive universal quantifiers barı da ‘all’, böten ‘whole, all’, as well as collective numerals ike dä ‘both’, öč dä ‘all three’, etc., can form agreeing partitive constructions (62).

(62)	a. finite predicate
	Böten-egez	šul	sorau-nı	birä-sez.
	all-2pl	this	question-acc	give.ipf-2pl
	‘All of you ask this question.’ [CWT]
	b. nominalization
	Ike-bez-neŋ	dä	gomer-ebez-dä	ber-enče	märtäbä
	two-1pl-gen	ptcl	life-1pl-loc	one-ord	time
	xämer	äč-ü-ebez	bit!
	alcohol	drink-nml-1pl	ptcl
	‘It is the first time in our life that we both drink alcohol!’ [CWT]
	c. possessive construction
	Kil-äčäk-tä	barı	da	bertigez,	bar-ıbız-nıŋ
	come-fut-loc	all	ptcl	equal	all-1pl-gen
	bala-lar-ıbız	da	urıs	bul-ačak.
	child-pl-1pl	ptcl	Russian	be-fut
	‘In the future, all will be equal, children of all of us will be Russian.’ [CWT]
	d. postpositional phrase
	Šun-nan	soŋ	ir-em	tagın	botka	pešer-de
	this-abl	after	husband-1sg	again	porridge	cook-pst
	ike-bez-neŋ	aldı-bız-ga	da	kuj-dı.
	two-1pl-gen	before-1pl-dat	ptcl	put-pst
	‘Then, my husband cooked porridge again and placed it in front of both of us.’ [TT]

Note also that non-distributive universal inflected quantifiers participate in PFG constructions (63a) and combine with plain denominal postpositions (63b). It is also significant that they receive genitive case as postpositions’ arguments, which is a hallmark of nominals bearing a marked person feature.

(63)	a.	Ike-bez-neŋ	dä	äti-lär	sugıš-ta.
		two-1pl-gen	ptcl	father-pl	war-loc
		‘Fathers of us both are (serving) in the war.’ [TT]
	b.	Duslık	bar-ıbız-nıŋ	ara-da	da	bar	dip
		friendship	all-1pl-gen	between-loc	ptcl	cop	comp
		ujla-dı	ul.
		think-pst	this
		‘He thought that there was friendship between all of us.’ [CWT]

At the same time, the default agreement pattern is also attested:

(64)	a. finite predicate
	Sineŋ	matur	ǯırla-gan-ıŋ-nı	barı-bız	da	belä.
	you.gen	beautiful	sing-pf-2sg-acc	all-1pl	ptcl	know.ipf
	‘We all know that you sing beautifully.’ [CWT]
	b. nominalization
	Operacija	wakıt-ı-nda	sez-neŋ	ike-gez-neŋ	assistentlık
	surgery	time-3-loc	you-gen	two-2pl-gen	assistance
	it-ü-e-n	dä	šart	it-ep	kuj-dı.
	make-nml-3-acc	ptcl	condition	make-cvb	put-pst
	‘He set a condition that during the surgery, both of you should assist him.’ [CWT]
	c. possessive construction
	Ärmänstan-da	ǯir	teträ-gän-nän	soŋ	böten-egez-neŋ
	Armenia-loc	earth	quake-pf-abl	after	all-2pl-gen
	adres-lar-ı-n	jugalt-tı-m.
	address-pl-3-acc	lose-pst-1sg
	‘After the earthquake in Armenia, I have lost addresses of all of you.’ [CWT]
	d. postpositional phrase
	Xuǯalık	bülag-e	mödir-ebez	barı-bız	ald-ı-na	da
	household	good-3	manager-1pl	all-1pl	before-3-dat	ptcl
	berär	čemetem	toz	sal-ıp	čıga.
	one	pinch	salt	pour-cvb	exit.ipf
	‘Our hardware store manager poured a pinch of salt in front of all of us.’ [CWT]

Native speakers of Tatar prefer the person agreeing pattern in all agreement configurations but report that the default pattern is also acceptable. The collective vs. distributive distinction does not seem to play a role in choosing agreement pattern, cf. (66).

(65)	a.	Barı-gız	da	miŋa	bulıš-ır-(sız)		inde?
		all-2pl	ptcl	I.dat	help-fut-2pl		q
		‘You will help me, all of you, won’t you?’
	b.	Öjrän-ü	böten-ebez-neŋ	teläg-ebez	/	teläg-e		i-de.
		study-nml	all-1pl-gen	wish-1pl		wish-3		aux-pst
		‘The wish of all of us was to study.’

(66)	a.	Barı-bız	da	izrä-p	jokla-p	kit-te-(k).
		all-1pl	ptcl	doze-cvb	sleep-cvb	leave-pst-1pl
		‘We all dozed off and fell asleep.’
	b.	Barı-bız	da	kitap	kibet-e-ndä	očraš-tı-(k).
		all-1pl	ptcl	book	store-3-loc	meet-pst-1pl
		‘We all met in the book store.’

Distributive universal quantifiers härber ‘each’, härkem ‘everyone, each’, härkajsı ‘whatever, each’ can be distinguished from non-distributive quantifiers in that they do not support the collective reading, cf. (67). They generally form non-agreeing partitive constructions, which exhibit the default agreement pattern across all the agreement contexts; see elicited examples in (68) and corpus examples in (69). Furthermore, the PFG construction is not attested, and denominal postpositions are only licit in their agreeing form. Note also that postpositions combine with a nominative (caseless) form of the inflected quantifier, cf. (68b), (69d).

(67)	a.	Härber-ebez	izrä-p	jokla-p	kit-te-(*k).
		each-1pl	doze-cvb	sleep-cvb	leave-pst-1pl
		‘Each of us dozed off and fell asleep.’
	b.	*Härber-ebez	kitap	kibet-e-ndä	očraš-tı-(k).
		each-1pl	book	store-3-loc	meet-pst-1pl
		Int.: ‘We all met in the book store.’

(68)	a.	Bez-neŋ	härber-ebez	sez-gä	bulıš-ırga	teli-(?*bez)
		we-gen	each-1pl	you-dat	help-inf	want.ipf-1pl
		‘Each of us wants to help you.’
	b.	Ukıtučı-bız	a-lar-ga	härber-ebez	tur-ı-nda	/	*härber-ebez-neŋ
		teacher-1pl	this-pl-dat	each-1pl	about-3-loc		each-1pl-gen
		tur-ıbız-nda	söjle-de.
		about-1pl-loc	tell-pst
		‘Our teacher told them about each of us.’

