The antipassive derivation and the lexical meaning of the verb

The antipassive derivation and the lexical meaning of the verb Descriptions of antipassive constructions in individual languages show that these constructions are often compatible with only a subset of transitive verbs. There are significant typological similarities between the sets of verbs that allow antipassivization. The following properties are typical of these verbs: i) agentive A, ii) specification of the manner component in the verb meaning, iii) lack of inherent telicity (the transitive use can be compositionally transitive, but this is cancelled under antipassivization), iv) narrow class of potential Ps, and v) affectedness of A. Verbs with all of the properties in i)-v), such as ‘eat’, constitute the core of “natural antipassives”, whereas verbs with only some of these properties are at the periphery of this class. Apart from being especially prone to enter antipassive constructions, the fuzzy class of natural antipassives is relevant for a number of phenomena. i) Polyfunctional valency-related markers or constructions tend to yield antipassive reading when applied to natural antipassives. ii) Natural antipassives tend to choose the less marked construction in languages with two antipassive constructions. iii) Lexicalization of antipassives is more likely for verbs that lack natural antipassive properties, and a typical scenario of lexicalization involves coercion of some of these properties. Ultimately, I conjecture that it is the relevance of the P-argument for the meaning of the verb which accounts for the rarity of fully productive and semantically uniform antipassive constructions in the world’s languages. 1. Introduction In principle, antipassive can be unlimitedly productive and regular: applied to any transitive verb in a given language such a valency-changing operation will yield parallel syntactic and semantic outcomes. In particular, according to the definition adopted in this volume, the P-argument of the transitive construction will be deleted or demoted to an oblique position, the A-argument of the transitive construction will be encoded as the unique core argument of the intransitive clause and the lexical meaning of the verb will not be affected. This situation is reported, for example, for the famous antipassive marked with –ŋay in Dyirbal [Dixon 1972: 65] (note, however, that antipassive in Dyirbal is exceptional in many respects). However, more often than not language-specific constructions which meet the typological definition of antipassive significantly interact with individual transitive verbs. An obvious pattern of interaction is observed if antipassive can only be applied to a subset — whether productive or unproductive1 — of transitive verbs. Lexical restrictions on antipassivization have In the sample used in [Polinsky 2013], there are 24 languages with a “productive” antipassive, 14 languages with “partially productive antipassive” and 2 languages with unproductive antipassives. This distribution can be somewhat biased towards productiveness because i) as far as I can judge, languages are allotted to types based on the most productive antipassive construction that they have and ii) “productive antipassives” are not meant to be unrestrictedly productive. 1 1 been acknowledged since the early classical literature on antipassives [Heath 1976: 211; Cooreman 1994: 60]; they are probably especially typical of accusative languages [Janic 2013: 27]. In some languages, lexical restrictions are severe. The most radical scenario is when antipassive is only possible or one verb, as in ooc (< Siouan), where the “detransitive / slot iller (wa-)” is used regularly only with the verb ‘eat’: (1) wa-haac=gįnį OBJ.3PL-eat\1E.A=already ‘I ate already.’ [ ooc example No. 430; artmann 2013] Lexical restrictions and idiosyncrasies are typically described for individual languages in the form of lists of verbs that exceptionally allow or disallow antipassivization or have semantically non-transparent antipassive counterparts. However, typological generalizations that underlie these lexical patterns are understudied. This exploratory paper is aimed at filling this gap. The data used in this study are secondary: they are taken from available descriptions of individual languages with antipassives. Based on these secondary data, in 2.1 I identi y the class o verbs that can be re erred to as “natural antipassives”; for example, ‘eat’, as in (1), clearly belongs to the core o this class, cf. [Malchukov 2015: 105]. The next step to take is to identify those semantic properties o ‘eat’ and other natural antipassives which are responsible for the specific behaviour of natural antipassives. In 2.2 I put forward a list of properties which contribute to the special behaviour of natural antipassive verbs with respect to antipassivization: i) agentivity of A, ii) specification of manner, iii) lack of inherent telicity (even if the transitive clause is compositionally telic, telicity is not inherent to the verb and is cancelled under antipassivization), iv) narrow class of potential objects, and v) affectedness of A. In 2.3, I show that although animacy distinctions of P are highly relevant for the functioning of antipassives in individual languages, neither value of this feature is a natural antipassive property. The most obvious manifestation of natural antipassives is that if an antipassive construction in a certain language is unproductive or not fully productive, then it will first of all accommodate natural antipassives; I discuss this issue in 3. In sections 4–6, I explore three other manifestations of the special status of natural antipassives vis-àvis antipassivization. i) In languages with polyfunctional mechanisms that are lexically distributed between antipassivization and other functions (e.g. reflexive, anticausative or reciprocal), natural antipassives tend to yield antipassive interpretation (section 4). ii) In languages with multiple antipassive constructions, natural antipassives tend to choose the less marked construction (section 5). iii) When antipassive derivation is applied to verbs that are far from the natural antipassive prototype, idiosyncratic lexical effects are often observed, including lexicalization and coercion of natural antipassive properties (section 6). In section 7, I summarize my findings and speculate on possible reasons for the omnipresent lexical sensitivity in antipassives. 2 2. The limits of natural antipassives 2.1. Lexical lists It is by no means a new idea that two-place predicates can be hierarchized based on the relative ease with which they enter valency-affecting operations. In an early but influential proposal, Tsunoda [1985] claimed that antipassive, as well as other transitivity-related processes (passive, reflexive and reciprocal), are more likely for predicates that are higher on the “transitivity scale”, where transitivity scale is the scale which reflects the likelihood with which a predicate selects the transitive coding frame. Tsunoda’s transitivity scale is shown in (2). (2) Direct Effect > Perception > Pursuit > Knowledge > Feeling > Relationship > Ability Tsunoda’s generalization was based on a survey of nine languages with ergative flagging. Subsequent research based on more extensive data brought modifications to the original scale [Malchukov 2005; Haspelmath 2015]. Even more importantly, it is now clear that individual valency-changing operations differ in terms of sets of verbs to which they are most easily applicable. For example, Kemmer [1993] identified “naturally reciprocal verbs”, such as ‘ iss’, ‘meet’, ‘agree’ or ‘ ight’, and showed that these verbs often involve zero-marked reciprocals (as in they met at noon) or light reciprocal markers in languages where other verbs require explicit or heavier reciprocal markers. Verbs li e ‘ iss’ or ‘agree’ are relatively low on the transitivity scale; they undergo reciprocalization with great ease simply because they denote actions which are typically mutual forms of social behaviour. Likewise, actions that are often performed by agents on themselves, such as