Bi-absolutive constructions in Chechen

2.1 Argument marking in Chechen

Chechen has 10 grammatical cases, viz. ergative, absolutive, genitive, dative, allative, lative, ablative, instrumental/comitative, comparative, and locative. They are all morphologically marked on the noun, except the absolutive case, which is unmarked. The S argument of an intransitive verb is in the absolutive case, as shown in (3). As already indicated in Section 1, the A argument of a transitive verb is typically marked as ergative, whereas the P argument is in the absolutive case (4). However, A arguments of experiencer verbs expressing perception, emotion, feelings, etc. take the dative case, as shown in (5).^[4]

In addition to the case frames described above, a number of other case frames exist for core arguments in Chechen. For instance, (6) and (7) show examples with the A argument in the absolutive case and the P argument in the allative and lative cases, respectively.

Finally, Chechen has a large number of noun-verb complex predicates. The case frames of these complex predicates deviate from the simple verbs illustrated in (3)–(7). For further details on complex predicates and their argument marking see Witzlack-Makarevich and Molochieva (submitted).

2.2 Gender agreement

A sub-class of Chechen verbs shows gender agreement with their absolutive (typically S or P) argument. In Chechen, each noun belongs to a gender class. The gender is not morphologically marked on the noun itself; it appears as a prefix on some verbs (and on adjectives). There are four grammatical genders, marked morphologically as prefixes j-, v-, b-, and d- on the verb, and glossed in our examples accordingly with capital letters. The gender-agreeing verb class makes up only about 30% of the verbal lexicon (Nichols 1994). However, since this class includes auxiliaries and many high-frequency verbs, the frequency of gender agreement in discourse is close to 50% (as shown in Komen et al. [2021]).

Only absolutive arguments can trigger gender agreement. For instance, the S argument in (8) is a v-noun and the gender agreement is marked with the prefix v- on the verb v-odu ‘run’. In transitive constructions, the absolutive P argument triggers gender agreement, as shown in (9): quor ‘pear’ belongs to gender b-, whereas the A argument belongs to gender v- and does not trigger gender agreement. In (10) the verb b-ieza ‘want, love’ takes an A in the dative case. Like in (9), it is the absolutive P argument that triggers gender agreement. Note that gender agreement occurs regardless of whether the absolutive argument is singular or plural (although plural forms can have a different gender compared to their singular counterparts).

3.1 Bi-absolutive constructions in Chechen

As shown in Section 1, Chechen progressive clauses display an alternation between EC and BC. Progressive aspect is expressed by the simultaneous converb, which has the suffix -(u)sh, plus an auxiliary. Aspect is furthermore morphologically marked by the ablaut of the verbal stem. For instance, the root vowel /a/ in the infinitive form of the verb ‘eat’ da’a changes into /u/ in the simultaneous converb du’u-sh, whereas the perfective form is de’a-na. The auxiliary can have the present or past form; for instance, d-u for the present form (see [11a]) and d-ara for the past form (see [11b]; for more details see Molochieva [2011], Molochieva and Nichols [2018]). Note that if the lexical verb belongs to the gender-agreeing class, it will agree with the P argument, whereas the auxiliary agrees with the gender of A.

BCs express an ongoing event; they are often used to describe a situation which is ongoing at the moment of speech, as in (11a), or a situation which happens simultaneously with some other event, as in (11b).^[5]

As an alternative to the BC, an ongoing event can also be expressed by means of a progressive EC. Since ergative A cannot trigger gender agreement, both the auxiliary and the converb agree in gender with the P argument. For instance, in (12) the P argument ‘mouse’ belongs to gender B and both verb forms show B-agreement.

Forker (2012: 82) observes that while most ND languages do not allow BCs with experiencer verbs, Chechen and Ingush are exceptions to this generalization (see Section 3.2). (13a) illustrates a Chechen BC with an experiencer verb. This construction occurs in contexts where the event is in focus (13b) or happens during a restricted time period (13c). (13b) was produced while watching a TV show. The speaker referred to the characters in the show when it was obvious that one of the characters was not honest about his feelings. The sentence in (13c) was also produced while watching a TV show. The scene showed a man standing far away from two people having a conversation. The conversation was audible for the viewers, but it was not clear whether it was audible for the person in the scene.

However, not all experiencer verbs can appear in BCs. Experiencer verbs such as douza ‘ recognize, be acquainted’, xieta ‘seem’, dagadaa ‘ remember’, k’ordada ‘ get tired’, etc. cannot be used in a BC. Experiencer verbs can be subdivided into three groups: the first group, represented by ‘love’ in (13), is very flexible; the second group contains verbs which are usually used with dative A but in a certain context can be used in BCs, for instance, xaza ‘hear’; and the third group comprises verbs which do not allow BCs at all, for instance, douza ‘ recognize, be acquainted’, xieta ‘ seem, etc.