(69)	a. finite predicate
	Sport	belän	šögıl’län-ü	sälamätlek-kä	uŋaj
	sport	with	exercise-nml	health-dat	positive
	täjesir	itä-čä	genä	häkem-ebez	ıšana.
	influence	make.ipf-agt	only	every-1pl	believe.ipf
	‘Each of us believes that sporting activities only have a positive effect on one’s health.’ [CWT]
	b. nominalization
	Härber-ebez-neŋ	ig’tibarlı	häm	ujau	bul-u-ı	kiräk.
	each-1pl-gen	attentive	and	vigilant	be-nml-3	necessary
	‘Each of us should be attentive and vigilant.’ [CWT]
	c. possessive construction
	Kuj-ıl-gan	maksat-lar-ga	ireš-ü	härkajsı-bız-nıŋ	uŋıš-ı-na
	put-pass-pf	goal-pl-dat	achieve-nml	each-1pl-gen	success-3-dat
	bäjle.
	dependent
	‘Realization of our goals depends on success of each of us.’ [CWT]
	d. postpositional phrase
	Bu	ǯır	tatar	xalk-ı-nıŋ	ačı	jazmıš-ı		tur-ı-nda,
	this	song	Tatar	people-3-gen	bitter	fate-3		about-3-loc
	bu	ǯır	Ilham-nıŋ	üz-e	tur-ı-nda,
	this	song	Ilham-gen	self-3	about-3-loc
	bu	ǯır	härber-ebez	tur-ı-nda.
	this	song	each-1pl	about-3-loc
	‘This song is about the bitter fate of Tatar people, this song is about Ilham himself, this song is about each of us.’ [CWT]

Existential quantifiers containing ber ‘one’—berničä ‘some’, berkadär ‘several’, bernikadär ‘several’, bereü ‘one, alone’, beraz ‘a few’, berär ‘certain, one’, kajber ‘a few’—pattern with distributive universal quantifiers in that they form partitive constructions which trigger the default agreement pattern exclusively, (70)–(71). Existential inflected quantifiers based on ber ‘one’ and berlär ‘ones’ do not attest the person agreement pattern either, cf. (72)–(73).

(70)	a.	Berničä-bez	a-ŋa	karšı	bul-gan-(nar)	/	*bul-gan-ıbız.
		some-1pl	this-dat	against	be-pf-pl		be-pf-1pl
		‘Some of us were against it.’
	b.	Bernikadär-egez-neŋ	miŋa	jardäm	it-ü-e-n	/	*it-ü-egez-ne
		several-2pl-gen	I.dat	help	do-nml-3-acc		do-nml-2pl-acc
		teli-m.
		want.ipf-1sg
		‘I want some of you to lend me support.’

(71)	a.	Berničä-bez	jarıš-ta	katnaša	al-ma-dı.
		some-1pl	competition-loc	participate.ipf	can-neg-pst
		‘Some of us could not participate in the competition.’ [CWT]
	b.	Berničä-gez-neŋ	ǯawab-ı-n	tıŋla-p	ütä-r-bez.
		some-2pl-gen	answer-3-acc	listen-cvb	fulfill-fut-1pl
		‘We will pay attention to the answer of one of you.’ [CWT]
	c.	Kemder	šunduk	ül-de,	bernikadär-ebez	su-ga
		some	at_once	die-pst	several-1pl	water-dat
		bar-ıp	töš-te.
		go-cvb	fall-pst
		‘Some people died at once, some of us fell into the water.’ [CWT]
	d.	Moŋa	bereü-lär-ebez-neŋ	üz-lär-e-n	genä	dahi	sana-p,
		here	one-pl-1pl-gen	self-pl-3-acc	only	genius	believe-cvb
		Parnas	taw-ı	tübä-se-ndä	üz-lär-e	genä
		Parnassus	mount-3	under-3-loc	self-pl-3	only
		utır-ırga	čamala-u-lar-ı…
		sit-inf	suppose-nml-pl-3
		‘Some of us suppose that only they are geniuses and sit on Parnassus…’ [CWT]
	e.	Ä	kajber-lär-ebez	ber	aša-u-da	aša-p	beter-de.
		and	a_few-pl-1pl	one	eat-nml-loc	eat-cvb	finish-pst
		‘And a few of us ate (it) up at one time.’ [CWT]
	f.	Beraz-ıgız	gına	zatlı	näsel-dän	i-de.
		a_few-2pl	only	noble	origin-abl	aux-pst.
		‘Only a few of you are of a noble origin.’ [CWT]
	g.	Jaŋa	jıl-nı	berär-ebez-neŋ	öj-e-ndä	karšı	ala-bız.
		new	year-acc	one-1pl-gen	home-3-loc	meet	take.ipf-1pl
		‘We celebrate the New Year at the house of one of us.’ [CWT]

(72)	a.	Ara-bız-dan	ber-ebez	genä	tik	jata.
		between-1pl-loc	one-1pl	only	ptcl	lie.ipf
		‘Among us, only one is lying.’ [TT]
	b.	Äti	ber-ebez-neŋ	genä	bul-sa	da	Kazan-ga
		father	one-1pl-gen	only	be-cnd	ptcl	Kazan-dat
		kajt-u-ı-n	telä-de.
		return-nml-3-acc	want-pst.
		‘Father wanted that anyone of us returned to Kazan.’ [TT]
	c.	Ike-bez	jal	itä,	ber-ebez	kara-p	jata.
		two-1pl	rest	make.ipf	one-1pl	look-cvb	lie.ipf
		‘Two of us take a rest, one of us keeps watching.’ [CWT]
	d.	Ber-lär-ebez	jaza,	gazeta	čıgara,	ike-nče-lär-ebez
		one-pl-1pl	write.ipf	newspaper	publish.ipf	two-ord-pl-1pl
		isä	a-nı	tarata.
		ptcl	this-acc	distribute.ipf
		‘Some of us write, publish the newspaper, others distribute it.’ [CWT]
	e.	Sez-neŋ	ber-egez	dä	šul	süz-neŋ
		you-gen	one-2pl	ptcl	this	word-gen
		mäg’nä-se-n	bel-mi	bit.
		meaning-3-acc	know-neg.ipf	ptcl
		‘None of you knows the meaning of this word.’ [CWT]

(73)	a.	Šu-nıŋ	arka-sı-nda	ber-ebez	isän	kal-dı-(*k).
		this-gen	because-3-loc	one-1pl	untact	stay-pst-1pl
		‘Because of him, one of us survived.’
	b.	Ber-lär-egez-neŋ	äti-äni-se	/	äti-äni-lär-e	/
		one-pl-2pl-gen	father-mother-3		father-mother-pl-3
		*äti-äni-gez	jardäm	itä	ala.
		father-mother-2pl	help	do.ipf	take.ipf
		‘Parents of some of you can help.’

Interestingly, when existential inflected quantifiers have a narrow scope with respect to negation, they can trigger the person agreement pattern as well (74). I believe that this behavior of ber ‘one’ and hičber ‘any’ is attributable to their NPI status. Indeed, hičber ‘any’ is only licensed under negation, and ber ‘one’ is ambiguous between the PPI and NPI readings. It is in the latter case that ber ‘one’ gives rise to the agreeing inflected quantifier.