shaving, washing or (un)dressing, correspond to “natural reflexive verbs” (the notion was also introduced by Kemmer [1993]). These verbs tend to select the least marked pattern of reflexivization. Following this line of research, I suggest to use the label “natural antipassives” for those predicates which are cross-linguistically most easily compatible with antipassivization. Lists of verbs that show the least marked pattern of antipassivization have been reported for some individual languages. For example, Fillmore [1986: 95] notes that verbs like eat, read, sing, cook, sew and bake easily allow indefinite object deletion in English, which is not equally available for many other transitive verbs. Næss [2007: 124–141] compares lists of verbs that allow indefinite object deletion in an extensive convenience sample and concludes that there are two main types of such verbs: affected-agent verbs and effected-object verbs. Affected-agent verbs are centered around verbs of ingestion — ‘eat’ and ‘drin ’ — but also include verbs such as ‘read’ or ‘learn’. Effected objects are objects that come into being as the result of the verbal action; typical effected-object verbs are ‘sew’, ‘coo ’ and ‘write’, but they also encompass sound-emission verbs such as ‘sing’, ‘shout’ or ‘whistle’. A major achievement in the cross-linguistic study on the lexical basis of valencychanging operations is the Leipzig Valency classes project (ValPal) [Hartmann et al. (eds.) 2013]. This project was based on a sample of 80 verbs (verb meanings), which 3 were tested for coding frames and compatibility with valency-changing operations in a world-wide sample of 36 languages. To date, ValPal is undoubtedly the most systematic and extensive source of information about lexical profiles of antipassives in the world’s languages. Unfortunately, ValPal does not make it possible to compare the 80 pre-selected verbs with other verbs. However, this database does make it possible to rank the 80 verbs according to their ability to enter “object-demoting/deleting” (= antipassive) derivations.2 In (3), I reproduce a variant of such hierarchy which was arrived at in [Malchukov 2015: 105–106].3 (3) ‘eat’, ‘shave’, ‘give’, ‘thin ’, ‘steal’ > ‘wash’, ‘cut’, ‘ta e’, ‘cover’, ‘wipe’, ‘see’, ‘search or’, ‘hit’, ‘throw’, ‘hear’ > ‘coo ’, ‘ now’, ‘as or’, ‘tell’ > ‘beat’, ‘tear’ > ‘pour’ > ‘ ill’, ‘climb’, ‘hug’, ‘loo at’, ‘help’, ‘name’ > ‘brea ’, ‘ ill’, ‘touch’, ‘load’, ‘teach’, ‘smell’ > ‘ ear’, ‘dress’ > ‘show’, ‘send’, ‘carry’, ‘tie’, ‘put’ > ‘sing’, ‘grind’, ‘dig’ > ‘ ollow’, ‘say’, ‘build’, ‘peel’ > ‘jump’, ‘li e’, ‘shout at’, ‘leave’, ‘live’, ‘play’, ‘meet’, ‘tal ’, ‘hide’ > ‘blin ’, ‘laugh’, ‘roll’, ‘burn’, ‘ righten’, ‘run’, ‘be dry’, ‘push’, ‘bring’ > ‘cough’, ‘sit’, ‘go’, ‘scream’, ‘ eel pain’, ‘sin ’, ‘be a hunter’, ‘boil’, ‘sit down’, ‘die’, ‘be sad’, ‘ eel cold’, ‘be hungry’, ‘rain’. According to the definition adopted above, verbs in the upper part of this hierarchy constitute the core of natural antipassives — inasmuch as one believes those ten or so language-specific alternations that served as the empirical basis for the hierarchy in (3) to be representative of antipassives in the world’s languages. Other verbs that are recurrently mentioned as allowing object deletion or demotion in individual languages are ‘read’, ‘swallow’, ‘sew’, ‘draw’, ‘plough’, ‘chew’, ‘bite’, ‘sweep’, ‘smo e’. Typically monovalent verbs (such as ‘laugh’, ‘run’, ‘cough’ or ‘die’) are expectedly found at the bottom of that hierarchy. However, there are also typically bivalent verbs which are far from the top of the hierarchy in (3), e.g. ‘ ill’, ‘brea ’, ‘touch’ and ‘hide’. As stressed by Malchukov [2015: 105], this class encompasses ‘ ill’ and ‘brea ’, the two verbs that are often thought to represent the lexical prototype of transitivity. Beyond the verb meanings from the ValPal database, verbs such as ‘hold’, ‘love’, ‘want’, ‘stir’ or ‘surprise’ are not easily compatible with antipassivization even if they are morphosyntactically transitive in individual languages. A generalization that suggests itself is that object omission or demotion is especially li ely with those verbs “which relate to an action which may be described just in general terms or, alternatively, with respect to some particular patient” [Dixon & Aikhenvald 2000: 20–21]. It is thus tempting to discern those more atomic and more palpable features that account for the somewhat abstract lexical property identified by Dixon and Aikhenvald. This is the objective of the next section. 2 The rankings discussed here do not discriminate between object-demoting and object-deleting antipassives. Arguably, sets of verbs that are most easily compatible with these two types of constructions are not identical. However, this issue is not discussed in the remainder of this paper and should be pursued elsewhere. 3 Similar hierarchies based on different subsets of data from the same ValPal database are presented in [Wichmann 2015: 166–167] and [Wichmann 2016]. 4 2.2. Properties The following semantic properties are typical of natural antipassives: - agentivity of the A-argument; - speci ication o the A’s manner; - inherent atelicity (with the possibility of compositional telicity in the verb’s transitive use); - narrow class of potential P-arguments; - affectedness of the A-argument. Natural antipassives are understood as a fuzzy category; the five properties above are not necessary or sufficient conditions, but rather they are thought of as contributing factors that determine the degree to which an individual verb is likely to behave as a natural antipassive. The five properties should be briefly discussed one after another. Agentivity is a shortcut term which re lects A’s animacy, volitionality and control. Since agentivity of the A-argument is an important component part of transitivity [Hopper & Thompson 1980], it is a typical feature of most transitive verbs, so this first property is not very restrictive. However, in many languages there are morphosyntactically transitive verbs, which select inanimate A-arguments or are necessarily or frequently used with A-arguments that lack control: ‘drop’, ‘ orget’, ‘ righten’, ‘surprise’, ‘fear’, ‘encompass’, ‘cost’, etc. Such verbs are not among natural antipassives. It has been repeatedly pointed out that A-preserving transitivity alternations are more typical o those verbs which “convey in ormation on the A’s state, purposes or mode o activity” [Kazenin 1994: 151]. Probably the most common way to capture this delimitation within the class of transitive verbs is to make a distinction between “manner verbs” and “result verbs”, as discussed in [Rappaport Hovav & Levin 1998] and many subsequent studies. Natural antipassives are typically manner verbs. Malchukov [2015: 106] asserts that manner verbs such as ‘wash’, ‘wipe’, ‘hit’ and ‘coo ’ are higher on his hierarchy for Object-demoting/deleting alternations than result verbs such as ‘ ill’, ‘brea ’, ‘ ill’ or ‘build’, see (3) above. Similar generalizations, whether in exactly these or in different terms, have been made for both individual languages (from observations on object omission in English [Rappaport Hovav & Levin 1998: 102] to antipassive in Adyghe [Arkadiev & Letuchij 2008]) and in cross-linguistic studies [Levin 2015; Polinsky 2017: 317]. A possible complication with this well-established idea is that some verbs that have a result component in their meaning easily enter antipassive constructions, most notably the verb ‘eat’ [Malchu ov 2015: 106]. A li ely explanation is that verbs like ‘eat’ can be used to lexicalize both the manner and the result components, but antipassive variants of these verbs highlight the former components. By contrast, causative verbs like ‘brea ’ or ‘ ill’ specify a result in the state of the P-argument without conveying any in ormation on A’s manner o carrying out the action; they are not so easily 5 compatible with antipassives.4 Under this refinement, the second natural antipassive property is the presence of the manner component in the verb’s meaning, even if this does not rule out the possibility that there are result components