Transitive Unergatives in Pazar Laz

The aim of this study is to show that all unergatives in Pazar Laz involve an overtly filled object position and behave simply on a par with regular transitives, as the availability of an initiator is strictly dependent on the availability of an undergoer argument in line with the Neo-Burzio Dependency proposed for Georgian by Nash (2018). We will show that differently from Georgian, in Pazar Laz the availability of an initator and that of an undergoer actually exhibit a mutual dependency for all eventive verbal predicates, which extends the transitive pattern also to unaccusatives. As the language requires a strict co-dependency between the undergoer and the initiator, all eventive verbal predicates project both an initiator and an undergoer position. Consequently, the language does not syntactically differentiate between unergatives and unaccusatives and lacks the true unaccusative patterns, where no initiator is available syntactically. We will further argue that the co-dependency requirement for initiators and undergoers is in line with Ritter & Rosen (2000)’s typology, where languages are classified as Initiation languages or Delimitation languages based on whether they define an event in terms of its initial bound or its terminal bound. As a very conservative Initiation-language, Pazar Laz has to syntactically express the initiator in all types of eventive predicates, which itself is dependent on the availability of an undergoer argument in line with the Neo-Burzio Dependency.

When we focus on unergative predicates in PL, we mainly see two patterns.One pattern, as given in (1b) and (2), mainly comprises verbs of sound/smell/light emission.The verbal root can be onomatopoeic.The subject can be animate or inanimate, but is always marked with ergative case: 2 (2) a. K'at'u-k p'iy-u.cat-erg meow-3pst 'The cat meowed.' b.
Ayna-k farfal-u.mirror-erg shine-3pst 'The mirror shined.' (4) Neo-Burzio Dependency: Voice assigns Agent role to its argument iff it selects an argument-selecting complement.(Nash 2018: 22) As a result of Neo-Burzio Dependency which Nash (2018) argues to hold in Georgian, the thematic property of the external argument is strictly correlated with the thematic property of the predicate.If the predicate selects a lower argument, the external is interpreted as an Agent or a Causer, and a dynamic/eventive reading becomes possible.
We will argue that this transitivity requirement stemming from Neo-Burzio Dependency in PL is not restricted to unergatives but also extends to unaccusatives, which are also eventive.In parallel to Öztürk & Taylan (2017), we will show that there is a strict co-dependency between the undergoer and the presence of an initiator.In other words, if the structure involves an undergoer, the initiator should also be available and if there is an initiator the undergoer position should also be activated.This condition implies that all eventive verbal predicates project transitively in PL and that the language does not syntactically differentiate between unergatives and unaccusatives.As unergatives and unaccusatives have the same architecture, there are no true intransitives in the language.Thus, PL lacks the true unaccusative pattern (e.g.anticausatives) found in languages like English, where no initiator is available syntactically.We will further argue that the transitivity requirement for eventive predicates is a consequence of how events are defined in PL.In line with Ritter & Rosen (2000)'s typology, where languages are classified as Initiation languages or Delimitation languages based on whether they define an event in terms of its initial bound or its terminal bound.As a very conservative Initiation-language, PL has to syntactically express the initiator in all types of eventive predicates, which blocks the true unaccusative patterns.
The paper is organized as follows.Section 2 would introduce the morphosyntactic properties of unergatives highlighting the parallelism between transitives and unergatives in PL.Section 3 will focus on the pronominal clitic nature of the morpheme i-in PL, which will play a crucial role in establishing the transitivity of unergatives.Section 4 will discuss the unaccusatives formed with -(u)r and argue that they are also not true unaccusative constructions but are syntactically transitive.Section 5 will focus on agentive unergatives and verbs of emission and claim that they also have a transitive syntax on a par with unaccusatives in the language.Section 6 will address the question why all eventive predicates require a transitive syntax based on Ritter & Rosen (2000)'s typology.Finally, Section 7 will introduce our concluding remarks.

MORPHOSYNTACTIC PROPERTIES OF UNERGATIVES IN PL
When we take a closer look at the morphosyntactic properties of unergatives in PL, we observe that the choice of ergative case for the subject is not the only parallelism between unergatives Another parallelism between transitives and unergatives that we observe in (5) is the choice of thematic suffixes (TS, as extensively discussed in Taylan & Öztürk (2014); Öztürk and Taylan (2017).As seen in ( 5a) and (5b), transitives and unergatives share the same TS -am, while the unaccusative in (5c) selects a different TS, i.e. -u(r). 5There are four TSs in PL; -e(r)/-u(r)/-am/um, which are obligatory in the imperfective both in present and past tenses, but do not surface in the perfective.Both eventive verbs and statives are compatible with TSs in PL and the choice of TS is sensitive to the lexical aspect and the argument structure of the verb used.As purely functional morphemes denoting imperfectivity, TSs in PL never occur in derived nouns such as masdars. 6Ramchand & Svenonius (2013) and Ramchand (2017) show that perfective and imperfective belongs to different functional zones.While the perfective belongs to a higher zone, imperfective (e.g. the progressive in English) which imposes selectional restrictions on the lexical aspect of the verb and which has a close connection with the verb and its arguments belongs in the lower verbal zone.They argue that imperfective is located in the projection called EventP, which is introduced above InitP where the initiator is located in a Ramchandian phrase 3 Note that we are only providing the singular set of agreement markers as plural morphology has an independent exponence for first and second persons and exhibits fusion with person for third person, thus, it exhibits a highly complex distribution (cf.Demirok 2013).The singular agreement paradigm suffices to make our point regarding the agreement split between unergatives and unaccusatives.