3.2 Previous research on bi-absolutive constructions in Nakh-Daghestanian languages

The most comprehensive overview of BCs is Forker (2012).^[6] She states that BCs are attested in all branches of the (ND) family, with the possible exception of Khinalukh (for which she lacks the relevant data). While Forker’s overview contains data on BCs in a large number of ND languages, detailed descriptions of the construction in individual languages are not common. Moreover, Forker (2012: 76) remarks that “because BCs are hardly ever attested in natural texts or speech, all examples collected during fieldwork are elicited”.

Forker (2012: 78–79) lists a number of formal properties of BCs that are recurrent across the ND languages: (i) both arguments (A and P) are in the (unmarked) absolutive case; (ii) the predicate is mostly periphrastic, with a non-finite lexical verb (a participle or a converb) and a finite copula or auxiliary; (iii) if agreement can be expressed, the lexical verb agrees with P and the finite verb with A; (iv) the clause has (some form of) imperfective aspect.^[7] One of the typical functions of BCs, according to Forker (2012: 80) is to describe the agent as involved in a concrete ongoing action. Similar lists of properties can be found in Ganenkov (2021) and Ganenkov and Maisak (2021).

In addition to these general formal and functional properties, Forker (2012: 82ff) goes on to describe a number of restrictions on the use of BCs (as opposed to ECs) that apply variously across ND languages. In particular, these restrictions apply to, among other things, the types of verbs that may participate in BCs, the animacy status of the A argument, and word order.^[8] First, as already mentioned, unlike Chechen most ND languages disallow BCs with verbs other than those with a default ergative-absolutive case frame, thus excluding experiencer verbs (and other non-canonical frames). Second, most languages do not allow inanimate A arguments in BCs (but see Chumakina and Bond [2016] on Archi). With respect to this restriction, Forker gives a quite detailed description of the situation in Chechen:

For human subjects the BC is either the only possibility or it is preferred. For inanimate subjects the EC is the grammatical construction. Animate, but nonhuman subjects lie in between. (Forker 2012: 84)

Finally, BCs in many ND languages do not allow word-order variants in which an element (either the A argument or some other constituent) intervenes between the P argument and the lexical verb; P must directly precede the verb.^[9] Moreover, the lexical verb and the auxiliary form a fixed unit. However, our Chechen data display more word-order variation than is generally considered possible in BCs. We will discuss this further in Section 5.2.

Since Forker’s comprehensive study, a few more studies on BCs have appeared, either in the form of language-specific analyses (Gagliardi et al. [2014] on Lak and Tsez; cf. Ganenkov [2016] on Lak; Chumakina and Bond [2016] on Archi), or as part of the description of a broader phenomenon in ND languages (see Ganenkov [2021] on ergativity and Foley [2021] on agreement).^[10]

Gagliardi et al. (2014) present a contrastive analysis of BCs in Lak and Tsez in the generative framework. With respect to the form of the construction, Lak is unusual in having a synthetic BC type, alongside a periphrastic type that resembles BCs in other ND languages including Tsez. Regarding function, the authors emphasize that the meaning of BCs in both languages is similar to its EC counterpart but that the BC alternant emphasizes the fact that the event is ongoing and that this event has come about through an active agent.^[11] In terms of restrictions, it is noteworthy that Lak patterns with Chechen (and Ingush) in allowing the BC alternation with experiencer predicates (cf. Section 3.1).

Chumakina and Bond (2016) observe that BCs in Archi display two possible converbal forms: -ši (progressive) and -mat (continuous).^[12] Interestingly, whenever the latter form is used, the BC is the only option; the progressive EC is ungrammatical. Moreover, Chumakina and Bond ascribe a special meaning to BCs with the -mat converb: these constructions express that the predicated event is or was continuing longer than anticipated. By contrast, BCs with the -ši converb, which alternate with ECs, have a broad progressive meaning, akin to BCs in other ND languages. In fact, this meaning is stable across progressive constructions that do not have both arguments in the absolutive case, similar to the observations for Lak and Tsez in Gagliardi et al. (2014).