as well. It has been pointed out that atelic verbs are “more permissive” to the antipassive alternations than telic verbs (see [Wichmann 2015: 167] and further references). At the same time, in many languages antipassive constructions are atelic, but transitive constructions with the same verbs are (see [Tatevosov 2011: 135–137] for an overview). In fact, there is no contradiction here. The verbs that enter antipassive derivation most easily are those verbs whose telicity in the transitive use is compositional: it is determined by the fact that their incremental P-argument is quantized (as in eat an apple). In the antipassive construction, telicity can be cancelled and the construction is interpreted as an actitivity (as in eat apples). This lexical property is consonant with the generalization that “[i]f an antipassive construction can have a perfective (telic) interpretation, it must also have an imperfective (non-telic) interpretation” [Polinsky 2017: 316]. By contrast, verbs that are inherently telic (achievements such as ‘ ill’) and stative verbs are not among natural antipassives. The three natural antipassive properties discussed so far are interrelated. The fourth property is orthogonal to them: other things being equal, antipassive alternations are more readily available for those verbs that semantically select objects from a relatively narrow class, i.e. verbs li e ‘eat’, ‘drin ’, ‘hunt’, ‘ ish’, ‘smo e’, ‘read’ or ‘ba e’ (the fullest realization of this property is found in cognate object verbs). Potential objects o ‘eat’ belong to the class o entities which can be described as food; likewise, hunting and fishing necessarily involves animals and fish as objects, etc. Although it is possible to point out the particular type of food being eaten, or animals being hunted, the nature of the object involved is often sufficiently predictable from the very verb to make its explicit mentioning unnecessary. These verbs contrast with those transitive verbs that can combine with many different types of objects (e.g. ‘see’, ‘li e’, ‘brea ’ etc.). At the extreme pole o this continuum are transitive verbs with fairly abstract lexical meaning, such as ‘ma e’: these verbs do not easily enter (absolutive) antipassive construction. An example of this is found in Adyghe, where many transitive effected-object verbs have stem ending in -ə but alternate with antipassive verbs with stems ending in –e, cf. txə-n / txe-n ‘write’, də-n / de-n ‘sew’ [Arkadiev & Letuchij 2008]. However, the antipassive counterpart is not possible for the transitive structure in (4): (4) se wEne se-I&E I house (3SG.P)1SG.A-make ‘I am building a house’ [Ar adiev & Letuchij 2008]. Arkadiev and Letuchij explain that although I&E-n is interpreted as ‘build’ in (4), this verb actually has a airly abstract meaning ‘do, ma e’; without an overt P the This idea is central to Basilico’s account o antipassives in Inuit, where consumption verbs can be used in antipassive constructions without morphological marking, whereas result verbs need a dedicated antipassive suffix –si [Basilico 2012: 87–91]. 6 4 nature of the object cannot be inferred from the verb, which makes the whole sentence hardly interpretable. The final natural antipassive property is affected A-argument. The most common type o A’s a ectedness is attested with ingestive verbs. Here, the Aargument is not only a causer but also the endpoint whose state significantly changes as a result of the action. Similarly, the A-argument is affected in those mental verbs where it can be conceptualized as an endpoint, c . ‘hear’, ‘understand’, ‘learn’, ‘read’, ‘see’, etc. Verbs with affected A-argument were identified as a class by Masica, who also concluded that this “group might be regarded as occupying a halfway station between intransitives and transitives, since the object in question can frequently be dispensed with in favor of concentration on the activity as such” [1976: 48]. The rationale behind the five natural antipassive properties identified in this section is identical: the presence of each property contributes to the likelihood of a situation in which it is possible or desirable to downgrade the P-argument and to highlight A’s activity — and this is exactly the main semantic / pragmatic function of antipassives. For example, all other things being equal, predictable P’s can be bac grounded with greater ease than those P’s whose nature cannot be in erred rom the meaning of the verb (cf. the fourth property above). The fuzzy class of natural antipassives as delimited in this section finds empirical support in various morphosyntactic phenomena discussed in sections 3–6. However, before proceeding to those phenomena, it is germane to discuss, and ultimately discard, one more potential natural antipassive property, viz. inanimacy of the P-argument. 2.3. Inanimacy of the P-argument It is sometimes assumed that demoted or deleted P’s in antipassive constructions are typically inanimate, cf. remarks in [Fleck 2006: 566; Heaton 2017: 15]. This assumption seems plausible, given that the main function of antipassives crosslinguistically is to background the P-argument and that inanimate NPs are more likely to be backgrounded. And indeed, there are some languages where P-arguments in antipassive constructions are obligatorily inanimate (e.g. some types of antipassive uses in Russian [Say 2005, Janic 2013: 150–156]) or at least non-human (e.g. the unspecified object prefix ta- in Pipil (< Uto-Aztecan) [Campbell 1985: 77]). Such patterns create lexical restrictions: those transitive verbs whose meanings entail animate P-arguments (‘punish’, ‘ ill’, ‘teach’, ‘dress’) are not available for antipassivization in these constructions. However, from a broad typological perspective there seems to be no justification for the claim that verbs implying inanimate / non-human objects are closer to the core of natural antipassives than other transitive verbs. Rather, in many languages antipassive constructions are sensitive, in one way or another, to whether the demoted / deleted P-argument is [±human] and [±animate]. Antipassives which are only possible with inanimate objects is just one of several scenarios which reflect this sensitivity. 7 Another possible scenario is attested in languages where the only antipassive construction requires the demoted or deleted P-argument to be animate; this is the situation in Matsés (< Pano-Tacanan) [Fleck 2006] (see also section 5). However, a more widespread scenario is the one in which a language possesses two or more antipassive constructions and they are distributed depending on the [±animacy] of the deleted or demoted P-argument. Pustet & Rood [2008: 342–345] report such a situation for Lakota (< Siouan). Lakota has three markers that signal suppression of the P-argument: wichá- suppreses exclusively animate objects, takúsuppresses exclusively inanimate objects, and wa- suppresses predominantly inanimate objects. It is not entirely clear whether these markers should be viewed as antipassives markers: an alternative interpretation is that these markers actually fill the object slot, rather than eliminate it [ibid. 342]. The same dilemma arises in many other head-marking languages, e.g. in Nahuatl, cf. discussion in [Janic 2013: 26–27; Heaton 2017: 186–187; Malchukov 2015: 98–99] and references therein. An interesting distribution is observed in Paraguayan Guaraní. ere, the mar er mba’e-, which is used in contexts with implied inanimate objects, actually means ‘thing’; its use in contexts like (5) can thus be interpreted as noun incorporation rather than antipassive. However, the prefix poro-, which is used in contexts with intended human interpretation, as in (6), is not synchronically a nominal root and should be analyzed as a dedicated antipassive marker. (5) a-mba’e-jogua 1SG.ACT-ANTIP2-buy ‘I am shopping’ [Estigarribia 2017: 50]. (6) mbo’e-hára o-poro-mbo’e teach-NMLZ.AG 3.ACT-ANTIP1-teach ‘The teacher teaches people’ [Estigarribia 2017: 49]. In her discussion of languages with multiple antipassives, Heaton [2017: 265– 285] shows that animacy is among the most frequently attested factors determining the distribution of competing antipassives. Judging from her map on page 282, this possibility is particularly widespread in Americas, which can be related to headmarking. Finally, in some languages, the choice of the antipassive marker does not depend on the [±animacy] of the P-argument, but the case-marking of the demoted P does depend on its position on the animacy scale. This situation is observed in some PamaNyungan languages, e.g. Yidiny [Dixon 1977: 27] or Warungu [Tsunoda 2011: 464]. All these data indicate that animacy of the demoted / deleted P is an important factor which can determine the form of the antipassive construction or even its availability. However, there is no robust cross-linguistic preference for antipassives to be possible with inanimate P’s only (and, accordingly, with only the verbs that are normally used with inanimate objects). 