(74)	a.	Alla-ga	šöker,	ber-ebez-neŋ	dä	uka-bız	koj-ıl-ma-dı.
		Allah-dat	thank	one-1pl-gen	ptcl	lace-1pl	pour-pass-neg-pst
		‘Thank God, the lace didn’t get damaged on anyone of us.’ [CWT]
	b.	Läkin	ber-ebez-neŋ	dä	uj-lar-ıbız,	xıjal-lar-ıbız
		but	one-1pl-gen	ptcl	thought-pl-1pl	dream-pl-1pl
		tormıš-ka	aš-ma-dı.
		life-dat	realize-neg-pst
		‘But dreams of none of us come true.’ [CWT]
	c.	Ul	kön-ne	karaŋgı	töš-käč	tä,	hičber-ebez
		this	day-acc	dusk	fall-tmp	ptcl	any-1pl
		ker-ep	jat-ma-dı-k.
		enter-cvb	lie-neg-pst-1pl
		‘On this day, when dusk fell, none of us went to bed.’ [CWT]
	d.	231-lık	tur-ı-nda	hičber-egez	ber	süz	äjt-mi-sez.
		231-atr	about-3-loc	any-2pl	one	word	tell-neg.ipf-2pl
		‘About the 231st, no one of you says a word.’ [CWT]

Interrogatives in the partitive constructions are represented by kajsı ‘which’ and ničä ‘how many, how much’.29 Inflected quantifiers involving these elements are usually attested with default external agreement, cf. (75a–d); however, when used in rhetorical questions, they can support the person agreement pattern, cf. (75e).

(75)	a.	Kajsı-bız	adäm-neŋ	kijem-e-n	saldıra	al-ır?
		which-1pl	Adam-gen	suit-3-acc	take_off.ipf	take-fut
		‘Which of us can take off his birthday suit?’ [CWT]
	b.	Kajsı-bız-nıŋ	garaž-ı-na	jäšerä-bez?
		which-1pl-gen	garage-3-dat	hide.ipf-1pl
		‘In whose garage shall we hide?’ [CWT]
	c.	Inde	ničä-bez	kit-ep	bar-dı!
		already	how_many-1pl	leave-cvb	go-pst
		‘How many of us are already gone!’ [TT]
	d.	Ničä-gez-neŋ	art-ı-nda	ǯinajät’	eš-e	bujınča
		how_many-2pl-gen	behind-3-loc	criminal	case-3	for
		stat’ja	bula?
		article	be.ipf
		‘How many of you faced criminal charges?’ (Lit. How many of you have an article (of the Criminal Code) for a criminal case behind?) [CWT]
	e.	Kajsı-bız-nıŋ	satučı-dan	produkcija	sostav-ı-nda	GMO
		which-1pl-gen	seller-abl	production	content-3-loc	GMO
		komponent-lar-ı	bul-u-bul-ma-u	tur-ı-nda	sora-gan-ıbız	bar?
		component-pl-3	be-nml-be-neg-nml	about-3-loc	ask-pf-1pl	cop
		‘Which of us asks the seller about the presence of GMO components in the ?’ [CWT]

Finally, let us turn to numerals and other quantity denoting modifiers—küp ‘many, much’ and az ‘little, not much’. They exhibit variation as to agreement patterns attested with corresponding inflected quantifiers, and this variation is semantically non-vacuous. As corpus data suggest, inflected numerals like ikebez ‘two of us’ can have two interpretations: indefinite interpretation (two individuals out of the definite set ‘us’, the “canonical partitive”, according to Falco and Zamparelli 2019) and definite “appositive” interpretation (the definite set ‘us’ consisting of two individuals, ‘we two’, the “maximal pronominal partitive” in the typology of Falco and Zamparelli 2019). The same holds for quantity denoting küp ‘many, much’ and az ‘little, not much’: they derive both indefinite partitives (‘many/not many of us’) and definite appositives (‘we, which are many/not many’). Importantly, the agreeing pattern is attested with the definite “appositive” interpretation, cf. corpus examples (76)–(77) and elicited examples (78) constituting a minimal pair.

(76)	a.	Dürt-ebez	ber	bülmä-dä	jäšä-de-k.
		four-1pl	one	room-loc	live-pst-1pl
		‘We four lived in one room.’ [CWT]
	b.	Dürt-ebez-neŋ	ber-ebez	dä	čišä	al-mıj-bız.
		four-1pl-gen	one-1pl	ptcl	solve.ipf	take-neg.ipf-1pl
		‘No one of us four can solve (it).’ [CWT]
	c.	Süz	ike-bez-neŋ	ara-bız-da	kal-ır…
		word	two-1pl-gen	between-1pl-loc	stay-fut
		‘That stays between us…’ [CWT]
	d.	Annarı	min	ike-bez-neŋ	awıl-ga	kajt-u-ıbız-nı
		then	I	two-1pl-gen	village-dat	return-nml-1pl-acc
		küz	ald-ı-na	kiter-ergä	tırıša-m.
		eye	before-3-dat	bring-inf	try.ipf-1sg
		‘Then I try to imagine that we two return to the village.’ [CWT]
	e.	Tanılgan	ǯırčı-nıŋ	tawıš-ı-n	küb-ebez-neŋ	grammofon
		famous	singer-gen	voice-3-acc	many-1pl-gen	gramophone
		häm	magnitofon	jazma-lar-ı-nda	tıŋla-gan-ıbız	bar.
		and	tape-recorder	record-pl-3-loc	listen-pf-1pl	cop
		‘Many of us listened to the gramophone and tape voice recordings of the famous singer.’ [CWT]
	f.	Küb-ebez-neŋ	küz-lär-ebez-dä	kurku	katıš	šik-šöbhä.
		many-1pl-gen	eye-pl-1pl-loc	fear	mixed	doubt-hesitation
		‘There is fear and confusion in our eyes.’ [CWT]
	g.	Bik	az-ıbız	gına	kač-ıp	kotıla	al-dı-k.
		very	not_many-1pl	only	run-cvb	escape.ipf	can-pst-1pl
		‘Very few of us managed to escape.’ [CWT]

(77)	a.	Kal-gan	öč-ebez	tugız-ınčı	häm	un-ınčı
		stay-pf	three-1pl	nine-ord	and	ten-ord
		sıjnıf-lar-da	uk-ıp	jöri.
		grade-pl-loc	study-cvb	go.ipf
		‘Other three of us are studying in grade 9 and 10.’ [CWT]
	b.	Kal-gan	öč-ebez-neŋ	küz-lär-e	maŋgaj-ga	men-gän	i-de
		stay-pf	three-1pl-gen	eye-pl-3	forehead-dat	rise-pf	aux-pst
		šul	čak-ta.
		this	time-loc
		‘Meanwhile, the eyes of other three of us popped out of their heads’. [CWT]
	c.	Kit-te	bez-neŋ	ike-bez,	ničäü	kal-dı-k
		leave-pst	we-gen	two-1pl	how_many	stay-pst-1pl
		xäzer	bez?
		now	we
		‘Two of us have gone away, how many of us are there now?’ [CWT]
	d.	Öč-ebez	mäktäp-tä	ukıj,	ike-bez	xezmät-tä	jözmäk-tä.
		three-1pl	school-loc	study	two-1pl	service-loc	swimming-loc
		‘Three of us go to school; two of us are in service, out to sea.’ [CWT]
	e.	Küb-ebez	praktika-ga	akademijä-neŋ	üz-e-ndä	kal-dı.
		many-1pl	practice-dat	academy-gen	self-3-loc	stay-pst
		‘Many of us stayed at the Academy itself for practice.’ [CWT]
	f.	Axır-ga	kadär	bik	az-ıbız	gına	bar-ıp	ǯit-te.
		end-dat	up_to	ptcl	few-1pl	only	go-cvb	reach-pst
		‘Only few of us reached the end.’ [CWT]

(78)	a. Context: Our mother has three children.
	Öč-ebez	(dä)	mäktäp-tä	ukıj-bız	/	*?ukıj.
	three-1pl	ptcl	school-loc	study.ipf-1pl		study.ipf
	‘We three go to school.’
	b. Context: Our mother has five children.
	Öč-ebez	mäktäp-tä	ukıj	/	*ukıj-bız,	kal-gan	ike-bez
	three-1pl	school-loc	study.ipf		study.ipf-1pl	stay-pf	two-1pl
	zavod-ta	ešli	/	*ešli-bez.
	factory-loc	work.ipf		work.ipf-1pl
	‘Three of us go to school, the other two work in a factory.’

