Affective constructions in Tsezic languages

Comrie, Bernard, Diana Forker & Zaira Khalilova. 2018. Affective constructions in Tsezic languages. In Jóhanna Barðdal, Stephen Mark Carey, Thórhallur Eythórsson & Na'ama Pat-El (eds.) Non-canonically case-marked subjects within and across languages and language families: The Reykjavík-Eyjafjallajökull papers, 59–86. Amsterdam: Benjamins. Affective constructions in Tsezic languages Bernard Comrie, Diana Forker, Zaira Khalilova This paper addresses affective (“experiencer”) constructions in the Tsezic languages (NakhDaghestanian), which represent the most frequent type of non-canonical subject constructions in these languages. They differ from transitive constructions in a number of ways that go far beyond case marking and affect various domains of grammar (e.g. inflectional morphology, complex clause structures, reflexive and reciprocal binding, etc.). In this paper, we explore morphological, syntactic and semantic features of Tsezic affective constructions from a typological perspective. We investigate variation, stability and change between the constructions in the various Tsezic languages and try to give explanations for the observed patterns of variation. 1. Introduction This paper addresses affective constructions in the Tsezic languages, a group of five closely related languages within the Nakh-Daghestanian language family. By ‘affective constructions’ we refer to constructions with two-place predicates that exhibit the following morphosyntactic pattern: the predicates are bivalent with one argument marked by the dative/lative/IN-essive, depending on the language, and a second argument marked by the absolutive. Agreeing affective predicates agree in gender and number with the absolutive argument, though not all predicates exhibit agreement. To the predicates occurring in Tsezic affective constructions belong, among others, verbs with the English translations ‘see’, ‘hear’, ‘know’, ‘like’, and ‘remember’, but also ‘get’ or ‘lose’ and a few complement-taking predicates such as ‘can’ or ‘be allowed’. The two arguments of affective predicates largely fulfill the semantic roles of experiencer and stimulus, but there are also exceptions. Experiencers can be characterized as sentient arguments which undergo a sensory, cognitive, or emotional experience because they are mentally or in some other way affected by a stimulus (cf. Dowty 1991, Dahl to appear, Verhoeven 2007: 54–61). They are the animate though non-volitional locus of a sensory impression or psychological or emotional state or process and thus mainly occur with psychological and perception predicates. The stimulus, sometimes called ‘theme’, is the content of an experience or whatever causes a psychological response in the experiencer, be it emotionally positive, negative, difficult, easy or neutral. The stimulus is often not sentient, but can be an idea or a mental representation that is entertained or perceived, and it does not undergo a change of state. We call these constructions ‘affective constructions’ and not ‘experiencer constructions’ because some of the verbs can hardly be said to require an experiencer argument based on the common understanding of this semantic role. And by far not all verbs with the semantic 1 roles of experiencer and stimulus fall into the class of affective verbs analyzed in this paper (see Section 8). In the present study we take a semasiological perspective, i.e. we start from the linguistic structure by identifying affective constructions based on their morphosyntactic form (cf. Mosel 2006). We then proceed to the meaning and embed them into the wider context of experiential constructions and non-canonical subject constructions, thereby not relying on a single theoretical framework, but drawing on various approaches. However, our general approach is a functional-typological one. The paper attempts a detailed description of Tsezic affective constructions covering their morphosyntactic and semantic properties and discussing their syntactic status in contrast to intransitive and transitive constructions. The data used comes either from published sources or has been gathered by ourselves during fieldwork. For overviews on affective constructions in Nakh-Daghestanian including some information on Tsezic see Comrie & van den Berg (2006) and Ganenkov (2006). Before we turn to the main topic of the paper a brief introduction to the Tsezic languages is in order. The Tsezic languages form a subbranch of the Nakh-Daghestanian language family. They can be divided into West Tsezic (Tsez, Khwarshi and Hinuq) and East Tsezic (Bezhta, Hunzib). Tsezic languages generally have a very rich case system comprising minimally absolutive, ergative, genitive, and a large number of spatial cases. Many vowelinitial verbs in these languages show gender/number agreement with their argument in the absolutive case (e.g. S, P). The basic predicate classes are (i) intransitive predicates with one argument in the absolutive, (ii) extended intransitive predicates with one argument in the absolutive and another one marked with a spatial case, (iii) affective predicates, (iv) transitive predicates with one argument in the ergative and another argument in the absolutive, and (v) extended transitive predicates with one ergative, one absolutive, and a third argument mostly bearing a spatial case. The paper is structured in the following way: in Sections 2 and 3 we describe the formal properties of affective constructions and investigate the restrictions that affective predicates are subject to within the Tsezic verbal inflection systems. In Section 4 we turn to the peculiarities of affective constructions in simple clauses followed by a discussion of their behavior in complex sentences in Section 5. We then analyze the status of their arguments against the background of a general discussion of non-canonical subjects. We do this by applying a number of subjecthood tests and comparing the results with other argument types, first and foremost with agents. In Section 8 we briefly discuss the position of Tsezic affective constructions from the broader perspective of experiential constructions. Section 9 contains the conclusion. 2. Affective predicates: Formal properties and additional functions Affective predicates are two-place predicates, taking one argument in the dative (Hinuq, suffix -z), lative (Tsez -r, Bezhta -l, Khwarshi -l) or the IN-essive (Hunzib, the suffix is a copy of the preceding vowel) and a second argument in the absolutive case or alternatively a clausal complement.1 The dative/lative/IN-essive argument often has the semantic role of an experiencer and the absolutive argument functions as the stimulus. But since there are a few affective predicates for which this is not the case we refrain from calling these arguments The Tsez verb -esu- ‘find, come across, get’ can also be used as a one-place predicate in the sense ‘show up, turn up’ (Comrie 2004: 117), but all other affective verbs discussed in this paper are two-place predicates. 1 2 ‘experiencers’ and ‘stimuli’. Instead, we will use the terms ‘more prominent argument’ and ‘less prominent argument’, because the dative/lative/IN-essive argument cannot be lower on an animacy scale than the absolutive argument: the dative/lative/IN-essive argument must refer to an animate referent, whereas the absolutive argument can also refer to an inanimate referent or to a situation. Thus, our definition of ‘Tsezic affective predicates’ is essentially morphosyntactic in nature. However, there is a correlation with semantics because the majority of affective predicates takes arguments whose semantic roles are experiencer and stimulus.2 Affective verbs having agreement prefixes agree with the less prominent argument (1a, e). In the case of complement-taking affective predicates they exhibit local agreement or long-distance agreement with the absolutive argument of the embedded clause (1b). Here are some examples: (1) a. Tsez (Abdulaev & Abdullaev 2010: 244) ža ɣˤanabi […] žedur=no this woman(II) 3SG.M.LAT=and ‘He did not know this woman.’ y-iy-nanu II-know-UWPST.NEG b. Hinuq (Forker 2013: 606) debez kiy kʼilikʼ-a behezi y-iqqo gom 2SG.DAT blueberry(IV) wash-INF be.allowed IV-happen.PRS be.NEG ‘You are not allowed to wash blueberries.’ (because the taste goes away) c. Bezhta hollool 3PL b-öčö-yö-ɬi HPL-leave-PTCP-NMLZ äƛ-ä-lä village-OBL-GEN2 ädäm-lää-l person-PL.OBL-LAT hexzu šoʔ-na long.ago forget-UWPST ‘Long ago the villagers forgot their leaving.’ d. Khwarshi (Khalilova 2009: 324) tuq-un c’odoraw-il ʕadalaw-is hear-UWPST clever-LAT fool-GEN1 ‘Clever heard Fool talking to the bear.’ ze-qo bear-CONT iss-u tell-PTCP xabar talk e. Hunzib (van den Berg 1995: 123) ož-di-i kid y-ãc’ə-r boy-OBL-IN girl(II) II-see-PRET ‘The boy saw the girl.’ 2 We do not further extend our notion of affective predicates to oblique subject constructions or all constructions that employ non-canonical case marking of arguments because this would not yield a morphosyntactically coherent class of predicates and the semantic correlation with the roles of experiencers and stimuli would be completely lost. 3 Only Hinuq has a dedicated dative case that does not serve any spatial functions but marks the more prominent argument in affective constructions, and also beneficiaries, recipients, etc. The other Tsezic languages make use of their rich spatial case systems and encode the more prominent argument with a suffix expressing location in(side) (Hunzib INessive) or with a suffix expressing the movement to a goal (lative in Tsez, Bezhta and Khwarshi). Dabrowska (1997: 52–54) sketches a possible development of the Polish dative that can easily be transferred to the Tsezic languages and explain the use of spatial cases in affective constructions. The dative in Polish most probably goes back to an allative in Proto-Slavic with purely spatial meaning, namely the expression of motion towards a goal. This goal can be inanimate or animate, specifically human. Animate goals differ from inanimate goals because they are normally affected when something reaches them and enters their sphere. First, the affectedness arose as an implicature since, in principle, unaffected animate goals are also possible. However, in the course of time the implicature became conventionalized as a kind of affected recipient. After conventionalization this aspect of the meaning became increasingly more important because humans are normally more interested in how other humans are affected than in the final destination of objects and their affectedness. In the last stage the spatial use has basically disappeared. Dabrowska also notes that the various stages rather than occurring neatly one after the other are simultaneously attested in the language. We assume that this scenario is plausible for the development of the dative functions of the Tsez, Bezhta, and Khwarshi lative suffixes. In its spatial meaning the lative expresses motion to a goal (cf. Khalilova 2009: 86–88). The non-spatial functions that are now attested with this case (marking of experiencers, recipients, beneficiaries, etc.) arose from the spatial meaning in a similar fashion as happened in Polish, though the spatial meaning is also preserved. For Hunzib the scenario must be modified because in this language it is the IN-essive that has taken over the dative functions. Originally the Proto-Tsezic IN-essive denoted only the location of an object inside a container (Cysouw & Forker 2009). The location is normally inanimate (e.g. box, drawer, corner). However, body parts can also function as locations and then be marked with the IN-essive, e.g. the belly. In a later stage Hunzib has not only extended its spatial use of the IN-essive from container-like grounds to various inanimate grounds (van den Berg 1995:45–46), but also introduced a metaphorical use starting from the body-part containers. In the metaphorical use the ground is animate/human and conceived as the container not just of physical objects, but of mental objects and states such as perceptions, and feelings. Again, this usage of the IN-essive became more and more important and extended to other, less stative dative functions. The Tsezic situation is by far not unique among the languages of the world. Similar cases are attested for other Nakh-Daghestanian languages (e.g. Udi also has an IN-essive/dative syncretism, Schulze 2009), and are especially common among the South Asian languages. In Malayalam (Dravidian) dative-marked arguments such as experiencers are types of goals or possessors (Mohanan & Mohanan 1990). In Marathi (Indo-Aryan) experiencers are animate locations and the language shows a case syncretism of locative, possessive and dative (Pandharipande 1990). Based on these data experiencers have often been characterized as mental or abstract locations or goals of some sort (cf. Jackendoff 1990: 262, Landau 2010). 4 Though Hinuq is the only Tsezic language with a dedicated dative case, there is some variability with regard to the case marking of the affective predicates: the verbs ‘get, find’ and ‘can’ allow marking with a spatial case, the AT-essive, in addition to dative marking. There does not seem to be any semantic difference between affective verbs taking a dative and those taking an AT-essive. This variation can be explained with the well-known connection between experiencers and possessors. Possessors are animate locations, and in many language possessors and experiencers are marked alike (cf. Verma & Mohanan 1990, Næss 2007: 199). In Hinuq, the AT-essive marks, among other things, temporal possession. So its occasional employment with experiencers is most probably an extension of the possessor-marking function. Two affective verbs in Khwarshi have transitive cognates that differ in meaning from the affective verbs and assign the semantic roles of agent and patient to their arguments: affective loq- ‘to get’ vs. transitive loq- ‘to take’; transitive tuq- ‘listen’ vs. affective tuq‘hear’ (Khalilova 2009: 305). From a word-formation perspective affective verbs can be simple verbs or compound verbs consisting of loans or a native noun in an oblique case plus a native light verb (e.g. ‘understand’, ‘need’, ‘remember’). Hunzib also has derived affective verbs formed by means of adding the suffix -a.q’e to a transitive verb, e.g. -uwo- ‘do, make’ > -uw-a.q’e- ‘like to make’ (2a). When the same suffix is added to intransitive predicates, they only acquire the desiderative meaning without changing their valence pattern (2b). (2) Hunzib (van den Berg 1995: 112) a. hũl diʔi keč’ iƛ-a.q’-čo yesterday 1SG.IN song sing-DES-PRS ‘Yesterday I liked to sing a song.’ zuq’u-r be-PRET b. xišaƛ’ də hĩ̵də.no.ž k’ot’ Ø-əhdaː-q’e morning 1SG always well I-work-DES.PRS ‘In the morning I (masc.) feel like working.’ Tsezic languages show some variation with regard to the class of verbs occurring in affective constructions. Every Tsezic language has about a dozen affective verbs that occur in simple clauses and occasionally as complement-taking predicates in complex sentences (Table 1). Nevertheless these predicates form a coherent class within the individual languages, and the classes can be compared across the five languages based on common morphosyntactic properties. In addition, many of the predicates are obviously cognates. Such a comparison can even be more justified by the fact that similar classes of verbs can be identified across a large number of Nakh-Daghestanian languages (cf. Ganenkov 2006). Table 1: Affective predicates in Tsezic Tsez Hinuq ‘dislike, be -aci-ace-, surzi bored, hate’ iq‘want, like, -eti-etilove’ ‘see’ -ikʷad-ike- Khwarshi -ac-, ƛuralqoč-, goq- Bezhta rixzi -aq-, surzi aq-atʼ- Hunzib -acə- -ak- -egaː- -ãc’(ə)- 5 -at’(ə)- (dialectal: ukad-) teq- toq- tuq- tuq- ‘know’ ‘find, come across, get’ -iy-esu- -eq’i-aši- -iq’-us-, loq- -iqʼezoɣ- ‘find’ ‘forget’ ‘receive, get’ ‘get to know, notice, feel’ šuƛ’iiqič’aɬ- šuƛ’e- šuƛ’- c’aɬ- č’al- šöʔ-ĩq- ‘get’ č’aɬ-, čoq- ‘understand’ bič’zi -oq- bič’i -iq- bič’id-, bulh- ‘need’ ħažat -oq- q’ʷaraʡezi -iqrok’ƛ’o(r) -aq’e- / goɬ ‘hear, listen’ ‘remember’ ‘meet’ ‘lose’ č’al- ‘notice’; čoqe- ‘get to know’ häžät -aqlok’oƛ’o -eč- kezi -iq- yak’ ƛ’a õq’o- ‘recall’; yak’ ƛ’a -ec‘keep in mind’ kezi -aq-əʔə- ‘be enough’ -u- ‘prick’ ‘seem, think’ ‘can’ ‘be allowed’ bidži -aq-, adaxe- tuq- ‘hear’; nɨd- ‘hear’ -iq’ezəɣ- ‘show up, be found’; -ɨ̃qə- ‘find, get’ šoʔo- xäƛemüq- koƛ’i- Complement-taking predicates qebaːlokolgic’koƛ’ebehezi -iqbehid- tɨq’əgič’- Most of the predicates in Table 1 are psych verbs expressing sensory perceptions, cognitive processes or emotional states. But as already mentioned, not all bivalent predicates that assign dative/lative/IN-essive and absolutive to their arguments can be said to assign the semantic role of an experiencer with its typical properties (see the characterization in Section 1). For instance, ‘lose’, ‘can’, or ‘be allowed’ are not psych verbs and the semantic roles associated with these verbs are not perceivers, emoters, or cognizers. Nevertheless, we will treat them as affective verbs because they share most if not all morphosyntactic properties with typical affective verbs such as ‘like’ or ‘see’. On the other hand, we exclude verbs like ‘beat’ or ‘hit’. These verbs denote situations that imply an 6 experience of bodily affectedness, but they follow a different valence pattern.3 Similarly, we exclude all other predicates and constructions that denote experiential situations. See Section 8 for a short discussion of the relation between affective predicates and other verbs denoting experiential situations. When comparing the verbs listed in Table 1 to the structure of the domain of experience developed by Verhoeven (2007: 41–51), which comprises (i) bodily sensation, (ii) emotion, (iii) cognition, (iv) volition, and (v) perception, we notice that none of the Tsezic verbs occurring in affective constructions denote bodily sensations, but all other subdomains are covered. The lack of predicates denoting bodily sensation in the class of affective predicates in Tsezic can be explained by the fact that sensations are usually not directed towards a specific stimulus participant and therefore such predicates are more often than not monovalent. Before concluding this section we briefly want to point out an additional function of the affective construction in Tsezic, namely the expression of mirativity. In Tsez, Hinuq, and Khwarshi the verbs with the meaning ‘find, come across, get’ can be used to encode mirative meanings, a usage frequently attested in Nakh-Daghestanian languages (cf. Khalilova 2009: 231–237 for Khwarshi examples; Kibrik 1994 on Archi). For example, in (3) a man went to a city that to his surprise was completely empty of people. He visited the market, but nobody was there. Verhoeven (2007: 56) also mentions such an extended deployment of affective constructions within evidentiality or mediativity systems. (3) Tsez (Abdulaev & Abdullaev 2010: 60) yoɬ yāɬ-zā=tow q’ay-bi=n be be-LOC.CVB=PRT good-PL=and xeci-n leave-CVB bazargan-bi=n merchant-PL=and b-ox-äsi b-esu-n el-āy HPL-run.away-PTCP HPL-find-UWPST there-IN.ABL ‘Surprisingly the merchants had left their goods where they were and had run away.’ 3. Verbal inflection and derivation Affective verbs show some restrictions and in general a fairly heterogeneous behavior with regard to a number of verbal inflectional and derivational processes. 3.1. Imperative/Prohibitive In Tsez, Hinuq, and Hunzib (Comrie 2001, Forker 2013: 229, van den Berg 1995: 88) imperatives and prohibitives cannot be formed from affective verbs. In order to form imperatives of these verbs the verbs must be causativized such that they become standard transitive verbs, and then imperative formation is possible, but causativization can lead to a semantic change (see Section 3.3). In contrast, in Khwarshi and Bezhta affective verbs normally form imperatives and prohibitives (Comrie 2001, Khalilova 2009: 242–247), with the exception of the verbs ‘find’ and ‘see’ in Khwarshi which first must undergo causativization. As can be seen in (4c), the more prominent argument can be overtly expressed, just as other addressees of imperatives. 3 The referent of an affected participant in a beating situation is marked with the dative/lative/IN-essive in the Tsezic languages because it fulfills the semantic role of an experiencer, but the verb beat is trivalent and not bivalent, and does not behave like the affective predicates treated in this paper. 7 (4) Hinuq a. hado uži *Ø-eti-yom this boy(I) I-love-PROH ‘Do not love this boy! / / Ø-eti-r-om! I-love-CAUS-PROH b. Hunzib (van den Berg 1995: 88) Ø-ãc’ə-k’-(o)! I-see-CAUS-IMP ‘See (him)!’ c. Khwarshi (Khalilova 2009: 247) žu himon qoq-o dubul! that thing love-IMP 2SG.LAT ‘You like this thing!’ d. Bezhta dul wahli bidži<y>aq! 2SG.LAT this understand<IV>.IMP ‘You understand this!’ 3.2. Intentional future Tsez, Hunzib and Hinuq have a so-called ‘intentional future’ (or ‘definite future’). This verb form is employed when referring to events for which it is intended that they happen in the future, e.g. plans and goals of people. It is normally only used with first persons. In Tsez and Hinuq, this verb form cannot occur with plain affective verbs (5a, b) which must first be causativized as illustrated in (5c) (Comrie 2001, Forker 2013: 202). In Hunzib, the situation seems to be different; the intentional future is found with affective verbs in self-addressing questions (5d). (5) Hinuq a.*diž me šuƛ’-an 1SG.DAT 2SG forget-INTFUT (Intended meaning: I will forget you.) b. Tsez (Comrie 2001: 69) *dä-r meši b-es-an 1SG.OBL-LAT calf(III) III-find-FUTDEF (Intended meaning: I will find the calf.) c. Tsez (Comrie 2001: 69) di meši b-esu-r-an 1SG.ERG calf(III) III-find-CAUS-FUTDEF ‘I will find the calf.’ d. Hunzib (van den Berg 1995: 86) 8 že də sα-y aqe y-ẽƛ’-à, now 1SG who.OBL-IN wife(II) II-go-INTFUT ‘Now whom shall I marry, whom shall I find?’ diʔi 1SG.IN suk’u who Ø-ɨ ̃q-á? I-find-INTFUT 3.3. Causative verbs Tsezic languages have very productive causativization operations, allowing simple and often even double causativization of verbs with various valence frames. Causativization necessarily involves verbal derivation or auxiliary change. With intransitive and transitive predicates causativization usually adds one argument to the valence frame. Causativized transitive verbs are ditransitive with an ergative agent, an absolutive patient and a causee marked with the AT-essive case (Tsez, Hinuq, Hunzib, Bezhta) or the instrumental (Bezhta). With the exception of Hinuq qebaː- ‘seem’, which cannot be causativized at all, Tsezic affective verbs generally permit the formation of causative verbs. But in contrast to other two-place verbs (e.g. canonical transitive verbs) causativization does not necessarily result in a three-place verb. Thus, it is not always possible to detect a clear difference in the semantics between a non-derived affective verb and a causativized affective verb. A few affective verbs simply change the case marking from dative/lative for the more prominent argument to ergative, without changing the meaning or adding another argument to the valence frame, e.g. Hinuq šuƛe- > šuƛe-r- ‘forget’, Tsez -esu- > -esur- ‘find’, Bezhta šöʔ‘forget’, zoɣ- ‘find’. Thus, compare (6a) exemplifying the verb šuƛ’- ‘forget’ with its causativized counterpart šuƛ’ir- in (6b); there is no noticeable semantic difference between the two verbs. (6) Tsez (Khalilov 1999: 294) a. šow deber t’o šuƛ’-asi? what 2SG.LAT here forget-WPST ‘What did you forget here?’ b. Tsez di mi=n 1SG.ERG 2SG=and ‘I will forget also you.’ šuƛ’i-r-an forget-CAUS-FUTDEF Many affective verbs become canonical transitive verbs. They undergo a slight change in the meaning, usually by becoming more agentive, i.e. ‘want, like’ > ‘love, cuddle’, ‘dislike, be bored’ > ‘take offence’, ‘receive’ > ‘keep’, ‘hear’ > ‘listen’. For example, the Bezhta verb -at’means ‘like, love, want’ and takes an experiencer as the more prominent argument (7c). If it is causativized the experiencer changes into an agent and the situation denoted by the verb is more active describing a volitional action (7d). (7) Tsez (Khalilov 1999: 40) a. däːr nesis ʕamal 1SG.LAT 3SG.GEN1 character(III) ‘I dislike / hate his character.’ b-aci-x III-dislike-PRS b. Tsez (Khalilov 1999: 40) 9 nesä r-aci-r-si 3SG.ERG IV-dislike-CAUS-WPST ‘He was offended by what I said.’ di 1SG.ERG äƛ-iru speak-PTCP šebin thing c. Bezhta öždi-l kid y-at’-ca boy.OBL-LAT girl(II) II-love-PRS ‘The boy loves the girl.’ d. Bezhta öždi kid y-at’-il-ca boy.ERG girl(II) II-love-CAUS-PRS ‘The boy cuddles with the girl.’ Some verbs become three-place extended transitive predicates by adding an ergative agent, e.g. Hinuq -ike- ‘see’ > -ike-r- ‘show’, Tsez -iy- ‘know’ > -iyr- ‘inform, teach’ (ditransitive), ‘learn’ (transitive); Bezhta tuq- ‘hear’ > tuq-il- ‘make hear’, č’aɬ- ‘get to know’ > č’aɬil- ‘inform’. (8) Bezhta a. dil mi Ø-iƛe-yo-ɬi 1SG.LAT 2SG I-call-PTCP-NMLZ ‘I heard you calling.’ tuq-iyo hear-PST b. Bezhta kibba dil keč’ tuq-il-ca girl.ERG 1SG.LAT song hear-CAUS-PRS ‘The girl made me hear the song.’ A few causativized affective verbs display variability with respect to the case marking of the causee. Example (9a) shows the Khwarshi verb -ak- ‘see’. When the verb is causativized the causee may be marked with the case suffix normally expressing causees with transitive verbs, e.g. AT-essive in Khwarshi (9b). Or it may keep its case marking and therefore bear the lative suffix (9c). (9) Khwarshi (2009: 79) a. isu-l b-ak-un ɬe-ɬ 3SG.OBL-LAT III-see-UWPST water.OBL-INTER ‘He saw an apple in the water.’ giɬ down ẽš apple(III) c. Khwarshi (2009: 344) obu-t’-i kandɨqo surat b-ak-xʷ-i father-OBL-ERG daughter.AT picture(III) III-see-CAUS-WPST ‘The father showed the picture to (his) daughter.’ b. Khwarshi (2009: 284) 10 dubul Allahise b-ak-xʷ-i 2SG.LAT Allah.ERG III-see-CAUS-WPST ‘Allah showed me the medicine.’ daru medicine(III) 3.4. Potential verbs Tsezic languages have a potential construction that is used with potential agents that do not have all agentive properties. The same construction can also have an involuntary semantics for actions performed non-volitionally (Forker 2013: 499–502, Comrie 2001, Khalilova 2009: 305). These constructions normally do not permit affective verbs in those Tsezic languages for which the relevant information is available (Tsez, Hinuq, Khwarshi, Bezhta) as can be seen in the ungrammatical example (10a). There is one exception to this rule: Bezhta has two potential constructions, and one of these constructions is available with the verb yak’ƛ’a -ec- ‘remember’ > ak’ƛ’a -eče-yɬ- ‘be able to remember’ (10b). (10) Hinuq a.*diqo hayɬos roži-be šuƛ’e-ɬ-me 1SG.AT 3SG.GEN word-PL forget-POT-NEG (Intended meaning: I cannot forget his words.) b. Bezhta öždi-qa keč’ yak’-ƛ’a boy.OBL-AT song(III) heart.OBL-SPR ‘The boy was able to remember the song.’ b-eče-yɬ-iyo III-stay-POT-PST 3.5. Antipassive Another valence-changing derivation attested in Tsezic is the antipassive. This construction is not as productive as causativization, but nevertheless attested with a wide range of verbs. The antipassive normally has an iterative or habitual reading in these languages, thus it is probably due to this semantic impact that not all verbs can undergo antipassivization. Antipassive suffixes can be added to intransitive verbs with no impact on their valence frames. With those transitive verbs that allow for antipassive formation this operation leads to detransitivization, i.e. the agent becomes the S argument of an intransitive verb and is therefore marked with the absolutive. The patient is either completely deleted (e.g. Hinuq, Hunzib), or occurs in an oblique case (e.g. in the instrumental in Bezhta). In Bezhta it is impossible to form antipassive from affective verbs. The same seems to be true for Tsez, because the few examples provided by the dictionary (Khalilov 1999) were all rejected by our informant. In Hinuq, most affective verbs do not form antipassives (e.g. ‘understand’, ‘get to know’), but at least from toq- ‘hear’ and šuƛ’e- ‘forget’ it is possible to build antipassives: the verb toq-liː- ‘rummage’ is intransitive, and šuƛ’e-doː- ‘forget repeatedly’ remains an affective verb with the dative/absolutive valence frame. For Hunzib it seems that affective verbs do not form antipassives, at least the dictionary does not provide any examples (Isakov & Khalilov 2001), and Khwarshi does not have antipassives. 4. Affective predicates in simple clauses 4.1. Biabsolutive construction 11 Different types of biabsolutive constructions have been attested for Tsez, Hinuq, Khwarshi and Bezhta. In biabsolutive constructions, which are normally restricted to imperfective verb forms, the agent is promoted and the patient demoted. This is realized by changing the case marking of the agent from ergative to absolutive and the auxiliary agrees now with the promoted agent if it can agree at all (11a). Biabsolutive constructions do not admit affective verbs as the ungrammatical sentence in (11b) illustrates (cf. Forker 2012, Khalilova 2009: 309). (11) Tsez a. ʕali meši b-išer-xo Ali calf(III) III-feed-ICVB ‘Ali was feeding the calf.’ zowsi be.WPST b. Tsez * kid ʕazab b-ukay-xo zowsi girl torture(III) III-see-ICVB be.WPST (Intended meaning: The girl saw torture.) 4.2. Reflexive and reciprocal constructions The Tsezic languages, like many other Daghestanian languages, have reflexive and reciprocal constructions in which the less prominent argument (e.g. the argument in the absolutive case) functions as the controlling NP and the more prominent argument (e.g. the argument in the ergative, dative or lative case) functions as the reflexive or reciprocal pronoun (cf. Toldova 1999, Ljutikova 2001, Kibrik 2003). In general, affective verbs more often admit such typologically unusual reflexive and reciprocal constructions than canonical transitive verbs or extended intransitive and transitive verbs. An overview for the Tsezic languages is given in Table 2. Table 2: Reflexive and reciprocal constructions with affective predicates Reflexive Reciprocal Antecedent Reflexive Antecedent Reciprocal Hinuq Tsez Khwarshi Bezhta Hunzib DAT ABS DAT ABS ABS DAT ABS DAT LAT ABS (LAT ABS) ABS LAT ABS LAT LAT ABS (LAT ABS) ABS LAT ABS LAT LAT ABS ABS LAT IN-Essive ABS ABS IN-Essive # ABS LAT no information no information As can be seen from Table 2, all Tsezic languages allow a ‘more standard’ reflexive construction and a ‘less standard’ reflexive construction. In the more standard construction the controlling noun phrase occurs as the more prominent argument since it is marked with the dative, lative, or in Hunzib with the IN-essive case and the reflexive appears as less 12 prominent argument bearing absolutive case (12a). In the less standard construction the case marking of reflexive and antecedent is reversed (12b). (12) Hunzib (Kibrik 2003: 614) a. oždiː žu Ø-ãc’ə-r boy.IN REFL I-see-PRET ‘The boy saw himself.’ b. Hunzib (Kibrik 2003: 614) ože žiniː.žu Ø-ãc’ə-r boy(I) REFL.IN I-see-PRET ‘The boy saw himself.’ Reciprocal constructions show a more diverse picture. In Hinuq, again the more standard construction (more prominent argument in the dative, reciprocal in the absolutive) and the less standard construction (more prominent argument in the absolutive, reciprocal in the dative) are both equally available. In Tsez and Khwarshi, the second construction (more prominent argument in the absolutive, reciprocal in the lative) is more common whereas the other variant (with the reciprocal in the absolutive) is only marginally acceptable (Khalilova 2009: 453 on Khwarshi). In Tsez, as in other Daghestanian languages, not only the reciprocal pronoun, consisting of the reduplicated numeral sis ‘one’, but also plural reflexive pronouns can be used in reciprocal constructions. If the reciprocal pronoun is employed, it can have two different forms: lative (appropriate to its role) plus absolutive (reflecting the case of the controller) as in (13a), or a fossilized ergative plus lative (appropriate to its role) as in (13b). There does not seem to be any semantic difference connected to the two different forms. The variant with the controller in the lative is ungrammatical with either reciprocal pronoun (13c), hence it is given in brackets in Table 2. (13) Tsez a. Šamil=no Madina=n sider_sis Shamil=and Madina=and REC.LAT_REC.ABS ‘Shamil and Madina loved each other.’ b. Tsez Šamil=no Madina=n sida_sider Shamil=and Madina=and REC.ERG_REC.LAT ‘Shamil and Madina loved each other.’ c. Tsez *Šamil-er=no Shamil-LAT=and Madina-r=no Madina-LAT=and b-eti-x zowsi HPL-love-ICVB be.WPST ‘Shamil and Madina loved each other.’ 13 b-eti-x HPL-love-ICVB b-eti-x HPL-love-ICVB sider_sis REC.LAT_REC.ABS zowsi be.WPST zowsi be.WPST / sida_sis / REC.ERG_REC.ABS If, however, the plural reflexive pronoun is used, then both constructions are possible, but the variant with the reciprocal bearing the lative case suffix is nevertheless preferred and admits, for instance, a change in the word order with the pronoun preceding the controller. In Bezhta the variant with the controller in the lative is not available at all for reciprocal constructions (14a). This means that the antecedent in a reciprocal construction with an affective predicate must always bear the absolutive case (14b). (14) Bezhta a.*öž-di-l=na kibba-l=na sidi.hosso boy-OBL-LAT=and girl.obl-LAT=and REC.ERG (Intended meaning: The boy and the girl like each other.) b. Bezhta (Comrie et al. 2011: 45) Pat’imat=na Rasul=na sidil.hosso Patimat=and Rasul=and REC.LAT ‘Patimat and Rasul like each other.’ b-aːc-ca HPL-like-PRS b-aːc-ca HPL-like-PRS 5. Affective predicates in complex clauses In relative clauses and adverbial clauses there is no observable difference between affective predicates and predicates from other valence classes. For example, dative/lative/IN-essive arguments (15a) as well as absolutive arguments (15b) can be the target of relativization: (15) Hunzib (van den Berg 1995: 132) a. [(žini-i)i wə m-ac’ə-čo-s] REFL.OBL-IN dog(IV) IV-see-PRS-GEN1 ‘the girl who saw the dog’ b. Hunzib (van den Berg 1995: 131) [αbu-u _i y-ãc’ə-čo-s] father-IN ABS II-see-PRS-GEN1 ‘the girl whom father sees’ kidi girl kidi girl(II) Similarly, in coordinate constructions and adverbial clauses both types of arguments can be omitted when they are coreferential with some other overtly expressed argument. Thus, in the adverbial clause in (16a) the lative argument of the verb -ak- ‘see’ has been omitted because it is coreferential with the S argument uže ‘boy’ in the main clause. Likewise, in (16b) the absolutive argument of ‘see’ is not overtly expressed due to coreference with the S argument in the main clause. (16) Khwarshi (Kibrik 2003: 645-646) a. užei Ø-ot’q’-ɨ [ _i boy(I) I-came-WPST LAT ‘The boy saw the girl and came.’ kad=ɨn girl(II)=and b. Khwarshi (Kibrik 2003: 645-646) 14 y-ak-un] II-see-CVB kadi y-ot’q’-ɨ [užal=in girl(II) II-came-WPST boy.LAT=and ‘The boy saw the girl and she came.’ _i ABS y-ak-un] II-see-CVB Such deletion of coreferential arguments is the norm. Whether it is possible or not does not depend on the valence frame of the predicate but on the place of the omitted argument and the order of adverbial clause and main clause. In short, the only restriction is the ‘precede-and-command’ constraint, i.e. a zero anaphor cannot precede the coreferential noun phrase when the zero is part of the preceding main clause (Lasnik 1976, Reinhart 1981). This constraint holds not only for affective predicates, but also for complex clauses with other predicates. Most complement clauses do not make any difference between embedded affective predicates and other embedded predicates. Experiencers are the target of equi-deletion in complement clauses, in a parallel fashion to single arguments of intransitive predicates or agents. For instance, in (17a) the S argument hago ‘he’ in the matrix clause is coreferential with the more prominent argument of the complement clause. Therefore, the latter is deleted. In the Bezhta example in (17b) the lative-bearing pronoun can belong to the embedded affective predicate or to the matrix predicate niƛ- ‘give’ that assigns lative to recipient arguments. Therefore this example is presented without brackets indicating clause boundaries. (17) Hinuq a. hago Ø-uƛ-o [_ baru 3SG.M I-fear-PRS DAT wife ‘He fears to forget his wife.’ šuƛ-a] forget-INF b. Bezhta hogcol is Ø-ega-al ixtiyar niƛ-na 3SG.LAT brother(I) I-see-INF permission give-PST ‘He was allowed to see the brother.’ (Lit. He was given the permission to see the brother.) Nevertheless, a few types of complement clauses exhibit peculiarities with embedded affective predicates that are not attested with transitive predicates. First, in Hinuq, affective verbs are generally not permitted in complement clauses with the modal verb ‘can, be able’ (18). This constraint seems to be more semantic than syntactic in nature. That is, ‘can’ implies control over the relevant action, which is in contrast to the meaning of affective verbs.4 (18) Hinuq (Forker 2013: 642) diqo kʼwezi r-iq-me 1SG.AT be.able V-happen-NEG [debe 2SG.GEN1 roži-be word-PL *šuƛʼ-a forget-INF šuƛʼe-r-a] 4 Note, however, that Kibrik (2003: 634) cites a counterexample to (18) that needs to be checked. 15 / / forget-CAUS-INF ‘I cannot forget your words.’ In Khwarshi, there are two constructions with ĩχʷ- ‘can’. With the first construction the argument in the main clause is always marked with the AT-essive assigned by the matrix predicate. This construction is not available with embedded affective predicates (19a), but is with transitive or embedded intransitive predicates (Kibrik 2003: 642–643). The second construction seems to be clause union or backward control because in this construction it is not the verb ‘can’ that assigns case to its argument but the embedded predicate. This construction is equally permissible with extended intransitive, transitive and affective predicates (19b). (19) Khwarshi (Kibrik 2003: 642–643) a.*uža-qa [ _ kad y-akʷ-a] ĩχʷ boy-AT LAT girl(II) II-see-INF can (Intended meaning: The boy can see the girl.) b. Khwarshi (Kibrik 2003: 642–643) [uža-l kad y-akʷ-a] ĩχʷ boy-LAT girl(II) II-see-INF can ‘The boy can see the girl.’ The same restriction is attested for the Hinuq complement construction with the modal verb -aq’e- ‘must’ (Forker 2013: 625). The construction in which the modal verb assigns a spatial case to its argument is unavailable with plain affective verbs (20a). If they are causativized and therefore become transitive predicates, the construction becomes grammatical (20b). Otherwise a clause union or backward control construction must be used in which all arguments receive their case marking from the embedded affective predicate (20c). (20) Hinuq a.