4
The suffix -n is based on the third person singular form of the copula on.This split between the agreement suffixes is similar to the auxiliary (be vs. have) selection differences observed in unaccusatives and unergatives in languages like Dutch and Italian.

5
Georgian, a close relative of Laz, also marks different verb classes via TSs, such as -eb, -ob, -av.In contrast to the pattern we observe in PL, Lomashvili (2010) argues that there is no principled reason as to why a verb selects a particular TS and that these are simply used for morphological well-formedness in Georgian (Lomashvili 2010: 75).

6
Note that TSs can appear as part of the masdar forms of verbs in Georgian (cf.Nash 2018).

PERSON PERSON SUFFIXES
PAST PRESENT structure (cf.Ramchand 2008). 7In parallel to Nash (2017a), Taylan & Öztürk (2014) and Öztürk &Taylan (2017) also take the TS -the imperfective marker as the head of this projection. 8Dominating the predicate domain including all the arguments, TSs are selected based on the argument structure and the lexical aspect of the verb, as seen in ( 6). (6)
In PL, agentive unergatives as in (7a) and unergative verbs of emission as in (5b) take TS -am, in parallel to transitives as illustrated in (5a) and (7b).Note that in terms of lexical aspect, -am is compatible with atelic activities, as well as telic accomplishments.
Transitive Ahmedi-k dişk'a mo-ğ-am-s.Ahmet-erg wood pv-bring-ts-3prs 'Ahmet is bringing the wood.' As seen above, the TS that unergatives select are in parallel to the ones selected by transitives and are clearly differentiated from the TS -u(r) that unaccusatives select as seen in ( 8).Unaccusatives which always require the TS -u(r) include achievements (8a), degree achievements (8b) and verbs of directed motion (8c), which all express a change of state for the undergoer subject: (8) a.
Bere nca-şe ey-ul-u-n.child tree-all pv-climb-ts-3cop 'The child is climbing the tree.' Even though in terms of case, agreement and TS choice unergatives align with transitives and are clearly differentiated from unaccusatives, there is one particular construction which all 7 Ramchand (2008) decomposes verb meaning into several projections which match different subevents.In her architecture, InitP introduces the causation event and licenses the external argument similar to vP.The DP in the Specifier of InitP is the subject of cause, that is the initiator.ProcP specifies the nature of the change or process and licenses the entity undergoing change or process.This projection is available in all dynamic events.The spec of ProcP is the subject of the process known as the undergoer.Finally, ResP gives the 'telos' or 'result state' of the event and licenses the entity that comes to hold the result state.The DP in its Spec is the subject of result, namely the resultee (Ramchand 2008: 47-48).Note that in this framework, a single DP can be associated with more than one subevent projections, for example, in internally caused events a single DP can be simultaneously the subject of InitP, ProcP and ResP.This is reminiscent of Hornstein (1999) where it is possible for a single DP to be associated with more than one theta positions.Note that in the Ramchandian model, objects which are not subjects of subevents but are part of the description of the verb, such as objects of stative verbs, incorporated or cognate objects are called rhemes and are base-generated in the complement projections of the relevant subjevent projection.

8
The EventP assumed in the Ramchandian phrase structure is different from the ones proposed by the Harley (1995) and Borer (2005).Both Harley (1995) and Borer (2005) associate this projection with the external argument and the originator of the event, however, such arguments are associated with InitP in the current model, and EventP is associated with imperfective aspect, specifically with the progressive in the context of English as discussed in Ramchand & Svenonius (2013) and Ramchand (2017).