Interestingly, Paparounas and Akkuş (forthcoming) report that Turkish allows for both agreement patterns with inflected numerals as well, but the choice between the person agreement pattern and default agreement pattern is determined by clusivity. Specifically, the person agreement pattern is employed if the speaker is included in the subset (which can be a proper part of the superset), and the default agreement pattern is not specified with respect to the inclusion of the speaker. It is not clear how exactly this generalization is extended to a 2pl superset; I suppose that inclusion of the addressee is relevant in this case. An anonymous reviewer thus wonders whether Tatar data allow for the same generalization.

In fact, it is not easy to provide a context which would distinguish between the two generalizations. When the appositive interpretation occurs, the speaker (or the addressee with a 2pl superset) is automatically included in the subset, thus the clusivity effect is expected. In cases of partitive interpretation, my account predicts that the person agreement pattern would be illicit, whereas Paparounas and Akkuş’s generalization allows for the person agreement pattern if the speaker (addressee) is included (e.g., “three of us (which are five) including me”). The problem is that it is not quite clear whether the superset (“we”) in such cases is still the same and does not get recomputed as coinciding with the relevant subset. To ensure that the superset remains the same, we can make use of contexts of exhaustive listing like (78b). Since in such contexts, the speaker should be included in at least one subset, we expect that one of the inflected numerals can trigger the person agreement pattern, whereas all the others cannot. Examples like (77d) and (78b) suggest that this is not the case in Tatar: in exhaustive listing contexts, all partitives employ the default agreement pattern. Thus, for Tatar, I stick to my generalization and build the analysis upon it.

The revealed contrast suggests the following hypothesis. Partitive constructions with quantifiers allow for two interpretations: the true partitive interpretation, when the subset denoted by the quantifier differs from the superset, and the appositive identity interpretation, when the subset is equivalent to the superset. Quantifiers differ as to their ability to give rise to these interpretations. Distributive and existential quantifiers, as well as interrogatives, are only compatible with the true partitive interpretation; universal quantifiers and collective numerals are expected to produce the appositive identity interpretation, and cardinal numerals and quantity denoting expressions allow for both readings.

The next thing to observe is that these interpretational differences can be structurally represented in the grammar. For instance, Russian employs the appositive construction for the identity reading: my dvoe [we two.coll] ‘we two’ and the elective prepositional construction for the true partitive reading: dvoe iz nas [two.coll from us.gen] ‘two of us’, see, e.g., Paperno (2012). At the same time, the distribution of the grammatical patterns is semantically motivated only with quantifiers compatible with both readings. Non-distributive universal quantifiers, for instance, are attested in both construals (Russian vse my/my vse [all we/we all] ‘we all’ and vse iz nas [all from us.gen] ‘all of us’; cf. also English we all and all of us, Barker 1998; Shin 2016) but no interpretational effects arise.30 Existential quantifiers like ‘some’ or ‘one’ cannot give rise to the identity reading with personal pronouns denoting a superset and, consequently, do not form the appositive construction (Russian odin/nekotoryje iz nas [one/some from us.gen] ‘one/some of us’ vs. *my odin/nekotoryje [we one/some])31. Thus, the appositive construal can only be employed if the quantifier allows for the identity reading, whereas the partitive construal is available in any case, even if the quantifiers’ semantics produce the identity reading.

With this reasoning in mind, we turn back to Tatar inflected quantifiers. Let us make a reasonable assumption that the choice of the agreement pattern is ultimately determined syntactically, i.e., it depends on the structural/featural properties of the inflected quantifier. What we observe is that the person agreement pattern construal is available for non-distributive universal quantifiers and collective numerals, which yield identity readings, and for cardinals and quantity denoting elements, which are compatible with identity readings. The default agreement pattern construal is available for all the quantifiers, whereby for existential and distributive quantifiers, this construal is the only licit option.

The identity reading of inflected quantifiers can be tested independently in the floating quantifier construction. In footnote 23, I introduced the floating intensifier construction available for finite subjects. The floating quantifier construction has the same syntactic distribution. Importantly, the relation of the floating quantifier and its antecedent is that of identity. Thus, we expect floating of inflected quantifiers on the base of universal quantifiers, collective numerals, cardinals and quantity expressions. Inflected quantifiers on the base of existential quantifiers and interrogatives are not expected to float. This expectation is borne out, cf. (79).

(79)	a.	Bez	böten-ebez	Sovet	graždan-ı	bul-ıp	jaz-ıl-dı-k.
		we	all-1pl	Soviet	citizen-3	be-cvb	write-pass-pst-1pl
		‘We all signed in as Soviet citizens.’ [CWT]
	b.	Bez	ike-bez	dä	Räšat-nı	jarat-ır-bız.
		we	two-1pl	ptcl	Reshat-acc	love-fut-1pl
		‘We will both love Reshat.’ [CWT]
	c.	Bez	a-nı	bar-ıbız	da	bik	jarat-tı-k.
		we	this-acc	all-1pl	ptcl	very	love-pst-1pl
		‘We all loved her very much.’ [CWT]
	d.	Sez	härwakıt	dürt-egez	očraša-sız-dır,
		you	always	four-2pl	meet.ipf-2pl-q
		zur	mäsälä-lär-ne	bergäläp	xäl	itä-sez-der?
		big	problem-pl-acc	together	state	make.ipf-2pl-q
		‘Do you always meet, four of you, and solve important problems together?’ [CWT]
	e.	Bez	bit	küb-ebez	sugıš	jätim-när-e	i-de-k.
		we	ptcl	many-1pl	war	orphan-pl-3	aux-pst-1pl
		‘We were numerous war orphans.’ [CWT]
	f.	*Bez	a-nı	ber-ebez	/	bernikadär-ebez	/	kajsı-bız
		we	this-acc	one-1pl		several-1pl		which-1pl
		jarata-(bız).
		love.ipf-1pl

The importance of identity readings for the person agreement pattern is further supported by the following evidence. Existential quantifiers in the scope of negation are logically equivalent to universal quantifiers outside the scope of negation. Not surprisingly, the person agreement pattern is only attested with existential inflected quantifiers functioning as NPIs in negative sentences, cf. (74). The same regularity is observed with inflected interrogatives: in rhetorical questions, they are rather interpreted universally (which of you would ask… > none of you will ask/all of you will not ask), and this entailment gives rise to the person agreement pattern in (75e). Note also that negation licenses existential floating quantifiers, cf. (80).