8 3. Natural antipassives and dedicated antipassive constructions The most evident manifestation of the special status of natural antipassives with respect to antipassivization is the scenario in which an antipassive construction is only possible for a subset of transitive verbs. This is exactly the scenario which was generalized in the form of Malchu ov’s [2015: 105–106] “hierarchy o Objectdemoting/deleting alternations”, see (3) and the discussion in section 2.1 above. The hypothesis advocated here predicts that verbs with more natural antipassive properties identified in section 2.2 are more likely to enter this subset than other transitive verbs. In other words, the properties of natural antipassives are expected to be recurrent across languages, although the cut-off points that determine the productivity of antipassives in individual languages can be of course very different. The extreme scenario predicted from the hierarchy was mentioned above for ooc , see (1): here, just one verb can be used in the antipassive construction and this is ‘eat’, i.e. one of the few verbs that simultaneously have all the five natural antipassive properties from section 2.2. A somewhat different but similar situation is observed in Icari Dargwa (< Nakh-Daghestanian). Here, the antipassive construction does not involve any special verbal marking, but the contrast between transitive and antipassive constructions is reflected by both flagging (demoted P is marked with the ergative case) and indexing (in both constructions agreement prefix signals the gender of the absolutive argument). With most verbs, antipassive is limited to several TAM forms with resultative/stative meaning [Sumbatova, Mutalov 2003: 157]. There seem to be only two verbs which allow antipassivization in all their TAM forms, viz. ‘eat’ and ‘drin ’, the two verbs which are in the core of natural antipassives. The relevant examples for the transitive and antipassive uses o ‘eat’ in the (present) progressive form are in (7) and (8). (7) du-l t’ult’ b=uk-a-ta I-ERG bread [N]eat:IPF-PRG-1 ‘I am eating bread’. (8) du t’ult’-il uk-a-ta I bread-ERG (M)eat:IPF-PRG-1 ‘I am eating bread’. [Sumbatova, Mutalov 2003: 157] Other languages may allow antipassivization for a much wider set of verbs, but still have some limitations. For example, Campbell cites 25 verbs that can take the “unspeci ied object” pre ix ta- in Pipil; in more than a half of these verbs, the lexical meaning of the verb presupposes an object from a relatively narrow class and denotes a particular manner of acting upon that object: ‘smo e’, ‘spit’, ‘write’, ‘toast’, ‘sew’, ‘scrape’, ‘salt’, ‘roast’, ‘eat’, ‘coo ’, ‘weed’, ‘replant’, ‘grind (corn)’, ‘pic , harvest’, ‘shell (corn)’ [Campbell 1985: 77–78]. Among the five natural antipassive properties identified in 2.2, the one which has received most attention in the literature is specification of manner (as opposed to specification of result). A convincing example which shows the relevance of this property for antipassivization comes from Godoberi (< Nakh-Daghestanian), as 9 analyzed by Tatevosov [2011]. In Godoberi, antipassivization is morphologically realized with a special converb (“A-converb”). Tatevosov argues antipassivization in Godoberi is tightly interrelated with aspectual modification, namely detelicization. This analysis rests, among other things, upon lexical extent of A-converbs. These converbs are only possible for some 60 verbs; among transitive verbs, these are mainly manner verbs that get a telic interpretation if their direct object is a quantized incremental theme: ‘plough’, ‘thresh’, ‘chew’, mill’, ‘ma e’, ‘plane (wood)’, ‘lic ’, ‘gnaw’, ‘saw’, ‘si t’, ‘sni ’, ‘whistle’, ‘whisper’, ‘mow’, ‘drin ’, ‘dig’. With these transitive verbs, derivation of the A-converb does a double job: it detelicizes the verb and eliminates the P-argument, as shown by the contrast between the transitive use in (9) and the antipassive construction involving an A-converb in (10): (9) ʕali-di q'iru b-el-ata-da Ali-ERG wheat N-thresh-IPFV.CONV-AUX ‘Ali is threshing wheat.’ (10) ʕali w-ol-a-da Ali M-thresh-A.CONV-AUX ‘Ali is threshing.’ [Tatevosov 2011: 139] Importantly, A-converbs in Godoberi are not possible for typical result verbs, such as ‘open’, ‘brea ’ or ‘tear’. Thus, among the ive properties identi ied in 2.2, Godoberi seems to rely on the aspectual property and specification of manner. Lexically limited antipassivization is the most straightforward but ultimately just one of the several possible scenarios in which natural antipassives can manifest themselves. Slightly more complicated scenarios are discussed in sections 4–6. 4. Natural antipassives and syncretic antipassive markers Although dedicated antipassive markers / constructions which do not have other functions are attested in the languages of the world, it is a more frequent situation when the same affix or construction has other functions as well [Polinsky 2017: 314]. Very often, these other functions are also related to valency and two (or more) unctions are patterned across the verbal lexicon (c . the notion o “verb sensitivity”). The hypothesis advocated here predicts that, all other things being equal, verbs with more natural antipassive properties are more likely to get the antipassive interpretation with syncretic markers and constructions than verbs with fewer such properties. Kazenin [1994] claims that lexical distributions o “A-preserving” transitivity alternations, including antipassive, and “O-preserving” transitivity alternations are cross-linguistically motivated by similar lexical semantic factors. He stresses that these distributions work similarly regardless of the direction of derivation: valencydecreasing (e.g. antipassive vs. anticausative), valency-increasing (e.g. causative vs. applicative) or non-directed (two types of lability). The factor to which Kazenin attributed the observed distribution is the contrast between “agent-oriented” vs. “patient-oriented” verbs. owever, Kazenin’s study was only based on comparing three unrelated languages (Asiatic Eskimo, Boumaa Fijian and Bambara). More recent 10 studies do corroborate Kazenin’s generalization, but they also make it clear that the determining factor is not a strict dichotomy, but rather a hierarchy which determines the likelihood of getting an interpretation of a particular kind [Malchukov 2015]. A common type of syncretism involving antipassive-like constructions is the situation when some — but not all — transitive verbs allow objectless uses, that is, are A-labile, whereas other transitive verbs manifest other types of lability, including Plability, reflexive and reciprocal lability, etc. English is arguably the best-studied language of this kind. English transitive verbs yield different interpretations when used in a monovalent structure, viz. anticausative (the stick broke), the so-called “middle” (stale bread slices easily), reflexive (she dressed), reciprocal (they kissed) etc. Some verbs allow morphologically unmarked alternations in which the subject of the intransitive use corresponds to the subject of the transitive use while the object remains unspecified, as in she is drawing. These are instances of what Creissels [2014] terms “wea lability”: a situation where the constructions with