Öztürk
Glossa: a journal of general linguistics DOI: 10.5334/g jgl.828 the three verb types are compatible with.This is the active impersonal construction (AIC) in PL (cf.Öztürk & Taylan 2017).As we will show below, PL does not have separate intransitive constructions such as passive, inchoative, middle and anticausative constructions.Instead AIC is the construction available to meet these readings in the language, which we will argue to be a transitive construction in the following section. 9As it will become clearer shortly, AIC is not the counterpart of non-active voice found in languages like Greek or Albanian (cf.Kallulli 2007).Instead it is an active impersonal construction found in languages like Breton, Irish, Polish and Icelandic (cf. Legate 2014).This construction requires the valency marker i-, which we will show to be a pronominal clitic, and in the imperfective, the TS -e(r) is used.The subject in this construction is nominative and again the -n set agreement based on the third person copula is used in parallel to unaccusatives.The AIC denotes a strictly externally caused agentive reading referring to a human agent: (9) a. Dişk'a m-i-ğ-e-n.wood pv-val-bring-TS-3prs 'Someone is bringing the wood.' b.
Oxori i-tzopx-e-n.house val-build-ts-3prs 'Someone is building the house.' As seen in ( 9), transitives are compatible with AICs, so are unergatives and unaccusatives as shown in (10a) and (10b), respectively.The use of unergatives and unaccusatives with AICs lead to an impersonal passive-like reading: (10) a. Unergative Table 2 summarizes the morphosyntactic properties of different verb types in PL as also extensively discussed in Taylan & Öztürk (2014) and Öztürk & Taylan (2017).We observe that each verb type is associated with different case, agreement and TS patterns.As seen very clearly in the table, unergatives pattern with transitives in terms of case, agreement and the TS choice, hence they are highly parallel.We will take this parallelism to argue for the transitive nature of unergatives in PL.On the other hand, there is another significant parallelism that we observe between agentive unergatives and the AIC, that is the obligatory use of the valency marker i-.Unergatives and AIC behave differently in terms of case, agreement and TS choice but they both require i-.We will argue that this marker in both constructions is a pronominal clitic and it will play a significant role in arguing for the transitivity of unergatives so in the following sections, we will focus on the morphosyntactic properties of this marker in PL.
9 Georgian -a close relative of PL, for example, makes use of the suffix -d to derive inchoative unaccusatives.PL has no such derivational morphology to distinguish between passives, inchoatives, anticausatives and middles.

THE VALENCY MARKER i-
As discussed above, the valency marker i-in PL is obligatory in agentive unergatives and also in AICs across all tense and aspect series.Another structure where we find the same marker in PL is the reflexive constructions.As seen in ( 11), there are two ways to form reflexivization in PL.The coindexation between the subject and the object in (11a) can be achieved as in (11b) with the use of a reflexive pronoun çendi 'self' which is borrowed from Turkish kendi 'self'.Alternatively, the verb can be marked with the valency marker i-, which again indicates the coindexation between the subject and the object as in (11c).As seen in (11d), it is not possible for i-to co-occur with çendi: Both in reflexives and in AICs, the verbs marked with i-have one of the overt arguments missing.In reflexives, i-targets the internal argument, whereas in AICs the external argument appears to be reduced.The question is whether i-truly reduces the valency of the predicate yielding a true intransitive or whether it fulfills the argument function itself as some kind of a pronominal clitic keeping the valency of the predicate intact.We will argue for the latter.
Let us first check whether there is any evidence for the clitic status of this marker.As extensively discussed in the literature (Holinsky 1991; Kutscher et al. 1995;Lacroix 2009;Öztürk & Pöchtrager 2011; among others), the verbal complex in Laz can host both preverbal and postverbal markers.It has been observed that the postverbal markers exhibit an agglutinative nature and behave as suffixes, whereas the preverbal material which typically cover sets of person markers, valency markers, negation and modal adverbs on the other hand, compete for limited morphological slots exhibiting a clitic-like behavior.For example, based on a personcase constraint, prefixal agreement markers compete with each other for the same slot as illustrated in (12).While in (12a) the first person subject marker wins over the third person object, in (12b) the second person object wins over the first person subject.Note that in each case only one of the arguments can be realized preverbally on the verbal complex.Having both the subject and the object markers simultaneously leads to ungrammaticality as in (12c): We observe a similar pattern of competition among the valency markers which also include the marker i-.The reflexive valency marker i-in (13a) can also be replaced by a causative valency marker o-when an external causeP is introduced to the structure.Similarly in PL, the i-in AICs is replaced with the applicative marker a-in modal applicative constructions as in (14b) (cf.Öztürk 2012; Demirok (2018)): Öztürk Glossa: a journal of general linguistics DOI: 10.5334/g jgl.828 (13) a.
Ar oxori xordza-s mv-a-rg-e-n.a house woman-dat PV-appl-build-TS-3prs 'A house is being built for the woman.' In the literature, it has been shown that clitics cannot occur in nominalizations or in certain types of participles (cf.Wood 2015 for Icelandic; Kayne 1975; Pesetsky 1995 for Romance).Similary in PL, i-disappears under nominalization as in (15a) and in the formation of stative progressive participles as in (15b): (15) a.
Mbon-eri / *i-mbon-eri, bere-k lebe-pe-muşi ko-gam-i-yon-u.wash-ptcp / val-wash-ptcp child-erg dirt-pl-3poss pv-pv-val-rid.off-3pst'Washing himself, the child got rid off the dirts on him.'Given these patterns we take i-to be a pronominal clitic which stands for an argument in the spirit of Wood (2015), which then cliticizes onto the verbal complex.In the case of reflexives, we take it to occupy the object position standing for the undergoer which is bound by the subject as shown in ( 16).If it is indeed a pronominal clitic which fulfills the object function, this would also explain why i-cannot co-occur with the full reflexive pronoun çendi as illustrated in (11d): In the case of AICs, given that the overt DP stands for the object, i-should be replacing the ergative marked agent.The question is whether we have any evidence for the agentive argument status of i-in AICs which obligatorily denote external causation brought about by an agentive human subject.When we apply the well-known tests for agentivity to AICs, we see that the construction is compatible with an agentive interpretation despite the lack of an overt ergative marked subject.As seen in ( 17), AICs can be modified by purpose clauses, instruments and initiator-oriented adverbs (cf.Öztürk & Taylan 2017): (17) Cami k'asi-te amolva şeni ç'ak'uç'i-te i-t'ax-e-n.window intention-with enter for hammer-with val-break-ts-3prs 'The window is intentionally broken with a hammer in order to enter (the place).' The only agentivity test which AICs fail to pass is the use of an agentive by-phrase.Recall that PL does not have separate passive or anticausative constructions and simply makes use of AICs to accommodate these readings.The language simply lacks by-phrases or by itself phrases which are used to identify passives and anticausatives, respectively (cf.Alexiadou,