(80)	a.	Borčıl-u-ı-nıŋ	säbäb-e	bilgele	bul-gan-ga,
		worry-nml-3-gen	cause-3	known	be-pf-dat
		bez	ber-ebez	dä	däš-mä-de-k.
		we	one-1pl	ptcl	speak-neg-pst-1pl
		‘Since the cause of worry was known, we all kept silence.’ [CWT]
	b.	Sez	bügen	ber-egez	dä	keše	tügel,
		you	today	one-2pl	ptcl	man	neg.cop
		ä	intellektual’	milek.
		but	intellectual	property
		‘Today, you are not people, none of you, but intellectual property.’ [CWT]

An interesting parallel is found in Quechua (Muysken 1989, 1992), which possesses inflected quantifiers and intensifiers with various external agreement patterns. Muysken (1992, p. 271) emphasizes that obligatory subject (and optional object) agreement for person is attested with those obligatorily inflected quantifiers where the set of elements denoted by the quantifier is identical with the set of elements denoted by the inflection. This property is characteristic of intensifiers and universal quantifiers. Quantifiers denoting subsets, proper or not, when inflected, are optionally agreed with for person. Interrogatives are optionally inflected, too, but never trigger person agreement. Thus, Tatar is not alone in distinguishing between identity and partitivity relations in inflected quantifiers.

The correlation of the reading and agreement pattern of inflected quantifiers bears a direct consequence for anaphors. Reflexives and intensifiers are built on the noun üz ‘self’ which conveys the idea of identity by its lexical meaning. Reciprocals, on the other hand, employ the numeral/existential quantifier ber ‘one’ which is only compatible with a true partitive reading, unless used as an NPI. Therefore, the person agreement pattern with reflexives and intensifiers and the default agreement pattern with reciprocals fit perfectly into the general picture: reflexives and intensifiers form the identity partitive construction whereas reciprocals form the true partitive construction.

In the next section, I take this hypothesis as a point of departure for developing a formal analysis of external agreement with reflexives and reciprocals in Tatar.

5. Analysis

In order to account for the two agreement patterns with anaphors, we need two major ingredients. The first ingredient is the analysis of agreeing and non-agreeing partitives which would provide a syntactic representation for the two semantic relations attested with inflected quantifiers and anaphors; this representation, ideally, would yield different properties of true partitives and identity partitives with respect to external agreement. The second ingredient is the account of external agreement with anaphors which would be able to incorporate the assumption that bound anaphors contain pro with unvalued phi-features.

5.1. Structural Representation of Partitivity

In Section 2, I assumed a unified representation of possessive DPs and argumental nominalizations as containing a genitive subject controlling possessive agreement of the ezafe marker identified with D. In doing so, I followed the widely recognized assumption dating back to Abney’s (1987) dissertation that in Turkic languages, the highest functional layer of the extended nominal projection in nominalizations is responsible for their DP-like external syntax and for the possessive marking of the nominalization’s subject.

Importantly, in Tatar, the specifier position of D is reserved for argumental DPs theta-licensed in lower positions—in Spec, PossP/Spec, nP in possessive phrases and within the verbal domain in nominalizations (Pereltsvaig and Lyutikova 2014). Raising of possessors/subjects to Spec, DP is a side effect of their Agree relation with D, which case-licenses them. Given these assumptions, the internal syntax of partitives needs further clarification: how is it that the DP in Spec, DP is interpreted as a superset and the complement of D is interpreted as a subset, and where does the DP in Spec, DP raise from?

I propose that the partitive construction is yet another phrase that can be embedded under the DP-shell in Tatar, as illustrated in (81). Consequently, the interpretation of the genitive DP as superset is encoded in this embedded structure; the DP-shell itself only provides it with case.

(81)	a. possessive construction
	[DP DPi [uCase: Val] … [PossP … ti … Poss] … D [uϕ: Val]]
	b. nominalization
	[DP DPi [uCase: Val] … [vP … ti … v] … D [uϕ: Val]]
	c. partitive construction
	[DP DPi [uCase: Val] [PartP … ti … Part] D [uϕ: Val]]

This assumption allows us to explain the fact that true partitives, despite being semantically indefinite (cf. Barker’s 1998 Anti-Uniqueness condition), behave like definite DPs with respect to differential object marking and differential possessor marking (Lyutikova and Pereltsvaig 2015). Indeed, if it is the D head that bears the unvalued case feature, DP and only DP will need case-licensing, and hence appear in syntactic positions construed with case/agreement.

Another advantage of this hypothesis is that it provides us with a straightforward mechanism of encoding the semantic difference between true partitives and identity partitives. In analyzing the internal structure of partitive constructions, I build upon the widely acknowledged theory of predication (see den Dikken 2006; den Dikken and O’Neill 2017 for the fundamentals). It assumes that predicates and their subjects are systematically related to each other within an asymmetrical structure created by a functional head of the category relator (R); this structure can give rise to direct (subject asymmetrically c-commands predicate) and inverted (predicate asymmetrically c-commands subject) c-command relations between the predicate and its subject. Importantly, the relator phrase is argued to underlie not only clause-level copular constructions, where the different order of subject and predicate corresponds to the distinction of predicational and specificational copular clauses (John is the culprit vs. The culprit is John), but also various constructions within the nominal domain, including evaluative N-of-an-N constructions (this idiot of a doctor), double genitives, or of/z possessives (three books of John’s), alienable vs. inalienable possession constructions, linker constructions and true partitive constructions (see Alexiadou and Wilder 1998; Bennis et al. 1998; Sleeman and Kester 2002; den Dikken and Singhapreecha 2004; Martí i Girbau 2010; Franco et al. 2015; Jin 2015; den Dikken 2017; a.m.o).

In the spirit of this research paradigm, I propose that Tatar possesses two types of partitive constructions which are superficially identical but have differing underlying structure and different interpretation. The true partitive construction has a partitive relator R_part at its core. The partitive relator introduces the subset–superset relation between the subject and the predicate, as shown in (82a). On the other hand, the identity partitive construction is based on the standard relation of characterization, similar to that of predicational copular constructions, introduced by the relator R_ident (82b).

(82)	a. partitive predication
	[RpartP subset [Rpartʹ Rpart superset]]
	b. identity predication
	[RidentP subject [Ridentʹ Rident ppedicate]]

Now let us see which syntactic objects appear as subjects and predicates in these RPs. In the partitive RP, the predicate is the phrase denoting the superset (cf. Sleeman and Kester 2002; Jin 2015; Falco and Zamparelli 2019). Since true partitives require definite supersets (this requirement is known as Partitivity Constraint in the literature, first observed in Jackendoff 1977), the predicate of R_part is the DP. The subject of R_part, on the other hand, is a nominal projection with an indefinite interpretation (Jackendoff 1968; Barker 1998). Given that partitive constructions in general allow interpretable number morphology, unless the quantifier bans it, I propose that the subject position of R_part is occupied by NumP. Another well-known semantic constraint on true partitives is that the subset–superset relation associates the sets of the same kind (Selkirk 1977). I believe that this is the reason why the subset description cannot contain a non-vacuous lexical noun and only attests adnominal modifiers of the NP and NumP level. Thus, in (83b) I represent the assumed structure of the partitive predication underlying the true partitive construction bezneŋ ekibez ‘two of us’, (83a).

