The literature on reflexivity suggests that it is a universal part of language, observed in many languages across different language families, albeit encoded with different grammatical mechanisms. With respect to the strategies that encode reflexivity, we can talk about two main kinds of reflexivity: lexical reflexivity and grammatical reflexivity. Lexical reflexivity, which is also sometimes called inherent reflexivity, is a phenomenon we observe on predicates that express events that are prototypically done to oneself, such as “bathe”. The default interpretation of an intransitive sentence like I bathed in English is reflexive, I bathed [myself]. Predicates like “bathe” can, however, express non-reflexive events, as with the transitive variant I bathed the dog in English. Lexical reflexives in English are also often compatible with (redundant) grammatical reflexivity, as in I bathed myself (emphatic interpretations, which we discuss below, may be more natural here). Here, the reflexivity relation is not only expressed via the semantics of the verb but also the argument structure of the utterance. Grammatical reflexivity manifests itself either via marking on the verb (a) or marking on the arguments (b):

In (3a), the reflexive marker {-un}² is a verbal suffix. As for (3b), Faltz (2016/1977) observes that the reflexive pronoun “zich” in (3b) links the object of the verb to the subject, marking that the do-er and the patient are the same referent.

There is also variation within the categories of verbal and argumental reflexivity marking. Verbal reflexivity includes affixes like -un- in Turkish (3a) and clitics, such as the French se in s’habiller (“to dress oneself”). As for argument marking, we observe both free standing reflexive pronouns such as zich in (3b) as well as bound morphemes that shift a stem to a reflexive meaning (e.g., -self in English). It is not uncommon for languages to exhibit multiple reflexive strategies, or to combine them as part of a complex reflexive construction.

What unites these different strategies is that they are all subject to certain structural constraints. First of all, a reflexive requires an antecedent for co-reference, and the relationship between them is often called binding. Argument reflexives require being bound by a potential co-referent (4a-b) and are usually restricted as to what can bind them within what structural domain and/or configuration (4c-d), e.g., intra- or inter-clausal (see, among others, Chomsky, 1981; Reinhart and Reuland, 1993; Faltz, 2016/1977).

The literature on reflexives has long attempted to unify the structural constraints on the reflexivity. However, pinpointing these constraints isn’t always an easy task and is further complicated by the fact that reflexives are commonly observed to serve other functions (Reinhart and Reuland, 1993; Faltz, 2016/1977; Déchaine and Wiltschko, 2017 i.a.) that may not be subject to the same structural constraints as a traditional reflexive. Emphatic anaphors (5a) refer to one of the participants within the event, but their function is to put focus on or set apart a particular participant without affecting the argument structure of the verb. Logophors (5b), on the other hand, are anaphoric items that can get their co-reference from outside of the structural domain of a traditional reflexive. Thus, emphatic markers (5a) and logophors (5b) still express co-reference, which makes them anaphoric. However, they do not necessarily express co-reference between arguments of the predicate, so they are not reflexive.

Such multi-functionality has already been observed for KENDİ (Zeshan, 2002; Sevinç, 2006; Dikyuva et al., 2017), which we discuss in more detail below. However, what we first aim to show in this paper is that KENDİ can be used as a traditional reflexive. Before we move on to discussing that, however, we first provide a review of the existing reflexivity literature on sign languages.

As in other domains of linguistic structure, research on reflexivity in sign language is limited. However, the research that has been done observes key similarities across modalities. Sign languages, too, employ two main strategies to mark reflexivity: on the verb (6a), or on the argument of the verb (6b):

The verbal reflexivity of (6a) is expressed via the spatial agreement markers, indicated by the subscripted ‘‘a’’³. Here, the direction of the object agreeing verb’s path movement agrees with the subject, linking the grammatical object and the subject of the sentence in co-reference. Kimmelman (2009) also reports argumental reflexive pronouns in NGT, glossed as ZELF in (6b), similar to English -self. This parallelism across modalities is not unexpected considering that reflexivity reflects event participant structure, which may be conceptually and linguistically fundamental. This parallelism aside, what sign languages bring to the table is the way they use space for reference, which may lead to modality-specific effects on reflexivity. The role of space in modulating reflexivity is evident in (6a). These reflexivization strategies are not specific to the exemplified languages. Table 1 below shows various reflexivization strategies attested in sign languages.