Öztürk
Glossa: a journal of general linguistics DOI: 10.5334/g jgl.828 Anagnostopoulou, Schäfer 2006). 10We argue that this follows from the presence of the valency marker i-in AICs.As a pronominal clitic i-saturates the external argument of the predicate, disallowing the introduction of another initiator into the structure through by-phrases.Similar to the i-standing for the undergoer in reflexives, the i-in structures like (9) semantically closes the initiator (cf.Chierchia 1995), hence stands for the external argument.
An anonymous reviewer suggests that AIC might be similar to the Icelandic grammatical object passives which Legate (2014) discusses.Icelandic grammatical object passives are compatible with by-phrases, however, this is not the case in PL.AIC are more similar to the impersonals in Polish, Irish and Breton impersonals which Legate (2014) also discusses.These impersonal constructions similar to the AIC in PL are incompatible with by-phrases strictly requiring a human interpretation.Legate (2014) proposes that these impersonals hosts a D element in Spec, VoiceP referring to humans saturating the external argument of the verb.Given this referential element, for example, Irish impersonals can antecede a reciprocal as seen in ( 18b), fully in parallel to clauses with overt subjects as in (18a): Similar to the Irish impersonals, we see that the pronominal clitic i-can antecede reflexives in PL.Hence they act as syntactically active referential subjects.As seen in (19a), the reflexive anaphor çendi in the object position needs to be bound by the ergative subject to be licensed.Similarly,in (19b) in AICs, çendi in the object position is again licensed despite the lack of an ergative subject.We argue that it is the pronominal clitic i-in Spec, InitP which binds the anaphor çendi in Spec, ProcP:11 (19) a.
Ali-k çendi-Ø var msk'v-am-s. 11Ali-erg self-nom neg praise-ts-3prs 'Ali does not praise himself.' As it is well-observed in the literature beside by-phrases and by itself phrases passives and anticausatives can be differentiated from one another via other PP adjuncts which can bring out the causal structure such as from-phrases (cf.Levin & Rappaport-Hovav 1995;Kallulli 2007;Alexiadou, Anagnostopoulou & Schäfer 2015).PL does not allow for from-phrases either.The only adjunct type it allows for are the DPs marked with instrumental case as seen in (ia).Causal events are incompatible with instrumental case as in (ib) and they can only take locative case denoting co-temporality (ic), but not necessarily causation given that while (iia) would be a semantically compatible paraphrase of (ia), native speakers do not accept (iib) as the paraphrase of (ic): Cam-epe bomba-te i-t'ax-u.window-pl bomb-with val-break-3pst 'Windows were broken with the bomb.' b.
AIC Çendi-Ø var i-msk'v-e-n.self-nom neg val-praise-ts-3prs 'One does not praise himself.' In parallel to Öztürk & Taylan (2017), the representation we assume for AICs is given in (20), where the pronominal clitic i-occupies the external argument position.Hence it starts out in an argument position but then cliticizes on the inflected verb. 12As the clitic i-stands for an initiator in Spec, InitP, this implies that AICs always involve an initiator in the argument position and cannot consitute anticausatives or passives, which lack an initiator. 1320) Another interesting piece of evidence for the presence of an agentive layer in AICs can be seen in the translation of such sentences into Turkish by Turkish-Laz bilinguals, which was also discussed by Öztürk & Taylan (2017).Even though Turkish morphologically differentiates between passives and anticausatives, Laz informants translated the examples with AICs into Turkish with passive morphology which allows for an agentive reading and systematically rejected the anticausative version.For example, the corresponding Turkish form of the PL sentence in (21) can only be (22b), but not (22a) according to our bilingual informants.Yet Laz informants could never introduce a by-phrase to the construction in ( 21), even if they match this reading with the passive in Turkish, which allows for a by-phrase as in (22b).Given these patterns we argue that AICs involve an InitP whose Spec hosts the pronominal clitic i-which blocks the introduction of a by-phrase: Another case where we have i-surfacing is the agentive unergatives as discussed in the introduction.Such verbs always require the valency marker i-on the verbal complex in addition to an overtly ergative marked agent: (23) a.
Ali-k i-çaliş-am-s.Ali-erg val-work-ts-3prs 'Ali is working.' Given the above discussion regarding the argumental nature of the pronominal clitic i-, the question is whether i-in (23) also fullfills a similar function, that is, whether occupying the undergoer position, it is coindexed with the ergative subject similar to the case of reflexives as depicted in (24).In other words, do unergatives with i-have a transitive syntax in PL?: Before answering this question affirmatively we will first take a look at another intransitive pattern available in PL, namely, the unaccusatives formed with the TS -u(r) denoting a change of state from the perspective of transitivity, which will pave the way for us to argue that unergatives are also transitives in PL.