(83)	a.	bez-neŋ
		we-gen
		‘two of us’
	b.	partitive predication underlying the true partitive construction
		[RpartP [NumP eki [NP eN] Num] [Rpartʹ Rpart [DP bez]]]

In the identity RP, on the other hand, the definite DP is the subject, and the predicate is represented by NumP (84a–b). In principle, RPs of this kind are not confined to the nominal domain and can produce copular clauses like (85); however, in this case, no restrictions on lexical nouns apply.32

(84)	a.	bez-neŋ
		we-gen
		‘we two’
	b.	identity predication underlying the identity partitive construction
		[RidentP [DP bez] [Ridentʹ Rident [NumP eki [NP eN] Num ]]]

(85)	a.	Ike-nčä-dän,	bez	kürše-lär.
		two-ord-abl	we	neighbor-pl
		‘Secondly, we are neighbors.’ [TT]
	b.	Bez	köč-sez-lär	tügel.
		we	force-car-pl	neg.cop
		‘We are not forceless ones.’ [TT]

Thus, the two RPs differ not only as to the semantic relation established between the two nominal constituents, but also as to the syntactic positions of these elements: in the partitive predication, DP is the predicate, whereas in the identity predication, DP is the subject.

Our next observation concerning the structures in (83b)–(84b) is that each includes one DP with an unvalued case feature (the second nominal is NumP which, by assumption, does not have a case feature). Thus, (83b)–(84b) are like other complements of D in (81): they contain a DP which needs case-licensing and should enter the Agree relation with D.

The derivation of the identity partitive construction (84a) is straightforward: the RP’s subject DP enters an Agree relation with D, values its unvalued uninterpretable phi-features, is assigned genitive and raises to Spec, DP (86).

(86)	[DP [DP bezneŋ] [uCase:gen]i [RidentP ti [Ridentʹ Rident [NumP eki [NP eN] Num ]]] D [uϕ:1pl]]

The derivation of the true partitive construction (83a) requires predicate inversion—raising of the nominal predicate across the nominal subject. In principle, such operations can violate minimality, since A-movement crosses another A-position; den Dikken (2006 and elsewhere) proposes that head-movement of the relator licenses predicate inversion. In the partitive RP, however, the subject nominal does not need case and can only be considered as a defective intervener, or not an intervener at all. Therefore, I admit that R-to-D movement may be needed in this configuration, but I believe that nothing in the analysis hinges on this assumption. Thus, the derivation of the true partitive construction is very similar to that of the identity partitive construction, except that raising of the predicate nominal might require an additional head movement operation (87).

(87)	[DP [DP bezneŋ] [uCase:gen]i [RpartP [NumP eki [NP eN] Num ] [Rpartʹ tj ti]] Rj+D [uϕ:1pl]]

What we have achieved so far is that the true partitive construction and the identity partitive construction, being superficially similar, have different internal structure and different interpretation. What we need now is to ensure that the two partitive constructions have different phi-features.

The discussion in Section 2.3 leads to the conclusion that DP’s own phi-features which manifest themselves in external agreement are inherited from the lower heads in the nominal extended projection. I suggest that it is the relator which is the source of phi-features for DP in partitives. The relator, in its turn, is like a copula in that it agrees with its own subject. This agreement process may be construed as feature unification in the Spec-head configuration, as an upward probing of the Relator or as a downward probing of the complex head R+D; since I assume that the Agree relation can be established in various ways (see Section 5.2 below), the choice here is not decisive. Therefore, the two partitive constructions will have different phi-feature sets: the identity partitive construction will inherit the phi-features of the subject of the identity predication (the DP bez ‘we’ in (86)), whereas the true partitive construction will inherit the phi-features of the subject of the partitive predication (the NumP eki e_N ‘two’ in (87)).

(88)	a. the identity partitive construction
	[DP [DP bezneŋ] [RidentP eki eN] D [uϕ:1pl] ] [iϕ:1pl]
	b. the true partitive construction
	[DP [DP bezneŋ] [RpartP eki eN ] D [uϕ:1pl] ] [iϕ:3sg]

At that point, it is worth emphasizing that the analysis presented above treats the default agreement pattern attested with the true partitive construction as a standard agreement with a 3p nominal, and not as failed agreement, in the sense of Preminger’s (2014) model. This is a welcome characteristic of the analysis. Indeed, true partitive DPs cannot have a marked person feature (since it is never present at the NumP level), but can have an interpretable number feature. Not surprisingly, we find examples of number agreement with inflected quantifiers which contain a plural affix, cf. (89). Examples like (89) are problematic for any account which considers the default agreement pattern with inflected quantifiers as lack of agreement.

(89)	a.	Kajsı-lar-ıgız	näfs-e-nä	ijär-de-lär,	alar	šajtan
		which-pl-2pl	desire-3-dat	follow-pst-pl	they	devil
		art-ı-nnan	kit-te-lär.
		after-3-abl	leave-pst-pl
		‘Which ones of you followed their passions, they followed the devil.’ [CWT]
	b.	Kajsı-lar-ıbız	agačlık-lar	ara-sı-na	kač-u	ǯaj-ı-n
		which-pl-1pl	grove-pl	between-3-dat	run-nml	chance-3-acc
		kara-dı-lar.
		look-pst-pl
		‘Some of us looked for a chance to escape in the groves.’ [CWT]
	c.	Šunda	kajber-lär-ebez	kaber	jan-ı-na	bar-ıp,
		then	a_few-pl-1pl	grave	near-3-dat	come-cvb
		üz-lär-e	kür-de-lär.
		self-pl-3	see-pst-pl
		‘Then a few of us came to the grave and saw (it) for themselves.’ [CWT]
	d.	Kajber-lär-egez	doklad	äzerlä-de,	kajber-lär-egez
		a_few-pl-2pl	presentation	prepare-pst	a_few-pl-2pl
		inša-lar	jaz-dı-lar,	räsem,	plakat	jasa-dı-lar.
		essay-pl	write-pst-pl	picture	poster	make-pst-pl
		‘Some of us prepared presentations, some of us wrote essays, made pictures and posters.’ [CWT]

To conclude, our analysis predicts that inflected quantifiers based on the identity partitive construction have the phi-features of their supersets (i.e., ike-bez ‘we two’ is 1pl) whereas inflected quantifiers based on the true partitive construction have the phi-features of their subsets (i.e., ike-bez ‘two of us’ is 3sg). These features are not only revealed in agreement in the standard way but also trigger specific variation in possessive and postpositional constructions. As we saw in Section 2.1, 1–2p pronouns differ from other nominals in that they (i) form the PFG (ezafe-less) construction; (ii) are assigned genitive in postpositional phrases and (iii) combine with plain (ezafe-less) postpositions. 1–2p identity partitives share all these properties with 1–2p pronouns (cf. Section 4), which lends further support to our analysis.

5.2. External Agreement with Anaphors

In the previous section, I proposed that the two agreement patterns attested with inflected quantifiers ultimately result from the two types of predication underlying the partitive construction in Tatar. This proposal has a direct bearing on external agreement with anaphors. Reflexives are based on the identity-denoting noun üz ‘self’ and form the identity partitive construction; consequently, they possess phi-features inherited from their underlying subject, i.e., the nominal in the specifier of DP. Reciprocals, on the other hand, are based on the numeral/indefinite ber ‘one’ and can only produce the true partitive construction; this is why their phi-features are those of the NumP containing ber.