The multi-functionality of reflexive markers is also relatively well-documented in sign languages. As an example, SELF in ASL can also functions as copula (7a) and as an emphatic marker (7b):

All in all, the existing work, though limited, shows that reflexivity in sign languages is compatible with certain aspects of existing typologies. However, it is also the case that sign languages can provide unique and novel data to further explore the phenomenon of reflexivity, including how usage of space affects referential relations (Schlenker, 2018). Before turning to what TİD shows us about reflexivity, we first briefly explain how we collected and analyzed data in this study.

Much like traditional reflexives are semantically dependent on an antecedent, clitics are morphologically and phonologically dependent on a host. They cannot bear stress, and often come in a reduced phonological form, as exemplified by the reflexive clitic m’ in (18), which is a reduced form of the pronominal me. As the gloss suggests, the reflexive clitic m’ forms a phonological unit with its verbal host (auto-suggère). Moreover, though French is typically a postverbal object language and m’ is expressing the reflexive object of the predicate, the clitic appears in a preverbal position. Thus, the clitic has characteristic properties in terms of its morphophonology and its morphosyntactic distribution.

Turning to TİD, a default SOV language, note first that the typical position of object KENDİ is in the standard pre-verbal position, as illustrated in (9a) and similar examples. This distributional pattern is not inherently at odds with a clitic analysis, but if KENDİ were a clitic, we would expect it to form a phonological unit with a preceding or following element. However, as illustrated in (19), KENDİ can be morphophonologically separated from the verb (its following element) by an intervening adjunct (as illustrated in the repository video, the adjunct intervening in (19a) continues throughout the production of the verb, while the adjunct in (19b) is clearly sequential):

KENDİ is also morphophonologically independent from the elements that precede it. For example, KENDİ is adjacent to its antecedent, ELVAN in (19a), but separated from ELVAN by the intervening adverb NOW in (20).

Another piece of evidence that KENDİ is not a clitic is that it can dislocate to the left periphery, giving rise to a contrastive topic meaning (21).

In TİD, contrastive items are often observed with an eyebrow raise that is articulated simultaneously with the contrast-associated item (Gökgöz and Keleş, 2020 section 4.2). It is also the case that non-manual spread (annotated with the line above the manual sign glosses) (21) in sign languages has been associated with marking phrasal boundaries (Pfau, 2005, 2006). The non-manual marker associated with focus in (21) is not spreading over the subject ELVAN. Therefore, (21) not only shows that KENDİ can be linearly dislocated from what would have been its most plausible host, but also shows that it forms its own prosodic unit. This, then, backs up the narrative that KENDİ is a morphophonologically independent form.

Comparing KENDİ to other clitics that have been documented in TİD -namely, the clitic form of negation—also reveals differences. Zeshan (2003, 2004) and Kelepir (2020d:3.5.1.1) observe that manual negation in TİD can occur as a free form (Figure 2) or a clitic (Figure 3)⁵. In its cliticized form, negation loses its syllabicity (reduced movement, shorter duration) and assimilates to the location of its host, instead of the neutral signing position used in Figure 2.

Unlike clitic negation, the signing position of KENDİ does not get assimilated to that of its host. This could be due to the body-anchored nature of KENDİ. However, there are other ways KENDİ differs from the negative clitic. Building on Zeshan and Kelepir’s analyses, Gökgöz (2009) found that the non-manual marker for negation, a head tilt, patterns differently with the clitic and non-clitic form. With cliticized negation, the head tilt associated with negation spreads onto its morphophonological host, as indicated by the line above KNOW and cliticized ^NOT in Figure 4. With non-cliticized, free negation, however, the non-manual marker only spreads over the negation itself (Figure 5).

Here, too, (21) provides the relevant evidence: we might expect the eyebrow raise in (21) to spread onto ELVAN too, had KENDİ cliticized to it. Thus, based on evidence from intervening items and the properties of clitics in TİD, the relationship between KENDİ and preceding or following elements is linear adjacency, not morphophonological dependency, as would be expected of a clitic.