UNACCUSATIVES WITH -U(R)
Recall that unaccusatives compatible with the TS -u(r) in the imperfective typically denote an eventuality where the undergoer has gone through some change of state, as shown in (25).
The sole argument surfaces with nominative case and exhibits -n-set suffixal agreement with the verb: (25) a.
Bere mo-l-u-n.child pv-come-ts-3cop 'The child is coming.' As seen in (25), verbs denoting achievements, degree achievements and verbs of directed motion fall into this group, as extensively discussed by Taylan & Öztürk (2014).The common property of all these verbs is that they express a scalar change in the sense of Rappaport Hovav (2008) for the undergoer.Rappaport Hovav defines scalar change as 'one which involves an ordered set of changes in one particular direction of values of a single attribute and so can be characterized as movement in a particular direction along the scale' (Rappaport Hovav 2008: 17).Within a Ramchandian phrase structure, these verbs all involve a ResP -the layer associated with change of state, as depicted in (26).The nominative argument is simultaneously both the Resultee and the Undergoer.These verbs at first sight appears to form what one can call the set of anticausatives in PL.However, we will argue for the structure in (26) which not only involves a ProcP and a ResP layer but also an InitP layer: There are two arguments supporting the representation in (26).First, just as it was the case in AICs, it is again possible to detect the implicit initiator with unaccusative verbs which take -u(r) by using initiator-oriented adverbs, purpose clauses and instrumentals, even though they do not involve the valency marker i-unlike AICs ( 27), as discussed in Öztürk & Taylan ( 2017): (27) a.
Ham metali matzindi oyapu şeni ndrukh-u-n.this metal ring make for bend-ts-3prs '*This metal is bending to make a ring.' b.
Yaği xalva oyapu şeni ndgul-u-n.butter halva make for melt-ts-3prs '*The butter is melting to make halva.' c.Noti k'elemi-te Ali-s parti goşinu şeni nç'ar-u-n.note pen-with Ali-dat party remind for write-ts-3prs 'The note is written (in a written state) with a pen to remind Ali the party.' The fact that the verbs with -u(r) in ( 27) exhibits the same pattern with AICs in terms of the use of purpose clauses and instrumentals implies that these constructions also point to the presence of an initiator in addition to the nominative undergoer, even in the absence of the valency marker i-.
As pointed out by one of the anonymous reviewers, one question which arises at this point is whether -u(r) constructions can be analyzed on a par with passive constructions which also allow for external argument semantics, but are syntactically intransitive.In other words, can these constructions still be syntactically intransitive like passives?The second argument supporting the transitive nature of these constructions comes from the data in (28).As seen in ( 28a), these constructions can also license the reflexive çendi and be interpreted with a transitive meaning, given the right kind of context, in the same fashion as the AIC constructions illustrated in (28b).This implies that they are not on a par with passives:14 (28) a.
Çendi i-ndrikh-e-n.self val-bend-ts-3prs 'Somone is bending himself.'Furthermore, it is not possible to use by-phrases to show that these constructions are passive in nature, as the language simply lacks such adjuncts.Also -u(r) is compatible with the verbs of directed motion as seen in ( 25c), which are typically not compatible with passives across languages.Verbs of directed motion can also be used in AIC and both in their use in AIC and in their use with -u(r) and can license the reflexive çendi, not in the adverbial sense, but implying a transitive use.As seen in ( 29) they are compatible with purpose clauses and instruments: (29) a.
Çendi cari oşkhomu şeni araba-te m-ul-u-n.self food eat for car-with pv-come-caus-TS-3prs 'He is coming (i.e.bringing himself) to have dinner by car.' b.
Çendi cari oşkhomu şeni araba-te mo-i-lv-in-e-n.self food eat for car-with pv-val-come-caus-TS-3prs 'People are coming (i.e.bringing themselves) to have dinner by car.' Given these pattern, in parallel to Öztürk & Taylan (2017), we take the constructions with -u(r) as having a transitive construction involving both an initiator and an undergoer layer and we conclude that they are not like passives.At this point one can ask the difference between the use of AIC and -u(r) constructions, given that we argue both are transitive.Both unaccusatives with -u(r) and AICs take nominative marked subjects and -n-set agreement and hence are very similar to one another.However, there is a clear difference in meaning as shown in (30a-b), which was discussed in Öztürk & Taylan (2017).If one wants to highlight the natural property or the state of the undergoer (i.e. the metal has the intrinsic property of bending, e.g.copper, or it is in a bent state), then the verb takes -u(r) as in (30a).The verb in (30b), on the other hand, being marked with the valency marker i-necessarily implies the presence of an external factor, i.e. a human agent, that brings about the change.This means that a verb with an object that does not have the intrinsic property of bending (e.g.steel) is typically used with AICs, but not with -u(r).Thus, AIC strictly indicates that the external factor is a human agent, whereas -u(r) is compatible with non-human external factors and it is typically used when the event is an intrinsic property of undergoer.As we will discuss in section 6, the language pays particular attention to the types of initiator/causation as an Initiation language in the sense of Ritter & Rosen (2000).That is why it makes use of several different patterns highlighting the use of different types of initiators: (30) a.
Ham metali i-ndrikh-e-n.this metal.nomval-bend-ts-3prs 'This metal is being bent/bending.'It is possible to detect a syntactically active initiator position even in the case of internally caused change of state verbs, such as bloom, rot, decay, illustrated in (31) which are also compatible with -u(r).We again argue that these internally caused predicates also involve an InitP layer in their syntax, which introduces a coindexed covert initiator.It is possible to use these constructions with a purpose clause as in ( 32): (32) Ombri bere-pe oxelu şeni purk-u-n.plum.treechild-pl make.happy for bloom-ts-3prs 'The plum tree is blooming to make the children happy.'Furthermore, as seen in ( 33), it is possible to paraphrase the internally caused predicate with a reflexive construction.The pattern in (33) with the presence of the ergative case and the third person agreement marker -s looks very similar to reflexives which host the clitic i-in their object position as in the example (13a) above.The important point here is that (33) still retains the intransitive reading but implies that the undergoer has more internal control over the caused change in comparison to the construction with nominative subjects in (32).Similar to the case above it is possible to combine this construction with a purpose clause: Ombri-k bere-pe oxelu şeni i-purk-am-s.plum.tree-ergchild-pl make.happy for val-bloom-ts-3prs 'The plum tree is blooming to make the children happy.' To summarize, both externally and internally caused change of state verbs provide evidence for a syntactically active initiator position.Therefore, they do not pattern with intransitive constructions such as passives and anticausatives available in languages like English.The representations in ( 34) and ( 35 Given that we have established that there are no true unaccusative structures in PL, now we will move onto the discussion of unergatives, which we will also argue to have a transitive structure.