*xoddo-ƛ’o [_ baru y-et-a] y-aq’e husband-SPR DAT wife(II) II-like-INF II-must (Intended meaning: A husband must love his wife.) b. Hinuq xoddo-ƛ’o [_ baru husband-SPR ERG wife(II) ‘A husband must love his wife.’ y-eti-r-a] II-like-CAUS-INF c. Hinuq [xoddo-z baru y-et-a] husband-DAT wife(II) II-like-INF ‘A husband must love his wife.’ 16 y-aq’e II-must y-aq’e II-must Furthermore, there is arguably a similar difference between two types of complement constructions with the phasal predicate ‘begin’. According to Kibrik (2003: 600–648), the Tsezic languages make use of the construction illustrated in (21a) with extended intransitive, canonical transitive, and extended transitive predicates in the complement of ‘begin’. In this construction the matrix predicate assigns absolutive case to the NP referring to the beginner. In contrast, with embedded affective predicates the second construction is the norm in which the NP that denotes the person who begins an action is located within the complement clause and accordingly receives its case marking from the embedded affective predicate (21b). (21) Bezhta a. kid [_ öžö Ø-äƛ’el-al] girl(II) ERG boy(I) I-beat-INF ‘The girl began to beat the boy.’ b. Bezhta [kibbal öžö Ø-ega-al] girl.LAT boy(I) I-see-INF ‘The girl began to see the boy.’ y-ũ xöɬ-iyo II-begin-PST Ø-ũ xöɬ-iyo I-begin-PST transitive affective However, the data are not so clear. For Hinuq and Tsez most affective predicates as well as canonical intransitive and transitive and other predicates allow for both constructions (22b). In Tsez, but not in Hinuq, there is a slight difference in semantics between the two constructions. According to Polinsky & Potsdam (2002: 252, footnote 7), example (22a) means that the girl began to perceive the sounds of the father telling the story, while example (22b) has the additional meaning that the girl became consciously aware of the story. (22) Tsez (Polinsky & Potsdam 2002: 252) a. kid [_ babiw-s xabar teq-a] girl(II) LAT father-GEN story hear-INF ‘The girl began to hear the father’s story.’ b. Tsez (Polinsky & Potsdam 2002: 252) [kid-be-r babiw-s xabar teq-a] girl(II)-OBL-LAT father-GEN story hear-INF ‘The girl began to hear the father’s story.’ y-oq-si II-begin-WPST y-oq-si II-begin-WPST This all points to a preference for embedded affective predicates in modal and phrasal complement constructions to occur in clause union or backward control constructions in which only the affective predicate but not the matrix complement-taking predicate assigns case to all arguments of the construction. There is no obvious explanation for this preference, but perhaps overt case marking of the more prominent argument by dative/lative/IN-essive is important for the identification of the semantic roles and thus is better kept. This is only possible in constructions such as (20c), (21b), and (22b). One might suspect then that maybe this is part of a general tendency in Tsezic to have more overt 17 dative/lative/IN-essive arguments of affective verbs than, for instance, ergative-marked agents with transitive verbs. Actually, we have a few numbers that support this hypothesis. We counted the number of overt vs. covert arguments in 439 clauses of Hinuq narratives: from the 74 clauses with affective predicates only 17 (= 23%) lacked an overt experiencer, whereas from the 365 clauses with transitive predicates 223 (= 61%) lacked overt agents. However, further substantiation of this hypothesis by more data is obviously needed.5 6. Affective predicates and extended intransitive predicates In the preceding Sections 2–5 we pointed out the morphosyntactic differences between affective constructions and canonical transitive constructions. However, one might ask whether it would be more fruitful to compare affective constructions with extended intransitive constructions since the latter look at a first glance more similar to affective constructions than canonical transitives. Extended intransitives contain one argument in the absolutive case and a further argument in a spatial case or in Hinuq occasionally in the dative. Two examples are provided in (23a) and (23b). (23) Khwarshi (Khalilova 2009: 351) a. uže kandɨ-qo-l ʕezeʕan guc’-un boy(I) girl.OBL-AT-LAT much look-CVB ‘The boy was looking at the girl for a long time.’ Ø-eč-i I-be-WPST b. Hunzib (van den Berg 1995: 123) ože xõt’ox-a-ɬ Ø-ũxe-r boy(I) frog-OBL-CONT I-turn.into-PRET ‘The boy turned into a frog.’ However, it is hard to find any similarities between extended intransitive predicates and affective predicates. To begin with, the case marking of the non-absolutive argument of extended intransitives is lexically determined by the predicate and can vary. Thus, in the Khwarshi example in (23a) the AT-lative must be used; in the Hunzib example (23b) it is the CONT-essive. Other frequently used cases are the SPR-essive, the AT-essive, and the SPR-lative (Forker 2010). In contrast, for affective verbs there is no choice. The case marking of the non-absolutive argument is always dative/lative/IN-essive. And these latter cases are (almost) never used with extended intransitive verbs. Second, the default word order is different. Tsezic languages are predominantly verb final at the clause level. The most frequent word order in affective clauses is dative/lative/IN-essive before absolutive, whereas for extended intransitives it is absolutive before spatial case. Third, with affective verbs it is the dative/lative/IN-essive argument that is required to be animate. With extended intransitive verbs there is often no such restriction, and if there is any, then it would be rather the other way around, i.e. the absolutive argument must have an animate referent. Fourth, many extended intransitive verbs can be used as simple intransitives, e.g. if the Khwarshi verb guc’- ‘look’ occurs without a goal it simply means ‘look around’. Fifth, extended intransitive verbs never allow for a reversal of case marking in reflexive and 5 A similar tendency has been suggested for German by Barðdal (2006) to explain the difference between dative subject-like obliques in Icelandic and German with regard to conjunction reduction, control infinitives, and other constructions involving ellipsis. 18 reciprocal constructions (see Section 4.2). In sum, despite the fact that extended intransitives and affectives mark one argument with a spatial case (except for Hinuq) it is virtually impossible to find any other common properties between the constructions. 7. Is the more prominent argument of affective predicates a subject? Before trying to answer this question one caveat is in order. Is the classical subject relation at all applicable to the Tsezic languages; i.e. is there some indication that we can group together S and A under the heading of ‘subject’? Or should we rather assign S and P to the same category of ‘ergative subject’? Case marking and agreement obviously argue for the latter, but this might be an instance of pure morphological ergativity. In order to show whether Tsezic languages have subjects or not it is necessary to apply a number of tests to the transitive construction which due to lack of space is not possible in this paper. We rather refer to the existing literature which says that at least in Tsez and Hinuq there is, if at all, a weak indication of an S/A pivot (cf. Comrie 2004, Forker 2011). This fits the general claims on Nakh-Daghestanian languages which have repeatedly been analyzed as displaying only morphological ergativity and lacking any subject relation (Nichols 1980, Crisp 1983, Haspelmath 1991, Comrie et al. 2011). Yet one aim of this paper is to shed light on the syntactic properties of affective constructions. So it is legitimate to ask whether Tsezic dative/lative/IN-essive arguments can be compared to quirky subjects or dative subjects as for instance in Icelandic (Barðdal 2006) or in South Asian languages (Verma & Mohanan 1990). Thus we will apply the commonly used subject tests and check if dative/lative/IN-essive arguments pattern with S/A or rather with P. As and dative/lative/IN-essive arguments are case-marked and consequently do not trigger verbal agreement (Section 2). They can be the addressee in an imperative construction for those Tsezic languages that allow imperative formation with affective predicates (Section 3). They both can undergo valence-changing processes, though to various extents (Section 3). They both predominantly precede the verb and occur before the second argument in their clause. They can be the target of relativization, conjunction reduction and equi-deletion in complement clauses (Section 5), and they both control reflexive and reciprocal pronouns (Section 4). As has been shown throughout this paper, there are also several differences between As and dative/lative/IN-essive arguments. Nevertheless these differences do not support the inference that the latter are less subjectlike than As, but simply that they sometimes behave differently because their semantic properties are different. But before concluding that dative/lative/IN-essive arguments behave, with some restrictions, just like typical subjects, we must examine whether the absolutive arguments of affective verbs behave differently from dative/lative/IN-essive arguments or whether the absolutive arguments can also be compared to As. First, the absolutive arguments differ from As in case marking and verbal agreement, and they can normally not be the addressee of an imperative (24). (24) Bezhta * tušman, mi Ø-aːt’! enemy 2SG I-love (Intended meaning: Enemy, you be loved!) 