Bound noun reflexives are attested in languages such as in Halkomelem, where body part nouns attached to the predicate can be interpreted as co-referential with an argument, either the subject in an intransitive verb form (22a) or the object in a transitive verb form (22b). In the case of the predicate with an intransitive marker (-em) in (22a), only the reflexive interpretation is possible.

Data like that presented in the previous section also argue against a bound noun analysis of KENDİ’s. KENDİ does not display the morphophonological characteristics of a bound element.

Our next step is to check if KENDİ is a verbal agreement marker. As discussed earlier, marking reflexivity on the verb is a commonly employed phenomenon in signed (and spoken) languages. Moreover, sign languages often make use of space for modulating agreement, and we know that TİD is a sign language that marks agreement spatially (Gökgöz et al., 2020 section 2.1.2.3.1). Spatial markers on predicates parallel agreement markers in tracking the event participants [see contributions to Lillo-Martin and Meier (2011) for discussion]. In fact, even intransitive predicates with a single argument can spatialize this way (Costello, 2016). However, we argue that this is not what KENDİ is doing for two reasons: (i) it marks reflexivity without being assigned a locus and (ii) still allows the verbal reflexive agreement marker to appear (if compatible with the predicate in general; Kelepir, 2020b section 3.1).

Before elaborating on those arguments, we first provide an example of a reflexive that is of the agreement marker type in Shona. The reflexive -zví in (23) has the distribution of an object agreement marker in the morphological template of Shona⁶ (note that Shona also has a non-reflexive -zvì that differs in tone from the high-toned reflexive -zví):

As discussed above, KENDİ, as compared to -zví, does not form a morphophonological unit with the verb. Indeed, KENDİ is not even obligatorily adjacent to the verb, as one might expect of an agreement marker in general. Moreover, KENDİ is body-anchored and does not make use of an assigned locus in the signing space. Thus, it’s quite unlike how agreement is marked in TİD (Gökgöz et al., 2020:2.1.2.3.1) and in other sign languages (Cormier, 2012:124–125; Sandler, 2012: 44–45). Finally, if KENDİ were a kind of less common agreement marker in language that does not form a unit with the predicate it’s marking “on”, and an almost unattested kind of agreement marker in sign language that does not make use of space, we would expect it to show up post-verbally, because that’s where functional items typically occur in TİD (Gökgöz, 2020).⁷ The default position of KENDİ, however, is preverbal. Therefore, as above, both cross-linguistic and language-internal patterns argue against this analysis of KENDİ.

As for the intransitivizer category, Déchaine and Wiltschko (2017) point to the -iso suffix in Cree as an example. They observe that patient/object marking is absent on the verb when the reflexive marker is present (24a), but present then the verb is used non-reflexively (24b):

As described above, TİD and other sign languages can incorporate spatial locations to mark agreement. TİD also displays object agreement on some verbs through palm orientation (Kelepir, 2020c section 3.1.1.2). Figures 6, 7 illustrate how palm orientation marks object agreement in TİD. In Figure 6, the object is 1st person and the palm orientation is toward the signer, while in Figure 7 the palm orientation is toward an established spatial locus away from the signer’s body, the orientation of a 3rd person marker.

This palm orientation agreement marker is present in non-reflexive predicates like PROTECT or SUPPORT. Crucially, however, it is also present in reflexive usages of the same predicates. This is illustrated for reflexive and non-reflexive usages of the predicates PROTECT and SUPPORT in (25). In a sentence like (25a), the 3rd person palm orientation is toward a spatial locus away from the signers’ body, whereas in (25b) there is the reflexive marker KENDİ as well as palm orientation toward the signer’s body, just like Figure 6.⁸ Note that the signer has omitted the subject agreement marker in (25b), as has been observed elsewhere in sign languages (Padden, 1988, 1983).

Thus, KENDİ does not manipulate the argument structure like the intransitivizer -iso in Cree does, and we can conclude that KENDİ is not an intransitivizer because it co-occurs with a marker of transitivity, the palm orientation marker of object agreement.⁹

So far, we have seen that KENDİ does not align with the patterns of a clitic, bound noun, intransitivizer, or agreement type of reflexive. However, there is another kind of reflexive in the typology: a DP reflexive.