With regard to restrictions on BCs, Chumakina and Bond (2016) show that Archi (like Lak, Chechen, and Ingush) not only allows experiencer predicates to form BCs but also allows inanimate agents in these constructions. Typically, in such cases an instrument is interpreted as an agent; see (14) (Chumakina and Bond’s [2016] [40a]):^[13]

Importantly, Chumakina and Bond (2016) echo Forker’s (2012) observation that BCs are rare, especially in narrative texts. They explain this in terms of the discourse function of imperfective clauses in general; i.e. to provide contextual information, which is typically expressed by dependent converbial clauses in ND languages. However, they argue that the distribution of the two alternants suggests that the BC can be used in response to questions and for contrastive focus. For instance, (14) was produced in response to the question ‘Can I borrow your mower?’.

In the Oxford Handbook of Languages of the Caucasus, Foley’s (2021, Section 20.2.2.2) and Ganenkov’s (2021, Section 18.4.1) discussions of BCs in ND languages mostly summarize the points already covered in the studies cited so far. However, we consider it worthwhile to highlight Ganenkov’s observation that BCs have sometimes erroneously been considered a case of aspect-based split ergativity. The reason why this view is not accurate is that the use of the BC is – at least under a certain set of conditions – optional, rather than obligatory (although note that the above-mentioned continuous converb BC in Archi constitutes an exception). Thus, it is better to consider it an alternation rather than a split. Below, we will further scrutinize the factors determining Chechen speakers’ actual use of the BC.

In sum, previous research on BCs across ND languages identifies a number of common properties, in particular in terms of agreement pattern of core arguments, semantic function, and low usage frequency. Dimensions of variation include the amount and type of restrictions on BCs in terms of verb type(s) and animacy of the agent.

5.1 Stimuli

Our visual stimuli consisted of a combination of pictures and short video clips.^[18] These included 14 target transitive events intended to elicit BC sentences:

1. four videos with a human (agentive) A, e.g., a video of a man chopping wood;

2. four videos with an inanimate (agentive) A, e.g., a video of a river carrying away cars;

3. four pictures with a non-human animate (agentive) A, e.g., a picture showing a cat climbing up a tree (two examples appear in Figure 1);^[19]

4. two pictures with a human experiencer A, e.g., a girl looking at a boy.^[20]

Figure 1:

Two examples of visual stimuli involving animate non-human agents (Gagarina et al. 2019).

Our stimuli also included filler items (both videos and pictures) that depicted intransitive events, e.g., a picture of a boy and a dog sleeping (taken from Meyer 1969), a video of a person sitting on a chair.

It should be noted that the reason why we do not have a balanced set of stimuli with respect to both stimulus types (pictures vs. videos) and event types (agentive vs. experiencer) is connected with the difficulty in finding/creating visual stimuli for all types of events relevant for our study. In general, while Forker (2012) mentions that all BC data in her paper are elicited, we are not aware of any earlier studies which elicited BCs specifically on the basis of visual stimuli.^[21] In this context, our study should be seen as a first attempt to assess this construction type in a semi-controlled manner.

5.2 Results

We collected a total of 100 utterances, of which we had to exclude 46 that were not relevant for our study (we will discuss some of these cases later).^[22] Table 1 presents the distribution of different case frames with respect to the stimuli/event types in the remaining 54 utterances, which we further annotated with respect to the aspect used.

As in Section 4, we observe that progressive aspect does not systematically trigger a BC when A is animate (either animal or human), but it is by far the most common option (20 out of 21 progressive sentences with human As are BCs, and 6 out of 7 progressive sentences with animal As are BCs). This is illustrated in (23) and (24):

Speakers sometimes used a non-progressive, perfect EC with a human A, as shown in (25a). However, one speaker self-corrected and produced instead the BCs in (25b) and (25c).

However, even when the sentence is in the progressive form, a human or animate A argument can still be marked with ergative case. This, is infrequent, though, and we have only one case in our data, shown in (26). In this example, the A argument is actually unexpressed, but the gender agreement shows that it would have been ergative if it were overt, since it is the absolutive P argument that triggers gender agreement on both the converb and the auxiliary. With inanimate As, only 2 out of 8 progressive sentences were BCs. They are given in (27) and (28):^[23]

However, after uttering example (28), the speaker immediately self-corrected and produced the progressive EC in (29), presumably because A is inanimate.

As expected, based on the statement made by Forker ([2012: 84]; see Section 3.2 above), the stimuli with inanimate A arguments were described mostly with progressive ECs. Alternatively, a non-progressive EC was used, either in the perfect, as in (30), or in the present, as in (31).

With regard to verb class, we observe that, as expected, experiencer verbs are also allowed in BCs, as is shown in (32):

However, experiencer constructions were in general harder to elicit. Out of 10 datapoints that were meant to elicit a sentence with an experiencer A, only half actually worked in triggering such a sentence. (For comparison: for human agentive As, the rate was 20 BCs out of 26 responses, and for inanimate agentive As, it was 15 out of 21.) It is noteworthy that our stimuli with animal agentive As had an even lower rate of success (9 out of 24); we will return to this below.

