2.1 Copula

One important feature of Turkish is its copula, which plays a role in my proposed analysis. It has been proposed that Turkish has a (mostly) null copula (Kornfilt 1996). So example (1) can be revised to include the null copula.⁴

Kornfilt (1996) not only proposes the existence of this copula, but also argues that the majority of the agglutinization argued for in the Turkish verbal domain is actually cliticization of the copula, which inflects for person/number features, onto the verb, as opposed to the verb itself inflecting for person/number features. She argues that there are only two true verbal tenses, past(with the suffix -dI) and conditional (with the suffix -sE). Aside from these, the first instance of all other TAM markers is followed by the copula. Then when this first instance is followed by additional TAM markerss or affixes, those are not directly attaching onto the first TAM marker but instead attach onto the copula following the first TAM marker. A non-copular construction is given in (4), whereas a complex form with two TAM markers, and hence copula after the first, is given in (5).

The most straightforward argument for the existence of this copula is formal registers and local Anatolian dialects of Turkish which overtly pronounce i- in all the locations that the copula is proposed to exist. It is argued that this i- is the non-null realization of the copula in that register, as given in (6).⁵

2.2 Agreement paradigm

The previous examples demonstrate a variety of agreement morphemes on nouns and verbs inflecting for person. I give a more complete paradigm of this below.

Please note that glosses P₁ and P₂ are somewhat unconventional and normally glossed as GEN and POSS. I have chosen to gloss these categories differently because I would like to stay neutral as to whether or not this work connects to case theory, and because one of the aims of this paper is to reanalyze the cross-paradigm syncretisms as stemming from their being underlyingly the same morpheme.

An example with P₁ and P₂ is given in (8):

The person dichotomy in Turkish can immediately be observed in (7): the inflectional paradigms all show a stark distinction between the morphology of 1/2 and of 3. For example, one can see that 3sg morphology is null in V₁ and V₂, and 3pl inflection in those categories is -lEr or a null morpheme. These are very distinct from the morphemes for 1/2 in these same categories.

V₁ and V₂ are more well-known as the z-paradigm and k-paradigm, respectively, after the 1pl distinctions (originally coined by Good & Yu 1999b). In fact, the two verbal tenses not followed by the copula are exactly those which take k-paradigm endings, not z-paradigm endings, whereas those constructions with one TAM marker taking z-paradigm endings are actually inflecting the copula (as in (3)). Part of the goal of the paper to provide an analysis of the morphosyntax of verbal agreement in Turkish, taking into account the fact that there are so many distinct forms of this agreement. In this paper, I discuss the clitic-doubling syntax with respect to the z- and k-paradigms (V₁ and V₂). The aim is an analysis that can be extended to the other two nominal inflectional paradigms as well. As such, I provide vocabulary items and insertion rules in Section 5 for the inflectional categories of all four categories of verbal and pronominal categories,⁸ which abstract away from the syntax of possession and exactly how features are manipulated in the syntax. A future goal is to further refine the clitic-doubling machinery in order to account for the syntactic component of the nominal inflection.

2.3 1/2 vs. 3 in nominal possession

The syntactic distinction between the behavior of 1/2 vs. 3 has been previously observed, and there is an analysis of the nominal possession dichotomy by Kunduracı (2013) which is particularly relevant for this work.⁹

Kunduracı claimed that the third-person possession suffix -(s)I, as seen in column P₂, is specified only for a possession feature, as opposed to first- and second-person possession suffixes which are specified for both possession and person features. In other words, what is normally treated as the possession suffix marking third person is in fact just a possession suffix, and contains no person features. One implication of this would be that possession is an independent grammatical category in Turkish. A consequence of this is the fact that the first and second person possession markers are fusional, i.e. they mark both person and possession, whereas the third person possession marker simply marks possession. They do not co-occur because the possession suffix -(s)I is overridden by the more specific markers containing both person and possession, in the case of 1/2.

Kunduracı’s analysis is based on a few facts:

Kunduracı claims that this aligns with the fact that third person in Turkish is generally unmarked.¹⁰

This argument directly parallels much of what I will argue in the case of 1/2 vs. 3 dichotomy in the context of verbs—namely, that 3pl verbal morphology does not contain any person features, but rather is just the overt plural morpheme coming directly from the nominal subject.