DP reflexives are basically reflexives that act like any DP, except for the specific dependency relation that they require a co-referring antecedent. Previous examples have shown that KENDİ is a reflexive that behaves like any DP (object) argument of the verb: it occupies an A position as the canonical object in the sentence and, moreover, can be dislocated to an A′ position the left periphery.

Recall that Déchaine and Wiltschko’s (2017) typology includes syntactically parallel items for each of the reflexive types. For DP type reflexives, the parallel they observe is possessors. Note that in some DP type reflexives, the connection with possessors is transparent; the reflexive form in English contains a possessive: myself. For KENDİ, the connection with possessives is twofold. One, KENDİ can be combined with an overt possessor like the English “xself”, as in (27).

Two, KENDİ itself can also be used as an independent possessive (28), though this usage isn’t very common.¹⁰

There are also cases of complementary distribution, where possessive usages of KENDİ block another possessive:

With respect to reflexive multi-functionality, additional functions of the DP type reflexives that Déchaine and Wiltschko (2017) observe are serving as emphatic anaphors (30a) and a logophors (30b):

As a reminder, emphatic anaphors are used to focus some aspect of the event or the referent’s role in the event, often from a non-argument position, and logophors are anaphoric elements that seemingly skirt binding principles by getting their co-referent from a discourse antecedent outside of their local domain. Thus, emphatic anaphors and logophors are reflexive forms that are in grammatical positions that are not associated with reflexivity (emphatic anaphor), or have non-prototypical antecedents (logophor). We are currently investigating whether a logophoric usage of KENDİ possible, but we already know from previous observations by Zeshan (2002) and Sevinç (2006) that KENDİ does indeed function as an emphatic anaphor. We elaborate on the emphatic anaphor usage of KENDİ in “The Function of KENDİ Beyond Traditional Reflexivity.”

To summarize, our fieldwork reveals that KENDİ in TİD can function as a traditional reflexive, and its previously observed usage as an emphatic anaphor is connected to its status as a DP-type reflexive, similar to English xself, and unlike other syntactic categories of reflexives discussed above. Moreover, we have seen that KENDİ shares other features that characterize DP-type reflexives, such as a structural parallelism with possessors. In the next section, we further explore the traditional reflexive usage of KENDİ, providing a more detailed description of its binding domain and its antecedents.

A defining characteristic of a traditional reflexive is that it requires an antecedent. Whether this antecedent must be overt or not depends on whether the language in question allows null arguments. This is illustrated for KENDİ by the contrast in (31), where (31a) contains an overt antecedent (ELVAN) for KENDİ and is grammatical, but (31b) lacks an overt antecedent and is ungrammatical.

Judgments like those for (31a) and (31b) are in “out of the blue” contexts. However, language is rarely used in truly out of the blue contexts. Given that TİD permits null arguments, we would predict that KENDİ can be licensed by covert antecedents. The data in (32)-(33) illustrate that this prediction is borne out (“e” glosses the position of the null argument):

These data show us that the required antecedent for KENDİ can be null arguments that are licensed by earlier portions of the discourse (32) or by contextual salience (33). Like KENDİ, the null argument that binds the anaphor, e_i, is subject to its own licensing conditions (see Kayabaşı et al., 2020 for a detailed discussion of null arguments in TİD).

These data affirm that KENDİ patterns like a traditional reflexive in requiring an antecedent, though independent patterns of null argument licensing in the language mean that this antecedent need not be overt. These findings are in line with existing research on null arguments and reflexive pronouns in other sign languages (Lillo-Martin, 1986; Bahan et al., 2000; Koulidobrova, 2012; Kimmelman, 2018; Kayabaşı et al., 2020 i.a.).

We have already briefly described KENDİ’s relationship to overt and covert antecedents. In this section, we will talk about the structural logistics of KENDİ and discuss suitable structural positions for an antecedent to bind KENDİ, both in terms of hierarchy and proximity.

For a traditional reflexive like KENDİ, Principle A (Chomsky, 1981) is usually interpreted as requiring c-command of the reflexive by its antecedent. C-command is a structural relationship between two nodes, X and Y, neither of which dominates the other, but where every branching node that dominates X, also dominates Y (Reinhart, 1976; Büring, 2005b, i.a.). Figure 8 illustrates how A can bind B but not the other way around since A c-commands B but not vice versa.

