Causatives in Classical Armenian

1 Introduction

Classical Armenian, the oldest recorded variety of Armenian attested since the 5th century, expresses causation with lexical and synthetic causatives as well as three analytic constructions built with the auxiliary verbs aṙnem ‘make’ and tam ‘give’.^[1] The analytic constructions comprise aṙnem + noun/adj. (make-factitive), aṙnem + infinitive (make-causative) and tam + infinitive (give-causative).^[2] The description of these causative expressions in standard grammars of Classical Armenian remains incomplete. The present paper is a contribution to their study within a typologically oriented framework.

As the only productive valency-increasing derivational category of Classical Armenian, the synthetic causative has received the most attention out of the abovementioned causatives in previous scholarship (see Abrahamyan 1976: 179–186; Aṙak‛elyan 2010: 161–168; Jensen 1959: 36; Meillet 1913: 26; Minassian 1996: 76–77; Tumanyan 1971: 372–377). However, the existing accounts offer only a very general description of its lexical distribution and function. For example, Tumanyan (1971: 373) reports that the synthetic causative can be derived from the majority of transitive and intransitive verbs, cf. aršawim intr. ‘gallop’ → causative aršawecʽucʽanem ‘make gallop’, iǰanem intr. ‘go down’ → causative iǰucʽanem ‘bring down’, yišem tr. ‘remember’ → causative yišecʽucʽanem ‘make remember’. This generalization overlooks important constraints on the formation of the synthetic causative and its relation to other types of causatives. The cited grammars offer an even more cursory account of the analytic causative constructions. Characteristically, Minassian (1996: 77, 207–208) describes the make-causative as expressing causative meaning (“sens factitif”) in contrast to the give-causative expressing causative meaning with a permissive connotation (“sens factitif, permettre, autoriser”) but gives no further details on their lexical distribution, use, or relative productivity.

Aside from descriptive grammar, the study of Classical Armenian causatives is furthermore relevant for the discussion of cross-linguistic generalizations on the correspondence between the formal and semantic parameters of causatives (see Dixon 2000; Haspelmath 2016; Shibatani and Pardeshi 2002). The present paper aims to contribute to the discussion of the Classical Armenian evidence in the cross-linguistic perspective.

2 Causatives and the autonomy of the causee

The semantics of causation constitutes a complex domain with a plethora of parameters concerning the causing and caused sub-events and the nature of their relation, see Beavers and Koontz-Garboden (2020: 48–55) and Levshina (2022) for discussion. The present paper will address only some of those parameters and examine a possible correspondence between the formal type of the causative, on the one hand, and the semantic properties of the causee in relation to those of the first argument of the base verb, on the other. One of the key parameters, discussed in previous scholarship, is the degree of autonomy of the causee. This parameter is crucial for predictions that the more autonomous the causee, the more effort is required on part of the causer to initiate its change of state and the more complex the causative expression tends to be in languages that have multiple ways to express causation. Thus, Dixon (2000: 61–78) discusses the influence of control, volition, and affectedness of the causee on the morphosyntactic complexity of the causative. Similarly, Shibatani and Pardeshi (2002: 8) argue that the degree of volition of the causee is one of the two main parameters, along with spatiotemporal overlap, that determines the coding complexity of a causative. The scale of causative meanings characterized by different degrees of the causee’s autonomy can be grasped in terms of the direct-indirect causation scale; see Levshina (2022) on the semantic parameters associated with the direct-indirect distinction.

The autonomy of the causee partly depends on the semantic type of the base noncausal verb. Based on the results of the Valency Patterns Leipzig (ValPaL) project, Wichmann (2015: 171–172) concludes that verbs describing actions that are amenable to external control (e.g. ‘laugh’, ‘fear’, ‘sink’) more easily undergo causativization than verbs that describe actions of relatively more autonomous volitional participants which are less easily controlled by an external agent (e.g. ‘send’, ‘say’, ‘teach’). This explains in particular why ingestive and cognition verbs (‘eat’, ‘know’) belong to the transitive verbs which, according to the dataset of ValPaL, most often undergo causativization (Wichmann 2015: 171). Amenability to external control correlates with the parameter of affectedness of the first argument characteristic of many intransitive and transitive base verbs of the semantic middle domain; a participant that is typically affected by an external stimulus or one’s own action is also prone to be affected by an external agent (see Kemmer 1993; Næss 2007; Shibatani 2002: 6).