It is important to mention that Table 1 includes four examples of noun-verb complex predicates, as in (33) and (34) below, as well as (32) above. We included these sentences because form-wise they are BCs, given that the nominal element of the complex predicate is in the absolutive case, as is the A argument (which would be in ergative or dative in the default case). However, the notional P argument is in the dative.

Furthermore, some of the BCs we elicited in response to the video stimuli display a non-default word order pattern, in which the converb and the auxiliary are not adjacent. An example is given in (35).

We propose that this variation represents a focus construction. In Chechen, word order is not the only way to express focus; the enclitic =q and the proclitic ma= can also be used for this purpose. When the enclitic =q attaches to a finite verb, it scopes over the entire construction or clause, which becomes emphasized or in focus. This happens in (36) below. Notably, the same enclitic is attested in closely related Ingush. Nichols (2011:283) defines it as expressing ‘cumulative focus’, translated as ‘The point is…’, ‘What’s going on is…’.^[24]

Like the enclitic, the proclitic ma= also attaches to the main verb and emphasizes the entire clause, as illustrated in (37). The second part of the example shows how the proclitic combines with the interrogative marker -i.

The constructions involving a combination of a focus marker with non-default word order, as in (36) and (37), show that this word order pattern is compatible with the focus interpretation. These focus marking strategies may have been triggered by the speakers’ placing emphasis on the actions they perceived in the stimulus videos. Recall also from Section 3.2 that Chumakina and Bond (2016) describe the function of BCs in Archi as one of (contrastive) focus and/or of answering questions. Arguably, the constructions in (35)–(37) are answers to the question ‘What do you see in this video?’. Moreover, the case of non-default word order attested in our spoken corpus (see [17] above) also constitutes an answer to the question (prompted by hearing a breaking noise): ‘What is happening there?’. Whatever the exact motivation is behind the non-adjacent ordering of auxiliary and converb, our data clearly present counterevidence to the claim in earlier literature that the two verb forms form a fixed unit (cf. Section 3.2).

Finally, to illustrate some of the response constructions that we did not include in our analysis, consider the following examples, which were used to describe one of the videos, which showed an avalanche. Example (38a) is an intransitive progressive construction. The lexical verb is daa ‘come’, used as simultaneous converb. Thus, the speaker starts describing the event s/he sees with the progressive. However, in the subsequently uttered (38b) the present is used. The A argument cuo ‘3sg.erg’ refers back to the S argument in (38a).

In addition, many of the pictures with animal agents were described with complex constructions such as subordinate clauses. The following examples were produced to describe the picture stimuli illustrated in Figure 1 above. Example (39) is a purpose clause with an embedded infinitival clause hwoozarchij d-a’a ‘birds.abs d-eat.inf’. The absolutive cisk ‘cat’ is the argument of the intransitive verb d-ooda-sh ‘d-go-cvbsim’. The matrix clause is in the progressive tense. A syntactically more complex example is shown in (40). The matrix clause is chwogal q’aig-ie d-oozhaj-na ‘the crow knocks over the fox’ and several clauses describing other events are embedded into the matrix clause: oush laaca ‘to catch them’, gherta-sh ‘trying’, and te’qietta ‘attacked’. Some of the events are described using the simultaneous converb, which expresses the progressive aspect; for instance, doodash ‘going’ and ghiertash ‘trying’. However, it is not possible to reconstruct either a BC or an EC in these complex constructions since there is no verb (auxiliary or lexical) that shows gender agreement with an (unexpressed) absolutive argument.

Importantly, as mentioned earlier, speakers also used BCs to describe the animal picture stimuli, but they did so only with the inherently durative (simple) verb larbie ‘watch’, as shown in (41):

In sum, our elicitation data reinforced the corpus data to the extent that the BC is the most common construction when the A argument is a human agent. However, the BC still alternates with the EC. In response to our stimuli with animal agents, the BC was also used, but more often these pictures elicited complex constructions. Potentially, this happened because the stimuli contained multiple participants as well as background scenes, which may have enhanced the use of narrative style (comparable to our spoken-corpus data). With inanimate agents, ergative A arguments were preferred, whether in progressive sentences or not, but again this is not absolute since the BC was also attested. Regarding verb class, in response to stimuli with experiencer verbs, speakers did produce BCs as predicted, but overall descriptions of the relevant events were harder to elicit. Finally, we found several cases of non-adjacent ordering of auxiliary and converb and tentatively interpret this as a form of focus marking.