two vs. one core arguments do not show any formal difference except for the presence vs. absence of the second NP. It is not entirely clear whether objectless uses of weak A-labile verbs in English (and elsewhere) should be interpreted as antipassive constructions; an alternative interpretation is that they are transitive constructions where the object slot is filled with a null NP. Whatever the theoretical interpretation, the lexical distribution of labile constructions in English is largely predictable from the lexical meaning of the verbs involved ([Levin 1993] remains a groundbreaking classic in this domain, and more recent references are countless). The unspecified object alternation is mainly possible for agentive verbs which lexicalize the manner of activity and have a typical object which is implied by default (bake, cook, draw, drink, eat, paint, read, sew, sweep, teach, type, etc.) [Levin 1993: 33] and thus fit the hypothesis advocated here. A situation which can be more justifiably described as a syncretism between antipassive construction and another valency-changing construction is observed languages with “strong lability”5 in which A-lability coexists with other types lability. This is the situation in some varieties of Asiatic Eskimo, which will discussed in section 5. an in of be The remainder of this section is devoted to languages where antipassives are signalled by an overt marker which can elsewhere express a different valency-related alternation. Reflexive is the function which is typologically especially prone to show syncretism with the antipassive (see [Say 2008: 135–138; Janic 2013: 238–250; Sansò 2017: 193–197] for overviews). Other syncretism patterns including the passive/anticausative-antipassive pattern (this scenario often involves markers which simultaneously have other unctions in the “middle” domain, but this is not obligatory) and the reciprocal-antipassive pattern, attested in e.g. some Turkic (see [Janic 2013: 110–129] for an overview) and Austronesian languages (see a detailed analysis of the “Strong A-lability” (again, in Creissels’ [2014] terms) is a pattern in which the A-argument of the full-fledged transitive construction can also be used as the sole core argument with the same verb but the two constructions have some formal differences apart from the mere presence vs. absence of the P-argument (e.g. the ergative vs. absolutive flagging of the preserved argument). 11 5 acts rom To’abaita in [Lichtenberk 2008: 864] and a wider Austronesian perspective in [Lichtenberk 1999: 42–44]). Languages with syncretic markers differ in the lexical extent of those verbs that get the antipassive interpretation with these markers. At one pole of the continuum are languages where antipassivization is a relatively marginal function of a syncretic marker. This situation is typical o “re lexive verbs” in many European languages, e.g. Slavic and Romance. However, similar phenomena are observed elsewhere, e.g. in Australia, where antipassive suffixes are often identical or cognate with reflexive suffixes (see [Terrill 1997: 78] for an overview).6 A good example of verb sensitivity in Australia comes from Diyari. Here, verbal affix -tharri, which is in many respects a typical Pama-Nyungan syncretic reflexive marker, covers no less than five different functions: i) reflexive, ii) antipassive (with a demoted object), iii) a function which is labelled “antipassive” by Austin, but does not meet the de inition o antipassive adopted in this volume: the A-argument from the transitive verb is converted to Sposition, but the original P retains its absolutive encoding instead of being demoted or deleted, iv) passive, v) durative [Austin 2013: 157–163]. The second function is illustrated in (12), which is contrasted with the transitive construction (11). (11) ngathu 1SG.ERG nhanha 3SG.F.ACC wilha karlka-yi woman.ACC wait.for-PRES (12) nganhi karlka-tharri-yi nhangkangu 1SG.NOM wait.for-ANTI-PRES 3SG.F.LOC ‘I wait or the woman’ wilha-nhi woman-LOC Diyari transitive verbs can be divided into five mutually exclusive groups based on the function of –tharri [Austin 2013: 77]. The object-demoting antipassive is available for just eight transitive verbs (out of a total of 114 verbs analyzed by Austin). These eight verbs constitute a semantically compact group which covers meanings such as ‘ ind/discover’, ‘await’, ‘ ollow’, ‘ta e away rom’, ‘search or’ [Austin 2013: 81]; unlike most other transitive verbs, these verbs entail an affected A-argument (and speci y A’s manner o action rather than any direct result in the state of the Pargument), which echoes the natural antipassive properties identified in 2.2.7 A more balanced pattern of syncretism is found in Koyraboro Senni (< Songhay). Heath [1999: 166–167] identifies two homophonous suffixes –a in Koyraboro Senni, which have the “mediopassive” and depatientive (= antipassive) values correspondingly. Even though these markers are analyzed as homophonous, they clearly tend to be distributed lexically, which makes it possible to reinterpret –a as a Than s to Dixon’s pathbrea ing description of Dyirbal [Dixon 1972: 90], non-reflexive unctions o mar ers that can also signal re lexivity proper are o ten re erred to as “ alse re lexives”. 7 Another cluster of verbs with salient natural antipassive properties is formed by verbs with meanings such as ‘eat’, ‘drin ’, ‘coo ’, ‘try, test’, ‘sing’. The e ect o –tharri on these five verbs is also described as antipassive in [Austin 2013], but fails to meet the definition of the antipassive adopted here, because the P-argument in these constructions is not demoted. Whatever the appropriate label, verbs with the most natural antipassive properties stand apart from the majority of transitive verbs, which yield either a reflexive or a passive interpretation when marked with –tharri. 12 6 syncretic suffix with two distinct functions. Heath does not provide full lists of verbs yielding the two interpretations, but the examples he cites are suggestive. The mediopassive function is illustrated with verbs like dumbu ‘cut’ (c . dumb-a ‘be cut’), duubu ‘shut’, haw ‘tie, bind’, hanji ‘hang’, keyri ‘brea ’; the depatientive unction is illustrated with verbs feferi ‘peel’ (c . fefer-a ‘do some peeling’), neeši ‘measure’, kar ‘hit’, haabu ‘sweep’ [ibid.: 166–167]. These lists imply that the presence of the manner component in the verb’s meaning is a prerequisite for yielding the antipassive interpretation under a-derivation, but having a result component is not an obstacle to having this interpretation. As a result, many transitive verbs have a-derivatives which can have both a mediopassive and an unspecified-object interpretation (an example given is čin-a ‘be built’ or ‘do some building’). Finally, there are many languages where the pattern of syncretism is biased towards the antipassive function. In these languages, antipassive construction is productive, but with a minority of transitive verbs it is inapplicable or yields a different interpretation. The hypothesis advocated here predicts that this minority shouldn’t contain verbs with many natural antipassive properties; deviant behaviour is expected for highly transitive result verbs, as well as for natural reflexives and natural reciprocals. Exactly this scenario is observed in a number of languages. In Bezhta (< NakhDaghestanian), object-demoting antipassivization marked with the suffix -dah/-lah/rah is possible for some 60–70% of transitive verbs [Kibrik, Testelec 2004: 274]. However, there are several verbs that yield a reflexive interpretation when marked with the same suffix. Examples cited by Kibrik & Testelec include three natural re lexive verbs (‘wash’, ‘comb’, ‘shave’) and two verbs that speci y result rather than manner (‘hide’, ‘warm’) [2004: 285]. An even more biased pattern o syncretism is observed in Tz’utujil. Similarly to what is reported for many other Mayan languages, Dayley [1985] identifies two “antipassive” constructions in Tz’utujil. One o these is the “absolutive antipassive” marked with –oon, which fully meets the definition of the antipassive adopted here. The absolutive antipassive is productive; however, several verbs deviate from the regular pattern. These deviations include a) the verb ‘wash’ (a natural re lexive), which yields a reflexive interpretation when marked with -oon; b) a few verbs that can’t be used with this su ix at all; this potentiality is illustrated with elasaxik ‘ta e out’ — clearly a result verb; and c) a few verbs which are ambiguous between the regular antipassive interpretation and an anticausative interpretation; the latter group is illustrated with typical result verbs, such as ‘brea ’, ‘burst’, ‘spill/splash’, ‘spill over’. All of the examples surveyed corroborate the idea that in cases when syncretic affixes are verb-sensitive different valency-related alternations have their “centres of gravity”: semantic zones which attract specific valency-related interpretations. Natural antipassives occupy their niche in such systems, alongside with result verbs, which are most easily susceptible to the anticausative derivation [Haspelmath 1993], natural reflexives and natural reciprocals. 