The correlation between the semantic type of the base verb, the autonomy of the causee, and morphosyntactic complexity can be responsible for the lexico-syntactic distribution of competing causatives. Thus, increasing autonomy of the agent of the base verb correlates with increasing syntactic transitivity on the one hand (see Tsunoda 1985; Malchukov 2015, Comrie et al. 2015) and with the morphosyntactic complexity of (or else a constraint on) the derived causative, on the other (Haspelmath 2016). This correlation is confirmed by the evidence of languages with synthetic and analytic causatives, which are more likely to be derived from intransitive and transitive verbs, respectively (Nedyalkov and Silnitsky 1973: 7; Shibatani 2002: 5; Wichmann 2015: 171).

The aforementioned interrelated parameters have been presented by Haspelmath (2016) in a set of cross-linguistic generalizations on the formation of causatives. The principal generalization (Universal 1), on which a series of additional ones is based, reads: “[t]he higher the noncausal meaning of a causal-noncausal pair is on the spontaneity scale, the longer and the more analytic any causative marker on the causal verb form will be” (Haspelmath 2016: 38). Haspelmath’s spontaneity scale model includes the following types of meanings: agentive transitive (an agent impinges directly on a patient, e.g. ‘break’) > unergative = agentive intransitive (volitional actions that are not directed at another participant, e.g. ‘walk’) > unaccusative = anticausative^[3] (subdivided into “automatic” processes that are easily construed as occurring on their own, e.g. ‘freeze’, and “costly” processes that typically involve some energy input, e.g. ‘split’) > agentful = patientive (processes which have an agent-oriented manner component in their meaning and tend not to occur without an external agent, e.g. ‘be cut’). The degree of spontaneity decreases from left to right, see Figure 1. A series of additional generalizations, based on Universal 1, have also been proposed and will be discussed in Section 4.

Figure 1:

The spontaneity scale (Haspelmath 2016: 34).

Furthermore, Haspelmath argues that references to morphosyntactic complexity (e.g. synthetic vs. analytic) can be omitted in the formulations of the universals and that it is the phonological length alone that correlates with the position of the noncausal meaning on the spontaneity scale, which in turn is ultimately determined by the frequency of use (Haspelmath 2016: 54–55). According to this hypothesis, verbs with a higher autonomy of the subject occur less frequently in a causative context and therefore receive more overt coding. Haspelmath thus claims that there is a straightforward correspondence between the semantic type of the base verb and the direct-indirect distinction of causatives because “causatives of unergatives and transitives must express indirect causation (almost by definition), and only causatives of unaccusatives can express direct causation” (Haspelmath 2016: 46).

In order to scrutinize to what extent the Classical Armenian evidence conforms to the aforementioned cross-linguistic generalizations, its five causative formations will be specified with respect to their formal features (morphosyntactic complexity and phonological length), to the semantic type of causation, and to the semantic role of the first argument of the base verb. The last parameter will be assessed based on four major types of semantic roles: patient (a participant that corresponds to the first argument affected by an external agent), undergoer (a participant that corresponds to the first argument, the state of which changes spontaneously), experiencer (a participant that corresponds to the first argument, which is affected by a stimulus), agent (a participant that corresponds to the first argument, initiating a volitional action directed at itself or another participant).

These semantic roles of base verbs can be arranged in order of their increasing autonomy on the spontaneity scale: patient > undergoer > experiencer > agent. The relative autonomy of undergoer and experiencer is not straightforward, given that there is no clear-cut semantic boundary between anticausative verbs with an animate undergoer and experiential verbs of sensation and emotion (Kemmer 1993: 128). Whenever the stimulus is expressed, the experiential reading is clear in both the base verb and its derived causative. A contextual reading becomes ambiguous when the stimulus is not expressed, however. Yet given that the experiencer typically corresponds to a sentential participant which can, at least for some predicates, control the experiential event to some extent, it shares more semantic features with agent than undergoer and can therefore be allocated between the two on the spontaneity scale. Such a relative position of the semantic roles agrees with the relative position of the semantic types on the semantic map proposed within the VaLPaL project (Comrie et al. 2015: 8).