Crucially, there are patterns in language, like reflexivity, that show sensitivity to structural relationships like c-command. This is illustrated by the relationship between Charlotta and her/herself in (34). Note that the examples are structurally identical and Charlotta precedes the intended co-referential DP in both. Because Charlotta is “buried inside” the subject DP, however, it does not c-command the her/herself in either example. The resulting grammaticality differs depending on the use of a non-reflexive vs. reflexive pronoun. The non-reflexive pronoun in (34a) is grammatical because such pronouns are only optionally co-referent and do not need to be bound. However, a traditional reflexive is obligatorily co-referent, so the herself in (34b) does need to be bound. Consequently, expressing the co-reference via a reflexive, as in (34b), is ungrammatical.

A similar example for TİD is presented in (35). Here, there are two possible antecedents for KENDİ, the possessor ELVAN_i and the full possessive DP [ELVAN_i POSS_{3sg_i} SISTER]_j. Only [ELVAN_i POSS_3sg SISTER]_j, however, is in a c-commanding relationship with KENDİ. As predicted, this is the only DP that can bind KENDİ:

The fact that ELVAN_i cannot bind KENDİ tells us that TİD is no exception to the rule that the relation between the antecedent and a traditional reflexive is structurally determined, and one that cannot be characterized by linear precedence. Though critiques have raised questions about whether c-command accurately characterizes the structural constraints imposed on binding, what is relevant here is that (35) shows that there is a structural constraint at play. So far, we described what type of antecedents KENDİ can take and showed that it can be bound by both overt and null antecedents. Moreover, we have shown that the relation between KENDİ and its antecedent is subject to some type of structural constraint. To assess the proximity aspect of the antecedent-reflexive relation, we turn next to the pattern of KENDİ in subordinate sentences. In (36), KENDİ is in the object position of the subordinated predicate SEE, and co-reference with the subject (ELVAN) of the matrix predicate WANT is possible (either via direct binding or mediated by an intervening null argument, itself co-referent with ELVAN):

Note, however, there are no other viable antecedents in (36). In (37), we see how KENDİ behaves when multiple possible antecedents are present in the sentence.

Here, too, KENDİ is in the subordinate object position, but now two overt and distinct referents are available to serve as antecedents: the matrix predicate subject, IX_{1sg_j}, and the subordinate subject, ELVAN_i. The only permitted antecedent for KENDİ in (37) is the closer subordinate subject. Note that the WANT type verbs in (36–37) are usually associated with non-finite sentential complements. In (38), we see the same binding patterns with THINK, a verb type that is often associated with finite sentential complements. There is currently very little known about finiteness in sign languages (and almost nothing known about this phenomenon in TİD) within the existing literature, but the patterns are the same across these predicate types.

Together, (35)-(38) show us that KENDİ is sensitive to familiar, structural binding constraints of hierarchy and proximity. Having elaborated a bit on the co-referential relations of KENDİ as a traditional reflexive, we will now turn to KENDİ’s function beyond traditional reflexivity, emphatic anaphoricity.

Emphatic markers are anaphors that co-refer to a participant of the given event to cast focus on it and to contrast it from other participants in a possible set of participants (Kemmer, 1995; Stern, 2004, i.a.). They usually occupy non-argument positions, and they do not express reflexivity. Focusing on English, Ahn (2010) notes two distinct usages of the emphatic anaphor xself (here, himself), exemplified in (39). In (39a), himself creates argument focus on its event participant antecedent, a ‘‘specifically John and not someone else’’ meaning. In (39b), however, himself modifies the event (not an argument), and emphasizes that John performed the given event without help or the contribution of another causer/agent.¹¹

He labels the interpretation that arises from modifying the DP (39a) as “adnominal”, and the one that arises from modifying the DP as “adverbial” (39b).

In (40) we see the same two-way distinction with the emphatic function of KENDİ:

As the translations show, (39a-40a) and (39b-40b) resemble one another with respect to the emphatic contribution of himself and KENDİ.