13 5. Languages with multiple antipassive constructions Many languages have more than one construction which meets the definition of the antipassive construction. In eaton’s world-wide sample of 126 languages with antipassives, some 30 languages — almost every fourth language in the sample — have multiple antipassives [Heaton 2017: 266]. This quantitative estimate is to some extent subjective, because it depends on the sample used as well as on definitional assumptions, but it clearly shows that having multiple antipassives is not a rare phenomenon. Some languages mentioned in sections 1–4 do have multiple antipassives, but up to this point the competition between different constructions was not in the focus. Languages with multiple antipassives vary in terms of the main factor which determines their distribution. In particular, Heaton identifies three major types: lexical differences, aspectual/modal differences, and patient-related differences [Heaton 2017: 270]. The latter pattern was mentioned in 2.3, where I briefly discussed languages where the choice of the antipassive marker depends on the animacy of the P-argument. Here, I will concentrate on languages where the distribution is primarily lexical. Especially relevant for the discussion of natural antipassive properties are languages where the two antipassive constructions differ in terms of markedness. A clearer scenario of this kind is observed in languages where a syncretic marker, which yields the antipassive interpretation with a subset of transitive verbs, co-exists with a dedicated antipassive marker, which is typically more productive. I will discuss this pattern using data from Soninke (< Mande). A slightly different scenario is found in languages where an antipassive construction with a dedicated marker coexists with a morphologically unmarked antipassive construction, which can thus be analyzed in terms of A-lability. This pattern will be discussed for some Eskimo varieties and for Matsés. In both scenarios, the construction which involves a dedicated marker is more marked than the construction with the more bleached marker or no marker at all. The hypothesis which I put forward in this study predicts that if the two lexically distributed antipassive constructions differ in terms of markedness then the less marked construction will be available for verbs which have more natural antipassive properties, whereas the more marked construction will be used with verbs that have fewer natural antipassive properties. Soninke is a language where a syncretic antipassive marker coexists with a dedicated antipassive marker. The most frequent function of the former marker, detransitivizing suffix –i, is agent demotion, but it also has some other uses related to the middle domain. With a few verbs, it functions as an antipassive marker, cf. the transitive construction in (13) with its antipassive counterpart in (14): (13) Yàxàré-n dà máarò-n woman-DEF TR rice-DEF ‘The woman coo ed the rice.’ còró cook 14 (14) Yàxàré-n còré8 woman-DEF cook.DETR ‘The woman did the coo ing.’ [Creissels 2012]. Creissels [1991] analyzes lexical distributions of the various functions of -i in Soninke. In total, there are 35 verbs marked with –i which are discussed in this study. Out of these 35 verbs, there are four verbs which can be used in objectless (antipassive) constructions, viz. sòró ‘coo ’, as in (13)–(14), and also jígá ‘eat’, gòró ‘pound, grind’ and sòχó ‘plough’ (this verb also has the meaning ‘to close, shut’, but in this meaning, the antipassive interpretation of the –i-derivative is not possible); some of these –i-derivatives can also have non-antipassive interpretations. It is not clear whether the list of verbs provided in [Creissels 1991: 13-20] is exhaustive; at any event, it follows from his discussion that the antipassive derivation is lexically very limited and that it is possible with those transitive verbs which can be interpreted as activities, that is, closely correspond to the core of natural antipassives as identified in 2.2.9 The unavailability of the antipassive interpretation for –i-derivatives from other transitive verbs, including verbs that are far from the natural antipassive prototype, is in many cases compensated for by the dedicated antipassive suffix –ndì, which is relatively productive.10 In section 4, I mentioned that many languages allow several types of lability, which are distributed over their transitive verbal lexicon, so that the objectless use is only available for a subset of transitive verbs. The Chaplino variety of Eskimo, one of the Central Siberian Yupik idioms, has a system of this kind. The data on this idiom, especially the lexical lists, were taken from [Vakhtin 1981] and especially [Yemel’yanova 1982]; their analysis is inspired by the discussion in [Kazenin 1994: 145–147].11 In Chaplino Eskimo, some transitive verbs can be used in a monovalent construction and get the antipassive interpretation. There is no special marker in either construction, but the alternation manifests itself through the change in both flagging and indexing of arguments (this is thus a case o “strong A-lability” in Creissels’ [2014] terms), cf. the transitive construction in (15) and its intransitive counterpart with the demoted or deleted object in (16). (15) agna-m ukini-ᶄa atkuja-ᶄ woman-ERG sew-3SG.A:3SG.O dress-ABS ‘The woman is sewing the dress.’ 8 The form còré results from the fusion of the suffix –i with the root còró ‘coo (tr.)’, as in (13). 9 Some verbs in Soninke are simply A-labile verbs, that is, allow object deletion without any overt morphological marking [Creissels 2012: 6]. This possibility will not be analyzed here any further. 10 Judging from a cursory discussion in [Quesada 2007: 174–176], a similar distribution is observed in Guatuso (< Chibchan). Unfortunately, there is only one sentential example cited in this source in which a syncretic (reflexive, reciprocal etc.) marker fulfills the antipassive function; however, this example involves the verb ‘to eat’, which the patterns attested elsewhere. 11 Distributions which are similar to those discussed here for the Chaplino Yupik are also observed in other Eskimo languages, see [Say 2008: 236–237] for some references. 15 (16) agna-ᶄ ukini-ᶄuᶄ (atkujagmyn) woman-ABS sew-3SG dress.INST ‘The woman is sewing (a dress).’ [Kazenin 1994: 146] Most verbs that can be used transitively participate in the alternation which is structurally identical to the one illustrated in (15)–(16). However, some of these labile verbs get a reflexive or an anticausative interpretation in the intransitive construction [Vakhtin 1981]. Simplifying somewhat12, these possibilities yield a three-way classification: A-labile, P-labile and reflexive-labile verbs. Apart from this, many verbs which get a reflexive or an anticausative interpretation in their intransitive use participate in an alternation which is signalled by the vowel change (a => i), in which the verb ending in –i is obligatorily interpreted as the antipassive counterpart of the transitive verb in –a [Yemel’yanova 1982: 23–24]. In other words, this system simultaneously has i) a syncretic construction which covers the antipassive function along with other detransitivizing functions, the pattern discussed in section 4, and ii) shows a competition between two antipassive constructions, which is the focus of this section. Lexical distributions underlying the processes in i) and ii) are largely consonant and clearly have the familiar semantic rationale: antipassive alternation is morphologically unmarked for verbs that accumulate more natural antipassive properties, whereas the marked alternation is mostly found with verbs that have fewer natural antipassive properties. Some examples of A-labile verbs are shown in (17). (17) a. verbs o ingestion: ‘eat’, ‘drin ’, ‘swallow’; b. verbs o acting with one’s (usually speci ied) body-part: ‘bite (e.g. one’s tongue)’, ‘butt’, ‘chew’, ‘ ic ’, ‘lic ’, ‘pic (e.g. one’s teeth)’, ‘pinch’, ‘press upon something’, ‘touch’, ‘trample’; c. verbs of professional activity and/or activity with a (specified) instrument: ‘dig’, ‘drill’, ‘hone’, ‘knead’, ‘scrape’, ‘sew’, ‘shoot’, ‘sweep’, ‘throw a spear at somebody’, ‘weave’, ‘write’; d. verbs with a mentally involved or potentially a ected A: ‘as ’, ‘choose’, ‘ orget’, ‘look at’, ‘read’, ‘search or’, ‘steal’, ‘ta e pity on somebody’, ‘thank’; e. speech act verbs: ‘accuse’, ‘advise’, ‘call (many times)’, ‘explain’, ‘(make an) order’, ‘order, command’, ‘sing’, ‘shout’; f. other verbs: ‘do something in a hurry’, ‘play a jo e upon somebody’, ‘support’ ‘teach’. The verbs in (17) are classified into several conventional groups for the sake of readability. It can be seen that most verbs in (17) have A-oriented semantic components, that is, speci y A’s purposes, manner o activity, affectedness by the action, etc. Apart from that, verbs from groups (a), (b) and (c) semantically select objects from relatively narrow classes. Thus, as a group, these verbs show several natural antipassive properties identified in section 2.2 (except for probably the 12 For the sake of simplicity, I disregard verbs whose intransitive uses allow multiple interpretations as well as those verbs that can only be used in the transitive construction. 16 aspectual characteristics: these are hard to judge from the lexical lists in [Yemel’yanova 1982] and other sources). Although verbs in (17) are heterogeneous, they contrast with the set of Chaplino Eskimo verbs which yield anticausative interpretation in the intransitive use.13 Some examples are shown in (18). (18) ‘bring sb. in the middle’, ‘close’, ‘change’, ‘cool down’, ‘ inish’, ‘leave’, ‘make sb. tired’, ‘make sth. higher / grow (tr.)’, ‘make sth. red’, ‘make sth. smaller’, ‘ma e sth. visible’, ‘ma e sth. wet’, ‘move sth. aside’, ‘make tremble’, ‘prolong’, ‘take sth. up’, ‘turn’, ‘wea en’. The verbs in (18) are mostly causative change-of-state verbs; they do not specify A’s manner in any tangible way and only weakly specify the nature of the object involved. The contrast between A-labile and P-labile verbs is most clearly seen if one compares pairs or small sets of verbs which are thematically close to each other. In such contrastive groups A-labile verbs invariably have more A-oriented components or imply A’s involvement to a higher extent than P-labile verbs, cf. the following contrasts: inaxma ‘put’, ifkaǥa ‘drop’ (P-labile) vs. aŋaᶍᶄuǥa ‘drag (and drop)’ (Alabile); iflykaǥa ‘brea ’ (P-labile) vs. rypaᶄwa ‘crac (with a hammer)’, ᶄas’aᶄt ‘hit (with the palm)’ (A-labile); jakuǥŋīᶄa ‘ righten’ (P-labile) vs. akyᶄs’axt ‘threaten’, akinig ‘object’, aǥiva ‘tease’. It is important to remind that most P-labile verbs also have an intransitive counterpart marked by a vowel alternation, which functions as an antipassive verb. In other words, the lexical contrast between two types of lability in Chaplino Es imo is similar to the contrast between the less mar ed (“strong labile”) and the more marked pattern of antipassivization. An essentially similar contrast between two patterns of P-demotion is observed in Matsés (< Panoan), even though at first sight this language deviates from the typological expectations. In his analysis of antipassives in Matsés, Fleck [2006] stresses that the sets of transitive verbs that can and cannot be antipassivized with the dedicated suffix -an are typologically unusual. In particular, antipassives with the indefinite object reading are available exclusively for the verbs which specify human P’s and, more unexpectedly, “entail that Patient be significantly affected by the action, either physically or emotionally” [Flec 2006: 565]. Examples include verbs such as ‘ ill’, ‘topple (wrestling)’, ‘pierce, sting, stri e’, whereas verbs with less affected P’s, which typically convey more in ormation about A’s purposes or manner of activity, do not enter the antipassive alternation. However, Fleck proposes an elegant explanation for this seemingly paradoxical lexical distribution. He notes that transitive verbs whose P-arguments are typically unimportant can simply be used with an empty object slot without any verbal marking and yield the indefinite-P interpretation. Such uses can be interpreted in terms of weak A-lability. However, verbs such as ‘ ill’ normally ta e salient P’s and highlight P’s 13 For the sake of simplicity, I do not discuss verbs that get the reflexive interpretation in the monovalent constructions; many of these verbs belong to the class of natural reflexives. 17 change of state. With these verbs, the situation in which the speaker concentrates on A’s activity is unusual and requires explicit antipassive marking, as in (18). (18) kuessunne-an-onda-bi kill-ANTI-DIST.PAST-1SG ‘I used to ill.’[Fleck 2006: 565] Thus, although the set of verbs that can be marked with the antipassive suffix and yield an indefinite-object interpretation in Matsés is very different from the class of natural antipassive verbs as identified in 2.2, the rationale behind this set actually corroborates the hypothesis advocated here. In a nutshell, Matsés provides two options which can be used to delete an indefinite object, and marking the verb with a dedicated suffix is the more marked option of the two; hence, it is chosen by verbs that are further away from the natural antipassive prototype. 6. Lexicalization of antipassives 6.1. Lexical effects of antipassivization The definition of the antipassive adopted in this volume stipulates that verb’s lexical meaning should be identical in the transitive construction and its antipassive counterpart. However, an alternation that meets all the definitional criteria of the antipassive with some verbs can affect the lexical meaning of some other verbs. Typically, such semantic shifts are idiosyncratic, i.e. they are not fully predictable and are observed in some individual verbs. These verbs, which do not fully meet the definition of the antipassive adopted here, are often referred to as “lexicalized antipassives”. Even though “lexicalized antipassives” are irregular within their language systems, they show some cross-linguistically recurrent patterns. Antipassive’s lexicalization patterns of antipassives make it possible to put forward two generalizations that make recourse to natural antipassive properties. First, lexicalization is more likely with verbs that lack some of the natural antipassive properties. In other words, lexicalized antipassives are often found at the periphery of the set of verbs that can combine with antipassive markers in individual languages. Second, a typical lexicalization effect involves coercion of natural properties; this is observed if a certain transitive verb does not have, and its lexicalized antipassive counterpart does have, a certain natural antipassive property. Coercion is very common in the domain of aspect: it is a well-known fact that antipassivization often has aspectual effects. In particular, antipassivization often correlates with non-punctual, incomplete, habitual, iterative, conative and other atelic interpretations [Cooreman 1994: 57–58; Polinsky 2017: 315–316]. In some cases, antipassivization clearly affects the lexical meaning of the verb involved; for example, in many unrelated languages antipassive counterparts of verbs like ‘hit’, ‘beat’ have meanings li e ‘ ight’, ‘struggle’, ‘compete’ etc. [Cooreman 1994: 58]. In these cases, the transitive verb and its counterpart are two different lexical entries. However, there 18 are less straightforward cases, where it is not entirely clear whether the aspectual contrast should be interpreted as lexical (transitive verbs and its lexicalized antipassive counterpart are two distinct verbs) or grammatical. The latter possibility cannot be ruled out on a priori grounds, because different forms of a verb often differ in terms of their aspectual properties. This dilemma can only be solved based on detailed analysis of lexical and grammatical aspect in individual languages, which is often missing in available descriptions of antipassives. As a consequence, coercion of lexical aspectual properties in antipassive alternations, although potentially a recurrent pattern, will not be discussed in the remainder of this section. Instead, I will address coercion of two other properties: agentivity (6.2) and narrow class of P-arguments (6.3). 