TRANSITIVE UNERGATIVES
As seen in the discussion above, both AICs and change of state verbs with -u(r) which seemingly surface with an unaccusative pattern behave transitively in terms of their syntactic architecture in PL.There is no syntactic argument suppression which would yield a true unaccusative pattern consisting of only the ProcP and ResP layers in PL.Both the initiator and the undergoer(/resultee) positions are syntactically active, yielding a transitive syntax, but they do not have to be overtly realized.The initiator position is associated with ergative case, while the undergoer(/resultee) is associated with nominative case.By manipulating case, agreement and TS patterns it is possible to foreground or background either the initiator or the undergoer(/resultee) of the event that the verb denotes in PL.Given the strict transitive pattern, the initiator and the undergoer cannot be dissociated from one another in syntax.Keeping this pattern in mind let us now turn back to the question we have raised at the end of section 3, that is, whether unergatives with the pronominal clitic i-have a transitive syntax or not, which we will answer affirmatively below.

UNERGATIVES WITH THE CLITIC i-
Recall that unergatives which are agentive manner verbs in PL surface with the clitic i-and an overt ergative marked DP as in ( 36): Ali-k i-çaliş-am-s.Ali-erg val-work-ts-3prs 'Ali is working.' Ramchand (2008) takes such unergatives to have InitP and ProcP layers where the undergoer and the initiator positions are fulfilled by the same DP argument as in ( 37): (37)
We take the presence of the clitic i-in such unergatives in PL as the overt manifestation of the pattern which Ramchand argues for in (37).Unergatives with i-are highly similar to the examples with internally caused change of state verbs in (33) above where both the initiator and the undergoer positions are assumed to be strictly coreferential and cannot be realized by different individuals.While unergatives with i-denote internally instigated activities consisting of InitP and ProcP, the examples in (33) refer to internally caused change of state verbs with InitP, ProcP and ResP layers.
As shown in (38), we take the clitic i-to occupy the undergoer position in unergatives: Given the pattern in (37) where the undergoer and the initiator are co-referential, it is not possible to directly transitivize such unergatives in languages like English as shown in (39).

Öztürk
Glossa: a journal of general linguistics DOI: 10.5334/g jgl.828 the subject position.Baker & Bobaljik (2017) highlights the dependent nature of the ergative in Laz.They focus on Laz agentive unergatives which require ergative subjects and argue that the clitic i-might fullfill the formal syntactic transitivity requirement for dependent ergative case (Baker & Bobaljik 2017: 38).Following up on the notion of dependent case discussed in Marantz (1991), Baker &Bobaljik (2017) define dependent ergative case as the following: (43) (Baker & Bobaljik 2017: 4) a.
If NP1 c-commands NP2 and both are contained in the same domain (say, clause), then value the case feature of NP1 as ergative.b.
Based on (43), ergative would surface in the presence of another NP checking case in the structure, that is, when there is an object associated with a case feature is overtly realized we get ergative morphology.Thus, we also take the presence of overt ergative case on the subject as a consequence of the transitive nature of such unergatives in PL.
Given the discussion above, we assume that agentive unergatives in PL naturally involve an InitP and a ProcP layer and are syntactically transitive involving an overt initiator and an undergoer which are co-referential.17

VERBS OF EMISSION
Unergative verbs in PL also include verbs of emission with ergative subjects, which do not take the valency marker i-.Verbs of emission have been argued to have a causal implication, where their sole argument is taken as the causer of the event (Rappaport Hovav & Levin 2000;Potashnik 2012), as indicated by the paraphrase of (44a) as ( 44b): (44) a.
The flower smells.b.
The flower causes the smell.→ The flower is the causer.