The morphosyntactic features and lexical distribution of the five Classical Armenian causative formations are described in Section 3, while their comparative analysis in terms of a causative continuum is provided in Section 4.

3 Morphosyntactic coding and lexical distribution

The personal forms of the Classical Armenian verb express five synthetic tenses (the present, imperfect, aorist, present subjunctive, and aorist subjunctive) and a number of analytic tenses, the most productive of which are the perfect, pluperfect, and future perfect. In the synthetic tenses, valency alternations are coded on the verb by means of inflectional endings are often in conjunction with a productive valency-increasing suffix -oycՙ- (or its rare variants -oys- and -oyz-). The inflectional types are redistributed among four productive conjugations and a defective verb gom ‘exist’. Besides coding tense, number and person, the endings either express oppositional voice (active vs. mediopassive) or are indifferent to voice, in which case one form is used both transitively and intransitively (the latter type is conventionally termed ‘labile’ in the present paper; cf. Letuchiy 2009). Each verbal paradigm combines both oppositional voice and labile forms, although their ratio varies across the conjugations (see Table 1 for the endings of the present, imperfect and aorist in the four productive conjugations). Note that the use of the oppositional voice is lexically determined only in the present, whereas the imperfect is invariably labile and the aorist invariably distinguishes the two voices except for the 1st person plural (see Bubenik [1997] for a concise overview of the Classical Armenian verbal system).

Given that all of the five causative constructions discussed in the present paper involve personal verb forms, the syntactic transitivity is coded by the active voice in part of every causative paradigm. This common feature aside, the constructions differ in formal and lexico-syntactic properties, which are discussed in Sections 3.1–3.5, and compared in Section 4. The survey of the synchronic evidence is based on the following data. The account of the lexico-syntactic properties of synthetic causatives (Section 3.2.2) is based on the complete list of synthetic causatives attested in the Classical Armenian Bible, and their semantic classification published as a supplement to Kocharov (2023). Given the significantly lower text frequency of the analytic causatives and the potential influence of Greek in the Armenian translation of the Bible, the corpus for the study of the analytic causative has been extended with two non-translated early classical texts, De Deo by Eznik Kołbacʽi and The History of the Armenians by Agathangełos (see the supplementary material at https://doi.org/10.5281/zenodo.7756976). Given that the analytic factitive construction is typical for a specific semantic type of predicates (Section 3.3), it is characterized in general terms without recourse to quantitative data. Due to the lack of a formal difference and the uncertainty of semantic contrasts between lexical causatives and transitive base verbs, as discussed in Section 3.1, lexical causatives will also be characterized without quantitative data; the task of delimiting the two functional types in a comprehensive sub-corpus of early Classical Armenian and establishing the relative frequency of the different semantic types of lexical causatives goes beyond the scope of this paper.

4 The causative continuum of Classical Armenian

The formal and semantic parameters allow us to arrange the five Classical Armenian formations, described in Section 3, in a causative continuum as summarized in Table 11. The formal types are arranged according to their morphosyntactic complexity (lexical < synthetic < analytic), case coding of the causee (accusative < dative), and productivity (as reflected in the type frequency in the study corpus). The make-factitive is placed between the synthetic and analytic causatives because while being an analytic construction, it contains less dedicated verbal morphology (adjective/noun as opposed to the infinitive of the analytic causatives) and its causee is never coded by the dative. The synthetic and analytic causatives with a lower type frequency, as attested in the study corpus, are located to the left of those with the higher type frequency.

The lexical causative typically expresses direct causation with the full control of the causer and a strongly affected causee. This type formally overlaps with the type expressing transitive verbs, and their passive alternation yields a cross-linguistically common passive-anticausative syncretic pattern (see Bahrt 2021: 81–82). This most compact expression of causativity corresponds to noncausal verbs with patientive and costly anticausative meanings. A few suppletive pairs with lexical causatives and base verbs of the semantic types higher on the spontaneity scale (like tesanem ‘see’/cʽucʽanem ‘show’) are exceptional and constitute only very marginal counterevidence against Haspelmath’s Universal 1, see below.