(90)	a. reflexives based on the identity RP
	[DP [DP pro] [iϕ:1pl] [RidentP (üz-)üz] D [uϕ:1pl] ] [iϕ:1pl]
	b. reciprocals based on the partitive RP
	[DP [DP pro] [iϕ:1pl] [RpartP ber-ber eN ] D [uϕ:1pl] ] [iϕ:3sg]

However, there remains one problematic issue. In Section 3, I proposed that anaphors contain a silent pro with unvalued phi-features which only get valued as a result of binding. Therefore, at the stage when the anaphor’s partitive construction is built, the phi-features of pro are yet unvalued. As a result, they cannot value D’s uninterpretable phi-features (which both reflexives and reciprocals need) and cannot be inherited by the anaphor’s partitive DP (which is essential for reflexives).

Furthermore, at the moment when the external agreement with the reflexive takes place, its binder has not yet entered the derivation. In (91), the domains of the external agreement with reflexives are indicated with the dashed line; we observe that in all cases of external agreement with bound reflexives, the binder (shown in boldface) c-commands the phi-probe of the external agreement (italicized) and the whole agreement configuration. Consider (91b) as an example. The nominalization’s subject üzemneŋ ‘myself’ receives genitive and is expected to trigger possessive agreement of the DP shell the nominalized clause is embedded under. At that point, the binder has not entered the derivation yet. Therefore, the reflexive’s features remain unvalued until the nominalization first combines with the lexical verb and then the vP is projected and the external argument binds the anaphor. Consequently, the phi-features of the phi-probe in the nominal shell of the nominalization cannot be valued.

(91)	a. possessive construction
	b. nominalization
	c. postpositional phrase

This line of reasoning reproduces exactly the account of AAE based on an anaphor’s featural deficiency (Murugesan 2019): if the agreeing probe is lower than the binder, agreement with an anaphor fails or yields default values. Against expectations, Tatar exhibits co-varying agreement with reflexives in the very same configuration where the theory predicts it to be absent. Therefore, we have to come up with an idea for how to render the AAE inoperative in Tatar.

One option is to dispense with the assumption that the anaphor’s phi-features are unvalued. This is the path taken in Rudnev (2017, 2020) in explaining agreement with anaphors in Avar. Rudnev assumes that the anaphor’s phi-features are interpreted as presuppositional restrictions on the variable. Applied to Tatar, this step will make it impossible to maintain the distinction between bound and free pros, which the distribution and interpretation of Tatar anaphoric expressions requires.

Another option is to assume that agreement with anaphors is exactly what mediates the binding and the phi-features’ transmission from the binder to the bindee. Recently, evidence has been presented that binding can indeed trigger co-varying agreement in the predicate projecting an anaphoric argument (Murphy and Meyase 2022; Paparounas and Akkuş forthcoming). Applying this line of reasoning to Tatar data, we could say that agreement of a functional head with an anaphor is a prerequisite of binding this anaphor and that the binder transmits its phi-features to the anaphor via this functional head. Importantly, Agree-based binding is heavily restricted to coarguments of the same predicate; thus, Paparounas and Akkuş (forthcoming) argue that only coarguments, i.e., DP-arguments entering Agree with the Voice head, can be bound in this way. However, the configurations we are dealing with in (91) cannot be represented as involving a single functional head which agrees with the binder and the bindee. Specifically, in (91a–b) the anaphor does not enter the Agree relation with the matrix Voice head (instead, the matrix direct object does, but it has its own interpretable phi-features, 3p sg, which do not coincide with those of the anaphor or binder). In (91c), we are dealing with an adjunct (locative) PP which is an opaque domain for agreement of the clausal functional structure. Paparounas and Akkuş (forthcoming) note that in Turkish, anaphors inside adjunct PPs do not trigger co-varying agreement on the predicate; it is reasonable to assume that in Tatar also, binding of anaphors in this configuration is not mediated by agreement of Voice. Therefore, I conclude that binding through Agree mediated by a common functional head is not a viable option for accounting for (91).

Yet another option, which I am going to take, is to assume that the standard model of agreement is too restrictive and that these restrictions must be relaxed. It is clear that in configurations of agreement with anaphors, the valued phi-features of the binder have a direct impact on the valuation of multiple phi-feature sets: on the anaphoric pro, on the anaphors’ D, on the reflexive’s relator and DP, on the external probes such as possessive D and p. What we are dealing with is a situation of multiple agreement targets with one agreement controller, which is not covered by the standard one-to-one Agree model (Chomsky 2000).

Alternative formal models of agreement are especially concerned with multiple agreement targets. Some researchers assume that the probe attempts Agree more than once, finding all the goals in its domain (Nevins 2007, 2011), whereas others claim that multiple agreement is driven by the goal rather than by the probe (Zeijlstra 2012). The two approaches are best suited to different agreement configurations: the former works when the unvalued probe is higher than multiple valued goals (e.g., PCC configurations or omnivorous agreement), and the latter is appropriate when multiple probes need independent checking against a higher goal (e.g., negative concord or multiple case licensing).

It seems that the agreement configuration we are dealing with in Tatar looks like the latter case: the source of phi-features for lower elements is the highest DP-binder. Accordingly, one might suggest that the upward Agree (Zeijlstra 2012) is the right choice here: each of the constituents with unvalued phi-features probes separately until it finds a c-commanding DP with valued phi-features (92).

(92)

However, the problem with this account is that it only covers a very specific situation. If the lower element is not an anaphor but a DP with valued phi-features, the direction of probing and valuation is reversed: the heads F₁, F₂, F₃ probe downward rather than upward (93):

(93)

A model which would accommodate both situations (92) and (93) is the Cyclic Agree (Béjar and Rezac 2009): the probe is allowed to search upward after the downward search failed. A drawback of this approach is that it allows for the probe to look for the higher goal whenever there is no suitable lower goal, thus failing to distinguish between a well-formed agreement configuration featuring the anaphoric pro and a configuration where the lower goal is absent or has no phi-features at all.

I believe that attempts to subsume all the agreement-like processes under the same syntactic mechanism are on the wrong track: there exist syntactic dependencies of various kinds, and they differ as to the probing direction, correspondence between the number of probes and goals, etc. The standard predicate agreement procedure as presented in Chomsky 2000 is different from negative concord (Zeijlstra 2004, 2012) or anaphoric dependencies (Kratzer 2009; Rooryck and Vanden Wyngaerd 2011), although they can also be modeled as agreement. Thus, I believe that the best way to analyze external agreement with anaphors is to distinguish between feature unification through standard agreement and feature unification through binding, much along the lines of Wurmbrand (2017).33 Informally, the idea is that the standard agreement relation between the heads F₁, F₂, F₃ and the anaphor is established, but the valuation of phi-features is postponed until the anaphor is bound (94). Importantly, anaphors are allowed to probe upward, and only their binders (coindexed c-commanding DPs) are eligible for the agreement process.34

(94)

In implementing this idea, I rely on Pesetsky and Torrego’s (2007) feature sharing model. Its most appealing aspect is that it allows Agree between two elements which both bear unvalued matching features (95a). As a result, features in different locations α and β undergo a unification process which transforms occurrences of a feature F into instances of this feature, indicated in (95a) with coindexing [7]. When a subsequent Agree with an element bearing a valued feature F occurs, all the instances of F are valued simultaneously (95b).