As seen in ( 45), PL provides further evidence that the sole argument of such verbs are causers as the subject bears ergative case which is always associated with the semantic role initiator/causer in PL: (45) Purki-k msk'va gont'-u.flower-ergnice smell-3pst 'The flower smelled nice.' As discussed in the previous section, ergative case is licensed when there is another argument in the structure in line with the dependent case analysis of ergative by Baker & Bobaljik (2017).In the case of agentive unergatives the presence of the clitic i-in the lower argument position was what enabled the use of ergative case.In verbs of emission in PL, however, there is no overt morphological evidence for the presence of an undergoer licensing the use of ergative as a dependent case.The question which arises at this point is whether there is any syntactic evidence for a lower object position.This is what we will try to answer below.
As discussed by Öztürk & Taylan (2017), verbs of emission as in (46a) have nominal counterparts as illustrated in (46b).It is possible to paraphrase (46a) as (46c) with the overt light verb 'make/ do', having the nominal form of the verb of emission as the object.However, similar to the case we observed in agentive unergatives, the nominal form cannot be used as the cognate object of the verb of emission, as in (46d), which is again a pattern commonly observed in Turkish as shown in (47).Thus, PL behaves differently from Turkish.As Laz-Turkish bilinguals our informants  Hale & Keyser (2002) proposes the conflation model to account for such unergative verbs.In this model, the nominal complement conflates into a light verb by copying its p-signature, i.e. a phonological feature set, to the light verb whose p-signature is defective.Conflation is not a syntactic movement operation.However, under this derivation it is still possible to find cognate objects in languages such as English.As shown in (46d) above, this is not the case in PL, therefore, the conflation model which requires p-signature copying, is not compatible with PL.We would like to tentatively propose that rather than conflation what PL exhibits for the derivation of such verbs is in parallel to syntactic incorporation (cf.Baker 1988), where the head of an NP moves and incorporates into the light verb, but its trace is still syntactically visible.One piece of rather weak evidence for such a derivation comes from the availability of stranded modifiers in the object position: (48) Ntsa-k ar vorsi gurgul-u.sky-erg a good clap-3pst 'The sky thundered a great (thunder).' Note that such modifier stranding is not very common in PL unlike the case in languages like Hopi and Mohawk, which Baker (1988) discusses, and it does not work with all adjectives or with demonstratives.Furthermore, the adjectival vs. determiner status of ar 'one' is not clear-cut and requires further investigation in PL.Given that there is no morphological distinction between adjectives and adverbs, the use of ar here can be adverbial as well.Therefore, we take the data in (48) as a weak piece of evidence for a derivation based on head-incorporation.
Recall that in the case of agentive unergatives, we have concluded that the lack of cognate objects stems from the fact that at the syntactic level the object position is already full with the reflexive i-.If the incorporation analysis is on the right track for verbs of emission then the trace/copy of the incorporated object might be what is blocking the use of cognate objects.18Furthermore, when there is a syntactically visible object position, we get ergative case being morphologically realized on the initiator/causer in PL as a dependent case.The presence of a full nominative object NP as in regular transitives or the clitic i-is enough to fulfill this requirement for the realization of ergative case.Again if the incorporation analysis is on the right track, then the trace of the incorporated object is what conditions the use of ergative case in verbs of emission. 19Note that we are aware of the fact that we have not really provided conclusive evidence for the presence of a lower argument in the case of verbs of emission in PL.As such, verbs of emission might actually constitute an exception for the transitivity requirement observed in PL.However, to bring this verb class in line with what we have observed in other eventive verb classes in PL, we tentatively take them to be transitives based on the discussion of cognate objects and incorporation, even though these constitute weak pieces of evidence.