The synthetic causative is the default way to causativize intransitive verbs, the first argument of which is characterized by a certain degree of autonomy (the subject of automatic anticausatives, experiential and agentive verbs; only a few lexical passives have synthetic causatives). Depending on the semantic class of the base verb and its contextual readings, the synthetic causative typically expresses direct or sociative causation. Classical Armenian allows the causativization of a small group of transitive verbs of cognition and verbs with an affected agent-like participant. Synthetic causatives are not derived from canonical transitive and ditransitive base verbs. The first argument of noncausal verbs is strongly affected under causativization, and the contrast between the semantic roles of noncausal verb is largely neutralized. The strong degree of affectedness of the causee is the starting point for the secondary extension of the lexical compatibility of synthetic causatives to canonical patientive direct objects and the lexicalization of causatives to verbal forms with regular transitive semantics.

A few causatives in -oyz- and -oys-, being unproductive types, take an intermediate position between lexical and synthetic causatives. They constitute the periphery of overt causative coding. When not lexicalized, they are derived from anticausatives and express the direct causation (verbs of destruction) typical of lexical and synthetic causatives centered on the patientive and costly anticausative semantic types, on the one hand, and the automatic anticausative type, on the other hand.

The argument features of the make-factitive overlap with those of the lexical and synthetic causatives paired with anticausatives. Their relatively more complex form correlates with the aspectual difference that implies less spatiotemporal overlap between the causing and caused sub-events of the causative situation (cf. Haiman 1985: 110). Note that the degree of morphosyntactic complexity of the make-factitive in Classical Armenian contradicts the model suggested by Lehmann (2016: 926–927), according to which causatives derived from adjectives tend to be lower on the causation continuum than causatives of anticausatives.

The analytic causatives express sociative and indirect causation of agentive transitive verbs of dislocation and ingestion, less consistently of agentive intransitives and rarely of experiential and anticausative verbs. The marginal extension of the construction to the last semantic type can be partly accounted for by the influence of the dídōmi-construction in the Bible, especially when one is dealing with unique occurrences of the analytic causative. Nonetheless, the use of analytic causatives to express the causative-anticausative alternation can be considered as a genuine Classical Armenian phenomenon in a relatively high number of contexts, in particular with anticausatives expressing natural phenomena. The give-causative is not fully grammaticalized; it has a limited lexical distribution and coincides with the infinitival construction with tam ‘give’ expressing purpose. It is often difficult to distinguish between the causative and permissive readings of the give-causative. In causative uses, the first argument of the noncausal verb is not strongly affected and the causee retains the semantic features of the subject of the noncausal verb.

The grammatical status of the make-causative is controversial given that it predominantly translates the Greek poiéō-construction and is not found in early non-translated early classical texts. The make-causative of anticausatives and agentive intransitives expresses sociative or indirect causation. Altogether, the relatively stronger and weaker degree of the causer’s control may be correlated with the coding of the causee by the accusative and dative case, respectively; in support of this semantic approach, see Dixon (2000: 45–59), Kemmer and Verhagen (1994), Kulikov (2001: 890).

None of the examined causatives is productively derived from canonical transitive agentive verbs like ‘break’, ‘hit’, or ‘kill’. The causation of such predicates requires a polypredicative construction in Classical Armenian.

The semantic prototypes of the reviewed causative expressions generally agree with the cross-linguistically common tendency to express events with a higher degree of autonomy of the causee (or more indirect causative meanings) using more complex coding. The complexity of coding concerns morphosyntactic parameters such as the grammatical autonomy and linguistic distance between the morphological constituents expressing the causing and caused sub-events (see Dixon 2000: 74–78; Haiman 1985: 105), rather than the length of the causatives in terms of formal segments (Haspelmath 2008: 22–24, 2016: 54–55; Levshina 2016). In particular, the relative length of the Classical Armenian causatives shows systematic deviations from an expected coding of indirect causation with longer markers; cf. the 3rd singular forms in the present and aorist in Tables 12 and 13 (where the relative position of the make-factitive is established with respect to the synthetic causative derived from denominal anticausative verbs).