(95)	a.	feature unification
		Fα [] … Fβ [] → Fα [7] … Fβ [7]
	b.	simultaneous valuation
		Fα [7] … Fβ [7] … Fγ val []→Fα (val) [7] … Fβ (val) [7] … Fγ val [7]

This mechanism is exactly what we need for (94). Feature sharing will unify phi-feature sets on the anaphor, F₁, F₂ and F₃, assigning them a common index. When the binder enters the derivation and establishes Agree with the anaphor, all the instances of the phi-feature set will receive values provided by the binder.

Below, I provide example derivations of the person agreement pattern with the reflexive (96)–(97) and the default agreement pattern with the reciprocal (98)–(99). Both involve feature sharing and feature valuation under binding, and differ as to the number of elements involved in the agreement chain of the anaphor.

(96)	Min	jılan-nı	üz-em-neŋ	jan-ım-da	kür-de-m.
	I	snake-acc	self-1sg-gen	near-1sg-loc	see-pst-1sg
	‘I saw a snake near myself.’

(97)	derivation of (96), relevant parts
a.	anaphoric pro selected
	proi ϕ:__ []
b.	identity predication built; phi-feature sets of R and pro coindexed
	[RP proi ϕ:__ [5] [Rʹ R ϕ:__ [5] [NumP üz ]]]
c.	partitive construction built; phi-features of D and DP coindexed with those of pro and R
	[DP proi ϕ:__ [5] [RP [Rʹ R ϕ:__ [5] [NumP üz ]]] D ϕ:__ [5] ] ϕ:__ [5]
d.	reflexive embedded under pP; phi-features of p coindexed with those of DP
	[pP[DP proi ϕ:__ [5] [RP [Rʹ R ϕ:__ [5] [NumP üz ]]] D ϕ:__ [5] ] ϕ:__ [5] [PP janda] p ϕ:__ [5] ]
e.	binder enters the derivation; all the instances of ϕ:__ [5] valued
	[vP mini ϕ:1sg [5] … [pP[DP proi ϕ:1sg [5] [RP [Rʹ R ϕ:1sg [5] [NumP üz ]]] D ϕ:1sg [5] ] ϕ: 1sg [5] [PP janda] p ϕ:1sg [5] ] …]

(98)	Bez	jılan-nar-nı	ber-ber-ebez	jan-ı-nda	kür-de-k.
	we	snake-pl-acc	one-one-1pl	near-3-loc	see-pst-1pl
	‘We saw snakes near each other.’

(99)	derivation of (98), relevant parts
a.	anaphoric pro selected
	proi ϕ:__ []
b.	partitive predication built; phi-feature sets of R coindexed and valued against NumP
	[RP [NumP ber-ber ] ϕ:3sg [7] [Rʹ R ϕ:3sg [7] proi ϕ:__ []]]
c.	partitive construction built; phi-features of D coindexed with those of pro, phi-features of DP valued against R
	[DP proi ϕ:__ [5] [RP [NumP ber-ber ] ϕ:3sg [7] [Rʹ R ϕ:3sg [7]]] D ϕ:__ [5] ] ϕ:3sg [7]
d.	reciprocal embedded under pP; phi-features of p coindexed with those of DP
	[pP [DP proi ϕ:__ [5] [RP [NumP ber-ber ] ϕ:3sg [7] [Rʹ R ϕ:3sg [7]]] D ϕ:__ [5] ] ϕ:3sg [7] [PP janda] p ϕ:3sg [7] ]
e.	binder enters the derivation; instances of ϕ:__ [5] on pro and D valued
	[vP bezi ϕ:1pl [5] … [pP [DP proi ϕ:1pl [5] [RP [NumP ber-ber ] ϕ:3sg [7] [Rʹ R ϕ:3sg [7]]] D ϕ:1pl [5] ] ϕ:3sg [7] [PP janda] p ϕ:3sg [7] ] …]

6. Conclusions

In this paper, I presented evidence for the two agreement patterns available with inflected anaphors in Tatar. I showed that reflexives and reciprocals differ regularly with respect to external agreement: reflexives attest the person agreement pattern whereas reciprocals exhibit the default agreement pattern. The choice of the agreement pattern is consistent across all the agreement configurations.

In solving this puzzle, I examined the internal structure of reflexives and reciprocals and came to the conclusion that they possess a complex internal structure and that their properties with respect to syntactic and semantic binding are ultimately determined by the silent pronominal element in their highest specifier. Bound reflexives and reciprocals host an anaphoric pro, whereas unbound reflexives host a pronominal pro. Importantly, bound reflexives and bound reciprocals have the same binding-theoretical status; therefore, the difference in agreement patterns cannot be attributed to the AAE operative with reciprocal and inactive with reflexives.

In search of an explanation, I examined a broader set of data provided by inflected quantifiers. Inflected quantifiers belong to the same structural class of partitive constructions as reflexives and reciprocals do. They also show variation with respect to the patterns of external agreement. The distribution of agreement patterns available with inflected quantifiers allowed me to suggest that the agreement pattern is ultimately determined by the relation of the quantifier and the pronominal element underlying the partitive construction. I proposed that the partitive construction can be based on predications organized around two different relators—the partitive relator and the identity relator. Since the relators encode reverse subject-predicate relations between the pronoun and the quantifier, the resulting partitive constructions have different phi-features, those inherited from the pronoun or those inherited from the nominal projection containing the quantifier.

I extended this account to reflexives and reciprocals in arguing that reflexives, being based on the identity denoting noun üz ‘self’, only supports the identity predication, whereas reciprocals employing the numeral/indefinite ber ‘one’ are only compatible with the partitive predication. This account is supported by the fact that with respect to external agreement, reflexives pattern with inflected intensifiers, which are also based on üz ‘self’, whereas reciprocals pattern with inflected quantifiers based on ber ‘one’.

Finally, I developed a formal model of external agreement with anaphors incorporating the conjecture that anaphors enter the derivation with unvalued phi-features which are only valued by binding. I proposed that the apparent countercyclicity of the person agreement with anaphors can be addressed under the feature sharing account, whereby the ϕ-features on the anaphor and on the ϕ-probe are first identified as instances of the same ϕ-set and then valued by the anaphor’s binder.

This study is instructive in several respects. First, it contributes to the discussion of the AAE in that it provides a clear case of the co-varying agreement of phi-probes located between the anaphor and its binder. Secondly, it presents novel evidence concerning agreement with inflected anaphors and quantifiers in Turkic languages. It shows that the alternative to the person agreement with inflected anaphors and quantifiers is not agreement failure, but regular agreement with a 3p nominal. This fact constitutes a counterargument to approaches building on the absence or invisibility of phi-features on inflected anaphors and quantifiers, as well as to approaches assuming direct agreement of external probes with the pronominal element of inflected anaphors and quantifiers. Last but not least, this study lends support to the hypothesis initially put forward by Kornfilt (2001) which derives the properties of Turkic inflected anaphors from their internal structure, and, specifically, from the characteristics of their pronominal specifier.