2.4 Clitics in Turkish and the clitic/affix distinction

The distinction between clitics vs. affixes is a far-reaching issue in morphosyntax. Zwicky & Pullum (1983) established a set of well-known diagnostics to distinguish the two. Nevins (2011) argues against some of these diagnostics with various examples of elements which morphosyntactically pattern as clitics but would be wrongly categorized as affixes using the criteria from Zwicky & Pullum (1983), such as the condition that clitics not display allomorphy: Nevins (2011) argues that it is extremely difficult to find a clitic that does not demonstrate any allomorphy whatsoever (e.g. Romance clitics). Yuan (2021) also cautions against morphological diagnostics for distinguishing clitics and affixes, using a case-study of two related Inuit languages and demonstrating that object DPs in one of the languages, Inuktitut, express properties of doubled clitics, whereas in the other, Kalaallisut, they express properties of agreement. In both cases, the distinction is not perceivable from a morphological standpoint. Kramer (2014) also explores object agreement morphology, using Amharic, and argues for both—that there is clitic-doubling, but that the object marker also undergoes Agreement prior to movement and M-merger of the clitic.¹¹

In Turkish, previous work has proposed that the verbal person agreement morphemes in the z-paradigm originate from clitic-doubling of the pronominal subjects (Erdal 2000; Good & Yu 2000; Kelepir 2001). Erdal (2000) argues that the cliticized pronoun cannot bear stress, similar to the case in French, which is why the (non–cliticized) subject is often overt. The non–cliticized subject occurs because of the subject clitic, not the other way around, as is assumed with agreement morphology. Despite the fact that the verbal agreement has been acknowledged as originating via clitic-doubling, to my knowledge the specific machinery of proposed clitic-doubling in Turkish has never been worked out as this paper attempts to.¹²

I motivate the clitic-doubling analysis of 1/2 by first exploring the dichotomy of 1/2 vs. 3 through syntactic and morphological data, which naturally leads to the analysis of agreement as clitic-doubling for 1/2.

3.1 Syntactic puzzle

There is a plethora of syntactic data pointing to the presence of such a dichotomy between 1/2 and 3. This paper will examine data that can be divided roughly into three categories: optional vs. obligatory inflection, verbs with multiple TAM markers, and polar questions. I will discuss the person dichotomy via the unique behavior of 3pl inflection in comparison to 1/2 inflection by exploring these three types of data in turn.

3.2 Morphological puzzle

I explore the 1/2 vs. 3 dichotomy not only with the z-paradigm (as is the case with the TAM marker and question particle syntactic data), but with all four inflectional categories, repeated from the original table (7), below.

As before, the distinction between the 1/2 vs. 3 rows can be observed, and the similarities are insular within each category: each 1/2 row resembles another 1/2 row to a high degree, whereas the 3 rows are especially distinct with respect to V₁ and V₂ and do not resemble any 1/2 rows. There is an additional distinguishing fact of the 3pl agreement morpheme -lEr: it is syncretic with the nominal plural. Johanson (1976) and Keskin (2017) support this account by giving the historical perspective that most likely, the verbal -lEr is the same as this nominal -lEr, with the idea that the verbal -lEr originated as the nominal -lEr before changing type, from nominal to verbal. In my account, the verbal -lEr is actually still the nominal -lEr—the pl morpheme on the subject is exactly the morpheme that ends up on the verb and is identified as the 3pl verbal morpheme.

4.1 Morphological decomposition

I propose that the pronouns can be viewed as bimorphemic, and claim the following decomposition:²⁵

In this paper I use the [±Auth] (author) and [±pl] (plural) features to characterize person and number for 1/2.^26,27 Then the four pronouns are comprised of an initial morpheme, b- ([+Auth]) or s- ([-Auth]), determining whether the pronoun is first or second person, and the final morpheme, -en ([–pl]) or -iz ([+pl]), which indicates the sg vs. pl feature. On the other hand, the 3pl pronoun contains not the plural marker -Iz, but in fact the nominal plural marker -lEr. The 3pl pronoun can be then morphologically decomposed as in (23):

Then 3pl is really o, the 3sg pronoun (syncretic with the demonstrative ‘that’), followed by the standard plural marker –lEr, with [n] between the two.

There is good reason to believe that [n] is actually a part of the underlying form of demonstratives, disappearing with nominative case. Example (24) demonstrates what happens when three different demonstratives (including o) have different cases applied to them: all forms contain [n] except the nominative form.

As such, we take the underlying form of 3sg (and the other demonstratives) to be /on-/, and propose that the /n/ is deleted when adjacent to nominative case.

The fact that the pluralization of first and second person is a different plural marker –Iz, which I deem the true-human plural, is starkly different from third person subjects/person markers following the standard plural –lEr. This motivates the following proposed structure of the two types of pronouns:²⁸

In other words, the 1/2 pronouns have a structure involving the author features associated with personhood and number in the context of person, whereas 3pl is structured like a simple, pluralized noun. Re-examining the agreement markers of the z-paradigm/V₁, it can be observed that they are similar to these [±Auth] and [±pl] morphemes. The agreement morphemes may then be decomposed as well, so as to examine the relationship between the pronouns and the agreement morphology of the z-paradigm.³⁰

Having seen how pronouns can be decomposed into person and number, I now explain how apparent agreement works.