19 Moreover, they generally pattern like Ps when equi-deleted in complement constructions. Thus, it is common to use a personal or a reflexive pronoun whenever the absolutive, but not the dative/lative/IN-essive argument is coreferential with a main clause argument. For example, in (25a) the IN-essive argument in the complement clause has been omitted since it is coreferential with the S argument in the matrix clause. In contrast, in (25b) the coreferential argument in the complement clause is expressed as a reflexive pronoun rather than being omitted because it has the position of the absolutive and therefore the less prominent argument in its clause (see the Bezhta sentences in Kibrik (2003: 604) for an example with a personal pronoun in the complement clause). In Hunzib in such a case additional marking on the verb by the prohibitive suffix is required (25b). The same phenomenon is attested with transitive predicates. This means that P but not A arguments in complement clauses that are coreferential with arguments in the main clause usually appear as reflexive pronouns (van den Berg 1995: 131). (25) Hunzib (van den Berg 1995: 131) a. ože hĩč’e [_ kid boy fear.PRS IN girl(II) ‘The boy fears to see the girl.’ y-ãc’-á] II-see-INF b. Hunzib (van den Berg 1995: 131) kid hĩč’e [oždi-i žu y-ãcə-dor] girl fear.PRS boy.OBL-IN REFL II-see-PROH ‘The girl fears that the boy should see her.’ Nonetheless, this rule does not seem to be equally applicable to all Tsezic languages. In the Khwarshi examples in (26a, b) the experiencer as well as the stimulus can be coreferential with an argument in the main clause and left unexpressed in the complement clause. Although in (26b) the coreferential null is in the less prominent position, no reflexive or personal pronoun is needed. (26) Khwarshi (Kibrik 2003: 644) a. uža-l [_ kad y-akʷ-a] boy-LAT LAT girl(II) II-see-INF ‘The boy wants to see the girl.’ goqše want.PRS b. Khwarshi (Kibrik 2003: 644) kandɨ-l [uža-l _ y-akʷ-a] goqate girl-LAT boy-LAT ABS II-see-INF want.PRS.NEG ‘The girl does not want to be seen by the boy.’ There are also a number of similarities between As and absolutive arguments of affective verbs, e.g. they can both be the target of relativization and conjunction reduction (Section 5), though the former criterion is not very significant because the vast majority of Tsezic argument and adjunct types can be relativized on (e.g. locations, goals, instruments, etc.). Concerning reflexivization and reciprocalization, we can state that As and absolutive 20 arguments of affective verbs share some characteristics because As can normally be controllers, and in Hinuq they must be controllers. In contrast, in Bezhta reciprocal constructions As must be controllees (Comrie et al. 2011). Absolutive arguments of affective verbs can always be controllers (see Table 2 above). In Bezhta reciprocal constructions they must be controllers, and in Tsez and Khwarshi they are strongly preferred over the dative/lative/IN-essive argument. But the same can be said of A arguments and dative/lative/IN-essive arguments and of Ps and absolutive arguments, so this criterion is also not particularly noteworthy. In sum, from a syntactic point of view the dative/lative/IN-essive arguments display more subject properties than the absolutive arguments of affective verbs, and they share more characteristics with S/As than with Ps. 8. From meaning to form: Affective predicates in the wider context of experiential constructions As already mentioned in the introduction and as can be seen from the list of predicates in Table 1 (Section 2), not all affective verbs assign semantic roles to their arguments that fit the customary characterizations of experiencers and stimuli. On the other hand, in all Tsezic languages there are many more bivalent verbs that can be truly called experiencer predicates. Such variation in the morphosyntactic marking of experiencers and stimuli is cross-linguistically very common and fully expected given the fact that neither experiencers nor stimuli form fully coherent semantic classes (Næss 2007: 185-196). When we take all bivalent experiencer predicates together and group them according to valence classes we arrive at the following four classes (Table 3): Table 3: Experiencer predicates and their valence patterns Predicate class (i) Affective predicates (ii) Transitive predicates (iii) Extended intransitive predicates (iv) Transitive predicates Case marking Experiencer Stimulus dative absolutive ergative absolutive absolutive spatial case absolutive ergative Class (i) contains those affective predicates that have experiential semantics, and class (ii) consists of their causativized counterparts as they have been presented in Section 3.3. An example in case is (6a, b). Classes (iii) and (iv) also stand in the same morphosemantic relationship because class (iv) verbs are normally causativized predicates based on the noncausative verbs in class (iii). We illustrate this by examples from Hinuq and Bezhta. Bezhta has two verbs ‘wonder’, consisting of the invariable word äžaʔib and a light verb. The latter can be either the intransitive verb -aq- ‘be, become, happen’, or the transitive verb -o- ‘do’. In the first instance the result is an extended intransitive predicate (27a); in the second instance we get a normal transitive predicate (27b). The experiencer of both predicates is always the argument in the absolutive, whereas the stimulus takes the SPR-essive in the first instance and an ergative in the second instance. Similarly, Hinuq has an extended intransitive verb sut’ondeɬ- ‘be, become, get angry’ (27c) and a transitive verb sut’ondek’- ‘make angry’ (27d). 21 (27) Bezhta a. hollo-la m-üxo.mäxe-ya-ƛʼa 3PL.OBL-GEN2 HPL-behave-PTCP-SPR ‘We wondered about their behavior.’ äžaʔib wonder b. Bezhta hollo-la m-üxo.mäxe-ya-ɬi-la 3PL.OBL-GEN2 HPL-behave-PTCP-NMLZ-ERG ‘Their behavior astonished us.’ c. Hinuq sira me di-ƛ’o-r why 2SG 1SG.OBL-SPR-LAT ‘Why are you angry with me?’ d. hayɬo-y de 3SG.M.OBL-ERG 1SG ‘He made me angry.’ b-aq-iyo HPL-be-WPST äžaʔib wonder ile 1PL b-o-yo HPL-do-WPST ile 1PL sut’ondeɬ-o? be.angry-PRS sut’ondek’-iš make.angry-PST 9. Conclusion In this paper we have provided a detailed analysis of morphosyntactic and semantic properties that characterize affective constructions in Tsezic languages contrasting them with transitive and extended intransitive constructions. Most of the differences between affective constructions and canonical transitive constructions have a semantic explanation: affective predicates do not have an agentive semantics and are therefore excluded in constructions that imply agentivity, volitionality and control (e.g. imperative, intentional future, potential construction, certain complement clauses). This also explains why some but not all affective predicates allow for the antipassive construction: it depends on whether a plausible context can be found which allows for an affective predicate to have iterative semantics. From a syntactic point of view we showed that the more prominent arguments of affective predicates share more properties with As of transitive predicates than they share with Ps. Still we must note that the two arguments of an affective predicate are more similar to each other than the two arguments of a canonical transitive predicate. The crucial argument in favor of this claim is the peculiar behavior in reflexive and reciprocal constructions: affective predicates allow much more readily than transitive predicates a reversal of case marking by which the stimulus appears as the controller and the experiencer as the controllee (Section 4.2). In contrast, transitive predicates normally show a preference for the standard case marking pattern. This unusual phenomenon can be explained by the semantic properties of the arguments of affective verbs (cf. Ljutikova 2001 for a similar proposal for Tsakhur). Both arguments of affective verbs lack intentions and agentivity. They do not undergo a change of state nor are they typically in motion. They are not brought into being or destroyed by the event or situation denoted by an affective predicate (see also Butt et al. 2006, who point to similarities of experiencers and stimuli in some Asian languages). Thus, the semantic roles of arguments of affective verbs are not so clearly distinct as agent and patient. Or, as Næss 22 (2007: 190) puts it, “[experience events] lack a maximal distinction between one clearly controlling and one clearly affected argument.” Therefore, the case marking of the two arguments in reflexive and reciprocal constructions, in which the referents of the arguments partially or completely overlap, exhibits considerable freedom. The referent of the more prominent argument (e.g. the experiencer) is at the same time the referent of the less prominent argument (e.g. the stimulus). Among the affective constructions in Tsezic languages we find some variation in antipassive and imperative formation, reflexive and reciprocal constructions and complement constructions. These differences may be linked to semantic differences, e.g. Bezhta affective verbs perhaps do not have exactly the same meaning as Tsez or Hinuq affective verbs, but such a claim is hard to substantiate. At any rate they are not very surprising since affective predicates are commonly more variable in their properties and more readily undergo grammaticalization processes than for instance transitive agentive predicates (Primus 1999: 44–45, Butt et al. 2006). For instance, the possibility in Bezhta of forming imperatives from affective predicates may be due to an ongoing grammaticalization process by which affective predicates become more similar to transitive predicates. Abbreviations I – V gender classes, ABS absolutive, ABL ablative, AT AT-essive, CAUS causative, CONT CONTessive, CVB converb, DAT dative, DES desiderative, ERG ergative, FUTDEF definite future, GEN1 first genitive, GEN2 second genitive, HPL human plural, ICVB imperfective converb, IMP imperative, IN IN-essive, INF infinitive, INTER INTER-essive, INTFUT intentional future, LAT lative, LOC locative, M masculine, NEG negation, NMLZ nominalizer, OBL oblique, PL plural, POT potential, PRET preterite, PROH prohibitive, PRS present tense, PRT particle, PST past, PTCP participle, REC reciprocal, REFL reflexive, SG singular, SPR location ‘on’, UWPST unwitnessed past, WPST witnessed past References Abdulaev, Arsen K. & Isa K. Abdullaev. 2010. Cezyas folklor. Leipzig–Makhachkala: Lotos. Barðdal, Jóhanna. 2006. Construction-specific properties of syntactic subjects in Icelandic and German. Cognitive Linguistics 17, 39–106. Butt, Miriam, Scott Grimm & Tafseer Ahmed. 2006. Dative subjects. Talk presented at the NWO/DFG Workshop on Optimal Sentence Processing, Nijmegen. Comrie, Bernard. 2001. Love your enemies: Affective constructions in two Daghestanian languages. In Ferenc Kiefer, István Kenesei & Robert M. Harnish (eds.), Perspectives on semantics, pragmatics, and discourse: A festschrift for Ferenc Kiefer, 59–72. Amsterdam: Benjamins. Comrie, Bernard. 2004. Oblique-case subjects in Tsez. In Peri Bhaskararao & Karumuri Venkata Subbarao (eds.), Non-nominative subjects, Vol.1, 113–127. Amsterdam: Benjamins. Comrie, Bernard & Helma van den Berg. 2006. Experiencer constructions in Daghestanian languages. In Ina Bornkessel, Matthias Schlesewsky, Bernard Comrie & Angela D. Friederici (eds.), Semantic role universals and argument linking: theoretical, typological and psycholinguistic perspectives, 127–154. Berlin: de Gruyter. 23 Comrie, Bernard, Diana Forker & Zaira Khalilova. 2011. Alignment typology, reflexives, and reciprocals in Tsezic languages. Proceedings of the 37th Annual Meeting of the Berkeley Linguistics Society, 32–51. Cysouw, Michael & Diana Forker. 2009. Reconstruction of morphosyntactic function: Nonspatial usage of spatial case marking in Tsezic. Language 85, 588–617. Dabrowska, Ewa. 1997. Cognitive semantics and the Polish dative. Berlin: de Gruyter. Dahl, Eystein. Morphosyntactic variation and the differentiation of semantic roles: Evidence from Vedic Sanskrit. To appear in Silvia Luraghi & Fernando Zúñiga (eds.), Perspectives on semantic roles. Amsterdam: Benjamins. Forker, Diana. 2011. Grammatical relations in Hinuq. In Vittorio S. Tomelleri, Manana Topadze & Anna Lukianowicz (eds.), Languages and cultures in the Caucasus. Papers from the international conference "Current advances in Caucasian studies", Macerata, January 21-23, 2010, 553–568. Berlin: Otto Sagner. Forker, Diana. 2012. The bi-absolutive construction in Nakh-Daghestanian. Folia Linguistica 46, 75–108. Forker, Diana. 2013. A grammar of Hinuq. Berlin: de Gruyter. Ganenkov, Dmitry. 2006. Experiencer coding in Nakh-Daghestanian. In Leonid Kulikov, Andrej Malchukov & Peter de Swart (eds.), Case, valence and transitivity, 179–202. Amsterdam: Benjamins. Isakov, Isak A. & Madžid Š. Khalilov. 2001. Gunzibsko-russkij slovar’. Moskva: Nauka. Jackendoff, Ray. 1990. Semantic structures. Cambridge, MA: MIT Press. Khalilov, Madžid Š. 1999. Cezsko-russkij slovar’. Maxačkala: Institut JaLI DNC RAN. Khalilova, Zaira. 2009. A grammar of Khwarshi. Utrecht: LOT. Kibrik, Aleksandr E. 1994. Archi. In Rieks Smeets (ed.), The indigenous languages of the Caucasus, Vol. 4, The North East Caucasian languages II, 297–365. Delmar, NY: Caravan Books. Kibrik, Aleksandr E. 1997. Beyond subject and object: towards a comprehensive relational typology. Linguistic Typology 1, 279–346. Kibrik, Aleksandr E. 2003. Konstanty i peremennye jazyka. St. Petersburg: Aletheia. Landau, Idan. 2010. The locative syntax of experiencers. Cambridge, MA: MIT Press. Ljutikova, Ekatеrina A. 2001. Anaforičeskie sredstva. In Aleksandr E. Kibrik (ed.), Bagvalinskij jazyk, 615–681. Moscow: Nasledie. Mohanan, Karavannur P. & Tara Mohanan. 1990. Dative subjects in Malayalam: Semantic information in syntax. In Mandindra K. Verma & Karavannur P. Mohanan (eds.), 43–57. Næss, Åshild. 2007. Prototypical transitivity. Amsterdam: Benjamins. Pandharipande, Rajeshwari. 1990. Experiencer (dative) NPs in Marathi. In Mandindra K. Verma & Karavannur P. Mohanan (eds.), 161–179. Polinsky, Maria & Eric Potsdam. 2002. Backward control. Linguistic Inquiry 33, 245–282. Primus, Beatrice. 1999. Cases and thematic roles: Ergative, accusative and active. Tübingen: Niemeyer. Schulze, Wolfgang. 2009. A new model of metaphorization: Case systems in East Caucasian. In A. Barcelona, K.U. Panther, Günter Radden & L.L. Thornburg (eds.), Metonymy and metaphor in grammar, 147-175. Amsterdam: Benjamins. Toldova, Svetlana J. 1999. Mestoimennye sredstva podderžanija referencii. In Aleksandr E. Kibrik (ed.), Èlementy caxurskogo jazyka v tipologičeskom osveščenii, 629–674. Moscow: Nasledie. 24 van den Berg, Helma. 1995. A grammar of Hunzib. München: Lincom. Verhoeven, Elisabeth. 2007. Experiential constructions in Yucatec Maya. Amsterdam: Benjamins. Verma, Mandindra K. & Karavannur P. Mohanan (eds.) 1990. Experiencer subjects in South Asian languages. Stanford, CA: CSLI Publications. 25