The relative length of the 3rd person singular forms in Tables 12 and 13 depends on the type of the aorist stem from which the synthetic causative is derived. In verbs with a root aorist, the length corresponds to the scale of morphosyntactic complexity (lexical > synthetic > analytic) and rather neatly correlates with the direct-indirect causation continuum. However, the give-causative is shorter than the synthetic causative in by far the most productive Classical Armenian verb class characterized by an aorist stem in -ecʽ-, as illustrated by the causatives of bnakem ‘dwell’. Furthermore, the make-causative is phonologically longer than the give-causative but, unlike the latter, is not attested for agentive transitive verbs and thus should take a relatively lower position in the causation continuum.^[12]

The relative position of the three formal types of synthetic causative and the two types of analytic causative, indicated in Table 11, supports the generalization proposed by Shibatani and Pardeshi (2002: 109–116) according to which the position of a formation on the causation continuum depends on its relative productivity. The unproductive causatives in -oys- and -oyz- are derived from costly anticausatives, while the productive type in -oycʽ- is typically derived from automatic anticausatives and predicates higher on the spontaneity scale. Furthermore, the less productive make-causative is derived from a narrower range of semantic types that the more productive give-causative. Note however that the synthetic causative is more productive than the analytic constructions, although the latter are more typical for the expression of indirect causation, contrary to Shibatani and Pardeshi’s (2002: 112) generalization on the correlation between indirect causation and productivity.

The Classical Armenian causative continuum shows a significant correspondence between morphosyntactic complexity and semantic type of the base verb. Many of the universals suggested by Haspelmath (2016) hold true, see Table 14. Thus, Universals 3, 7, 9 (Haspelmath 2016: 49) and 10 uncontroversially apply to Classical Armenian. Other universals may apply with modifications. Thus, Universal 9 (Haspelmath 2016: 46) fails to predict the derivation of the synthetic causative from agentive intransitive (“unergative”) base verbs next to the more complex give-causative derived from agentive intransitive and transitive base verbs. One of the two universals that refer to syntactic transitivity (Universal 6) also works for Classical Armenian.

However, given that different formal types of causatives, characterized by varying degrees of affectedness of the causee, can be derived from the same semantic types, the Classical Armenian facts do not support the main generalization, Universal 1, and its special case, Universal 4. Thus, for example, the expectation that if a synthetic causative is used with base verbs of agentive intransitives, analytic causatives will not be required with anticausatives lower on the spontaneity scale, is not fulfilled. Universal 8 also does not apply to Classical Armenian because it is formulated with respect to the length and not morphosyntactic complexity of causatives. As mentioned above (see Tables 12 and 13), the give-causative that is typically used with transitive base verbs is phonologically shorter than the synthetic causative.

5 Conclusions

The present paper has offered a description of the lexical distribution and grammatical features of the five expressions of causativity in Classical Armenian, including the lexical and synthetic causative and three analytic constructions with the verbs aṙnem ‘make’ and tam ‘give’. The study shows that the morphosyntactic complexity of the causative formations partly correlates with a continuum of lexico-syntactic types of predicates, with the autonomy of the first argument increasing from patientive intransitives to agentive transitives. The lexical causatives are centered on patientive and costly anticausative verbs, while the synthetic causative is productive for a wide range of intransitive types starting from automatic anticausatives and predicates higher on the spontaneity scale. The make-factitive primarily corresponds to anticausatives and is opposed to the corresponding semantic type of synthetic causatives in aspectual features. The make-causative is a transitional type, which may be a product of Greek influence on the Armenian of the Bible translation, and is best attested with agentive intransitives. The give-causative is centered on agentive transitive verbs, primarily verbs of dislocation and ingestion, but excludes canonical agentive transitive verbs.

The aforementioned form-meaning correlation is not strict, and one finds areas of overlap in the lexical distribution of the various causative formations, which can be explained by the (in)directness distinction of causation and the degree of affectedness of the causee. Given that the latter parameter proves not to be fully dependent on the lexical type of the noncausal base verb, Classical Armenian provides evidence against a typological generalization, according to which the degree of spontaneity of the noncausal meaning of a causal-noncausal pair defines the phonological length and the morphosyntactic type of the causative in languages with several causative formations.

Non-Leipzig abbreviations

act

active

aor

aorist

aor_sbj

aorist subjunctive

impf

imperfect

lab

labile

mp

mediopassive