4.2 Clitic-doubling analysis for 1/2

As mentioned earlier, it has been proposed that the z-paradigm agreement morphemes derive diachronically from the pronouns themselves (the k-paradigm agreement morphemes are thought to have originated from the possession suffixes P₁ and P₂) and that the agreement morphemes themselves are clitics (Erdal 2000; Good & Yu 2000; Kelepir 2001). I will connect this to the larger body of work on the syntax of clitic-doubling, as the morphology of the pronouns and the agreement morphology of 1/2 are similar enough that they motivate a clitic-doubling analysis.³¹ I begin by stating the features of my proposal.

First, doubled clitics of 1/2 have a big-DP structure,³² where the double is the overt subject and the original clitic is what I have been referring to as the person agreement morpheme on the verb. The big-DP structure inspired by Uriagereka is the following:³³

In our case, then, the doubled clitic looks like:

The right child, [# [Person] [#]], is the adjoined clitic, not projecting its own phrase, which is the person agreement we’ve been referring to, while the left child is the pronoun itself. I refer to it as big-DP, to parallel Uriagereka, but of course in our case it is is ‘big-#P’.

This big-DP doubled-clitic structure originates in its argument position—in this case, VP. Throughout the derivation, the overt subject moves to spec-PersonP (clitic-licensing phrase) while the clitic ends up in Person-head.³⁴ Finally, one last important aspect of the analysis is that 1/2 number is also realized in Num-head, in addition to the number realized in the clitic. The motivation for this is due to morphological specifics of the clitics themselves, which are discussed in Section 5.³⁵ Finally, it can be observed these movements appear on the surface to violate various conditions on movement, such as the Head Movement Constraint. However, this is the case for the big-DP clitic structure in general, not just this analysis, as our analysis aligns with how others in the literature have used the big-DP structure for clitic doubling (Uriagereka 1995; Roberts 2010; Nevins 2011; Arregi & Nevins 2012; Tyler 2019). So these apparent violations, should they be taken as such, are a larger issue for the big-DP analysis of clitic doubling.

4.3 Null subject analysis for 3

On the other hand, I claim the 3pl verbal morphology is not doubling at all—it is simply the plural marker from the 3pl pronoun, when the subject is covert.⁴² In other words, the plural morpheme -lEr is always overt. It is the nominal subject that is either overt or covert—when overt, -lEr remains on the subject, as seen in the following. Note the usage of the NumP in this derivation as well: the probe is successful because it reaches the #P which is specified for number. Note also that there is no PersonP, as there was with 1/2, due to our claim that third-person is not specified for person (it has no [±Auth] feature) the way that 1/2 are. There is also no CopP, unless a TAM marker exists, as with the k-paradigm.

On the other hand, when the subject is covert, it is a null pro (without any number features). This lack of number features means that the Num-head probing cannot be satisfied by reaching its DP goal pro in spec-TP. The derivation is rescued by the existence of -lEr, which is inherently specified for plural and subsequently cliticizes onto Num-head.

Recall that example (10b), when -lEr is present both on the subject and the verb, had an acceptability jugement of #. Syntactically, this surface form can be understood as an instance of clitic-doubling of the plural morpheme -lEr, which then appears both on the subject and on the verb. This explains the redundancy of this example for those speakers who label it as such: the same exact morpheme is being doubled and repeated.

5.1 Pronouns

In the case of the pronouns, I have already proposed their structures. The vocabulary items are identical to the above, except for [–pl]—the vowel is e, not I.^44,45,46

The insertion rules that explain how these same morphemes combine in slightly different ways to yield the z-paradigm surface morphology are next given.

5.2 z and k-paradigm morphological rules

The V₁/z-paradigm and V₂/k-paradigm sections of the morphological table are repeated here:

Recall that the syntactic framework utilizes the number of the clitic being reflected in the Auth-head. As such, the rules will involve two instances of the number feature.

On the other hand, the same should not hold for the author feature: the author feature should be realized only once. Observe that that the clausal spine in (21) contained a PersonP as the highest phrase, which has a ±Auth feature in its head. This is in addition to the ±Auth feature housed in the subject clitic. The author feature should be realized in this subject clitic before the number features. In order to prevent two separate realizations of two author features, I claim that the second author feature, housed in the Person-head, is unrealized because the vocabulary items containing an ±Auth feature apply only to interpretable author features. The PersonP being a part of the clausal spine motivates its valued author feature to be uninterpretable and hence inaccessible for vocabulary insertion.⁴⁷

The place assimilation rule explains why the person agreement for 1sg is not -bIn, but rather /-Im/, because the labial feature of 1sg does not become [b] but rather labializes the [n] in /-In/ to become /-Im/.