Notice, though, that the different interpretations of himself in (39a-b) also correspond to different syntactic positions. Interestingly, the two distinct interpretations in (40) are possible with KENDİ in the same linear position. However, KENDİ in its emphatic function can also occupy different positions in the sentence (41), including the rightmost position (41c), which is ungrammatical for a traditional reflexive KENDİ as (42) shows (note also that emphatic KENDİ in (41) is optional):

Moreover, as the translations for (41a) and (41c) indicate, emphatic KENDİ has both adnominal and adverbial interpretive possibilities in each of these possible positions.

However, Ahn identified other diagnostics that distinguish adnominal and adverbial emphatic anaphora, such as: denying the event, specificity, thematic roles, context-free acceptability, stative verbs, prosody and stress. Here, we will use three of Ahn’s diagnostics to probe emphatic usages of KENDİ: (i) denying the event, (ii) specificity, and (iii) thematic roles of the co-referent (research is ongoing, and we do not have enough data to conduct all the diagnostics suggested by Ahn (2010) at this stage).

Denying the adnominal emphatic requires denying that the focused referent was, in fact, the relevant event participant at all; in (43a), this is accomplished by asserting that someone else did the activity. Denying the adverbial emphatic, however, doesn’t mean denying that the individual did the thing, just that they did the thing alone or without help. The felicitous adverbial denial pattern is given in (43b).

In (44), we see that both denials are felicitous with emphatic usages of KENDİ in each of the positions identified in (41)—that is, all three positions are apparently compatible with both adnominal and adverbial interpretations of KENDİ.

In addition to differences in deniability patterns, Ahn (2010) also observes that adnominal emphatic anaphors require a specific (but not necessarily definite) referent (45a,c), whereas no such restriction holds for adverbial emphatics (45b,d). “Specificity” here refers to the event participant being a unique entity as opposed to a generic one.

As for TİD, we again see a slightly different pattern. Non-specific referents—namely, WHICH CHILD (46) and SOMEONE (47)—are semantically compatible with the adverbial interpretation of emphatic KENDİ, as in English, but they are also compatible with adnominal interpretations. Here, too, these observations hold for KENDİ in multiple positions.

The third and final diagnostic that we will cover from Ahn (2010) concerns the thematic roles that are compatible with the emphatic anaphor. He observes that adverbial emphatic anaphors are only compatible with volitional and agentive subjects, whereas adnominal emphatic anaphors have no observed thematic role restriction. However, (48) shows not only that KENDİ is perfectly acceptable and grammatical with a non-volitional inanimate subject, but that it is compatible with an adverbial interpretation (EXTCL glosses an extension classifier sign):

So far, our assessment of the emphatic usages of KENDİshows us that they have more flexibility than (i) their traditional reflexive counterparts in TİD and (ii) their emphatic anaphor counterparts in English. However, it is important to note that the semantics and pragmatics of TİD is very understudied, as is the cross-linguistic typology of emphatic anaphors. Future investigation can investigate the source of these cross-linguistic differences and further assess if there are differences between adnominal and adverbial interpretations of KENDİ.

In this section, we will explore cases where KENDİ can be ambiguous between a traditional reflexive and an emphatic anaphor. Such cases are possible when the predicate of a sentence allows for object drop or null objects and a potentially reflexive event. The predicate VOTE GIVE can take a DP (49a) or PP (49b) as an object, and it also can be used intransitively (49c):

Thus, in (50), it is not clear whether KENDİ occupies an argument position as a traditional reflexive, or whether it’s an emphatic anaphor, and within the latter both adnominal and adverbial interpretations are possible.

We assume here that this is a case of structural ambiguity: despite KENDİ surfacing in the same linear position, it occupies different positions in the sentential structure. The intended interpretations can be contextually disambiguated but they can also be structurally disambiguated¹². Examples of structural disambiguation are illustrated in (51), where the presence of a preposition (51a) or a separate DP (51b) unambiguously express the intended transitive interpretation (recall from (49a-b) that VOTE GIVE can take its object as a DP or PP, which is why the FOR is “optional”).

The above examples are cases where we see a given token of KENDİ that is compatible with different types of interpretations. However, there are also cases where we see multiple tokens of KENDİ within the same utterance, as in (52).

For the first interpretation of (52), it might be the case that the doubling of KENDİ functions in a similar way to focus doubling (Makaroğlu, 2012). The second interpretation, however, provides clear evidence that KENDİ can serve distinct functions, which can be combined within the same sentence.