RITTER & ROSEN (2000): PL AS AN I-LANGUAGE
As seen in the above discussion, we have argued that unergatives in PL have a transitive syntax, which we take to extend to all types of eventive verbs in PL including unaccusatives.Hence, Neo-Burzio Dependency is very strictly observed in PL.All eventive verbs including unergatives and unaccusatives always form a transitive pattern retaining both the initiator and the undergoer positions available in syntax exhibiting a co-dependency.Hence, there are no truly single argument verbs in the language and also no voice phenomena yielding monovalent verbs, such as passives or anticausatives.In order to have an eventive dynamic structure transitivity requirement should be observed.However, even though there are no true single argument verbs in PL, the language still exhibits patterns analogous to unergative and unaccusative predicates found in languages like English.This is achieved through different case, agreement and TS patterns as extensively discussed above.
The question which raises at this point is why PL exhibits such a strong co-dependency requirement.Is there an underlying reason for this pattern to emerge given the general typological properties of PL are concerned?We believe the typology proposed by Ritter & Rosen (2000) might provide an answer to this.Ritter & Rosen (2000) propose that languages can be split into two: I(nitiation)-languages and D(elimitation)-languages.I-languages base event status to the initial bound, while D-languages focus on the terminal bound of the event.Given this split, I-and D-languages exhibit different clustering properties: 19 As indicated by one of the reviewers, Baker & Bobaljik (2017) show that in some languages like Chukchi, incorporation leads to detransitivization of the predicate which requires absolutive case on the subject blocking the dependent ergative as seen in (ib).This raises the following question: How is the trace of the incorporated object still takes part in dependent ergative assignment in PL?: (i) Chukchi (Baker & Bobaljik 2017: 20) a. ətləg-e mətqəmət (kawkaw-ək) kili-nin father-erg butter.absbread-loc spread.on-3sg>3sg'The father spread the butter (on the bread).' b. tləg-ən (kawkaw-ək) mətqə-rkele-nen father-abs bread-loc butter-spread.on-3sg>3sg'The father spread butter (on the bread).' (Baker and Bobaljik 2017:20) As extensively discussed in Baker, Aranovich & Golluscio (2005) and also in Baker (2014) incorporation does not always lead to detransitivization in all languages.The trace of incorporation can still be visible in terms of phi and case features in some languages, which they take to be subject to parameterization.For example, as seen in (ii), in Southern Tiwa the verb can still agree with the incorporated object and as seen in (iii) the object which undergoes pseudo-incorporation in Hungarian, which Baker (2014) analyzes as a specific kind of headincorporation is still visible for case.Given these, we take PL to be a language of the type like Southern Tiwa or Hungarian, but not like Chukchi, where the trace of incorporation is still visible for dependent case retaining the transitivity of the predicate: Southern Tiwa (Baker, Aranovich & Golluscio 2005: 141) a. Ti-seuan-mu-ban.

Table 1
Agreement Suffixes in PL.

Table 2
MorphosyntacticProperties of Verb Classes in PL.
As we will discuss shortly, these verbs are different from what is known as anticausatives which would only consist of ProcP and ResP layers lacking InitP in languages like English.We will argue that even though there is only one overt DP argument marked with nominative case in the structure, these constructions do not qualify as anticausatives, but they are underlyingly transitive and require an InitP layer which hosts a null initiator.Such unaccusative verbs can be internally or externally caused.If they are internally caused, the nominative DP would be simultanously associated with the roles of Resultee, Undergoer and Initiator in the representation in (26).If they are externally caused, while the overt DP fulfills the roles of Undergoer and Resultee, the InitP layer would host a covert initiator.
) depicts the structures we propose for externally and internally caused change of state verbs, respectively:15