The next observation is that in the syntactic structure, there are two number features, one in the clitic and one in Num-head. In the case of 2pl, /-sInIz/, it appears that the first realization of number is in fact /-In/, the singular number morpheme. This motivates an impoverishment rule which changes a plural number feature to singular if it adjacent to Person and adjoined to the Num-head. In particular, I follow Calabrese (2011) in assuming that this occurs via deletion of the marked value and then insertion of the unmarked version of that value. In this case, the unmarked version of the plural feature is [–pl], so impoverishment will result in [+pl] becoming [–pl].

Our final observation for the z-paradigm is that the entirety of the author feature and first number feature are not existent in the 1pl morpheme /-Iz/, and that there is only one singular number morpheme for singular subjects (not two, despite there being two heads with number features). This leads to the following obliteration rules:

The obliteration rule above deletes the entire node, along with the marked feature. Impoverishment replaces a deleted feature with the underspecified/elsewhere version, whereas obliteration removes this possibility altogether because the node along with the feature has been deleted. For the k-paradigm, one difference I note is that the second-person pronouns appear to have no /s-/ morpheme, which I claimed realizes the [–Auth] feature. As such, it appears that the locality to the k-licensing TAM marker causes obliteration of a [–Auth] feature.

The other k-paradigm difference is the fact that 1pl is realized as /-(I)k/, rather than /-(I)z/. The previous rules of impoverishing the first [+pl] to [–pl] and then obliterating the [–pl] feature when there exists a [+Auth] feature and a [–pl] feature, still hold. However, instead of obliterating the [+Auth] feature, as was done in the z-paradigm for 1pl, I instead have a rule which fuses the [+Auth] and the [+pl] feature bundles. This is realized as /-(I)k/.

Finally, note that this fusion rule must apply first.⁴⁹

Interestingly, note that in the case of the fusion rule (5), because this applies first (before any impoverishment) there are actually two [+pl] features for the [+Auth] to fuse with. I claim that it fuses with both, yielding the feature set [+Auth,+pl,+pl]. However, because this is a mathematical set, [+Auth,+pl,+pl] = [+Auth,+pl] (sets cannot contain repeated elements). This explains why 1pl in the k-paradigm does not contain two realizations of the number feature—they are both fused into a larger set, and then collapsed into one instantiation of +pl due to the fact that, with their identical valuations, they are identical elements of the set.

I demonstrate how this results in the correct vocabulary items in the following. Impover- ishment is denoted with an additional line where the updated (post obliteration/impoverishment) features are listed prior to the final line giving the different distinct vocabulary items. For convenience, I have repeated the vocabulary items defined in (35). These in combination with the rules just defined are used to derive the stated patterns of the z- and k-paradigms.

Next, I give the derivation of the morphemes for the two possession paradigms.

5.3 Possession paradigm modifications

At this point, I consider the two possession paradigms, P₁ and P₂. This is occurring isolated from the syntactic machinery, as that which has been developed this far has been for verbal constructions. In the case of P₂, there is reason to think there is a second instance of the number morphology, as will be demonstrated. These two instances are labeled with Num1 and Num2.

5.4 Discussion

While there are many ways to approach the vocabulary items, insertion rules, and decompositions, this section demonstrates that it is possible to account for the cross-category syncretisms with one morphological system. The implementation develops a more general analysis which can be applied to all four paradigms, despite the fact that they may appear to have syntactic differences.

One other observation I make is the connection between the P₁ morphemes, which only have one number morpheme, and the fact that they only appear on pronouns in the case of 1/2. As given in the previous pronominal decompositions, the pronouns already have one number feature being realized as a morpheme within them. Then it makes sense that the P₁ morphemes which attach to those only have one number morpheme as well: if there are two defaults, one to have two number features/morphemes in the endings and the other to only have one number feature realized (except in the case of 2pl), then some rule existing to obliterate one of the number features is consistent with the obliteration rules I have already proposed for other paradigms. It also makes sense that the P₁ endings have this impoverishment/obliteration, as opposed to the pronoun, since the pronouns have already been established to contain their number morpheme, and so the affix from P₁ must do the heavy lifting to compensate by impoverishing/obliterating to remain within the required amount of number morphemes. This is yet another indication that the syncretisms across the four inflectional categories are similar enough to motivate one morphological system that governs all of them.