6.2. Coerced agentivity Typically, antipassives are used to foreground the activity of an A-argument which is a full-fledged agent. In this respect, it is somewhat surprising that two experiential predicates, namely ‘see’ and ‘hear’, are found rather high in Malchu ov’s [2015: 105–106] antipassivizability hierarchy, see (3) above. A possible explanation is that Malchu ov’s hierarchy, based on data rom ValPal, re lects the verb’s ability to enter object-demoting alternations and does not take into account idiosyncratic shifts in meanings. In fact, the verbs ‘see’ and ‘hear’ are o ten coerced to have a slightly different lexical meaning when used in a construction that elsewhere functions as a regular antipassive construction. Apart from affecting aspectual properties, such shifts entail an increase in the agentivity of the experiencer. The two meanings that are particularly li ely or antipassive counterparts o ‘see’ and ‘hear’ are ‘loo (at)’ and ‘listen’ correspondingly.14 Ainu, which is one of about ten ValPal languages that have antipassives, has both of these lexicalized antipassives: inkar ‘loo at’ “can be traced back to the antipassive *i-nukar” (where nukar is ‘see’, and i- is the antipassive prefix) [Bugaeva 2015: 814]; i-nu is glossed as both ‘hear’ and ‘listen’ as opposed to the basic nu, which means ‘hear’ [ibid.: 831] ( or urther discussion see also [Bugaeva 2013]). Similar effects are observed in antipassivized perception verbs in languages outside the ValPal database. A good example is provided by Warungu antipassive, signalled by the suffix –gali. In his detailed analysis, Tsunoda [2011: 476–483] shows that meanings of antipassive counterparts to six transitive perception verbs in Warungu intricately depend on a number of factors, but each pair involves some degree of lexicalization. In most cases, idiosyncrasies involve an increase in the experiencer’s ability to control the situation, as can be seen rom the ollowing examples of transitive (19) and corresponding antipassive (20) constructions. (19) nyola nyaga-n 3SG.ERG see-NON.FUT ‘ e saw (or ound) bees.’ worriba-Ø bee-ACC (20) ngaya nyaga-gali-n waybala-wo 1SG.NOM see-ANTI-NON.FUT white.man-DAT Other attested meanings are ‘meet’ and ‘be sighted’ or lexicalized antipassives o ‘see’ and ‘understand’ and ‘thin ’ or lexicalized antipassives o ‘hear’. 19 14 ‘I loo ed at/watched the white man.’ [Tsunoda 2011: 478] 6.3. Semantic incorporation of the P-argument As was discussed above, inherent specification of the nature of the P-argument in the verb’s lexical meaning is one of the properties that acilitate the verb’s ability to participate in antipassive alternations. However, in some cases the relevant semantic property is coerced in the antipassive whereas in the transitive verb it is absent or weak. In other words, antipassivized verbs can entail specific properties of their unexpressed objects which are not obligatory for the P-arguments of the corresponding transitive verbs. This kind of coercion is well-documented for both morphologically marked antipassives and intransitive uses of A-labile verbs. An inchoate pattern of this kind has been identified for unspecified object alternation in English and elsewhere, i.e. for intransitive uses of verbs like bake, clean, draw etc. With most such alternating verbs, “[d]espite the lac o overt direct object in the intransitive variant, the verb in this variant is understood to have as object something that qualifies as a typical object of the verb” [Levin 1993: 33; bold ace mine — AUTHOR]. Thus, even here, intransitivization slightly affects the lexical meaning of the verb — to the extent that non-typical objects are excluded. With some other verbs this effect of narrowing down the range of possible objects is more drastic. For example, it has been observed that “[i]ntransitive eat is typically ta en to mean ‘eat a meal’ (I’ve eaten already), while objectless drink has as almost its only possible reading ‘drin alcohol’ (John drinks)” [Næss 2009: 35]. This means that syntactic deletion of the object has a rather strong effect for the lexical meaning of drink in English and many other languages with a similar idiosyncrasy. This effect can be interpreted as semantic incorporation of an object which belongs to a very narrow ontological class. Similar semantic effects, both weak (a typical object of the verb is implied) and strong (an object a narrow class is implied), are often observed in languages with morphologically marked antipassives. For example, the very same coercion of ‘alcohol’ as the understood object o ‘drin ’ is observed in the marked antipassive construction in Kaqchikel (< Mayan): (21) N-Ø-qum-un ri achin INCOMPL-3SG.ABS-drink-ANTI DET man ‘The man drin s (alcohol).’ [ eaton 2017: 327] Coercion of a specific kind of implied object is systematically found in one of the antipassive-like constructions in Russian (apart from lexicalization, it meets the definitional criteria of the antipassive). This construction is analyzed in [Say 2005] under the label o “lexical sja-antipassives” (sja- is etymologically a reflexive affix, which covers a wide range of functions in the middle domain). Many transitive verbs that allow derivation of lexical sja-antipassives inherently (that is, even in their transitive use) require objects belonging to very narrow classes, e.g. specific bodyparts, cf. naxmurit’ ‘to nit (one’s brow)’, vysmorkat’ ‘to blow (one’s nose)’. In such cases, sja-affixation has the syntactic effect of intransitivization, but the lexical 20 meaning of the verb is not significantly affected. However, other transitive verbs that participate in the same alternation can be combined with a wide class of overt objects; and yet, their sja-counterparts imply objects rom a particular class: “One may stroit’ (‘build’ — AUTHOR) houses, bridges, clubs, roads etc., stroit’sja means ‘to build a living place, a house, an edifice for living’; (...) one may tratit’ (‘spend’ — AUTHOR) one’s money, salary, stipend, paper as well as (metaphorically) one’s time, orces etc., but tratit’sja means ‘to spend one’s money, ( inancial) means’; (...) one may propit’ (‘drink away’ — AUTHOR) anything (without any lexical restriction), but propit’sja means ‘to drink away everything one possesses’ [Yanko-Trinickaya 1962: 175]. The Russian pattern just discussed is a lexical phenomenon: it is restricted in terms of productivity and semantically incorporated objects generally cannot be predicted rom the verb’s meanings. Despite this, the semantic relationships between the transitive verb and its sja-counterpart are fully transparent. A further possibility along the pathway of lexicalization is observed in fossilized derivatives that employ antipassive morphology but are no longer transparently related to transitive uses. For example, in Tatar (< Turkic) there is a reflexive/middle affix –n that functions as an antipassive (object-deleting) affix with some verbs. Apart from fully transparent alternations, it is also used in pairs such as e.g. čiš-en- ‘undress’ rom čiš- ‘untie, unbutton’ or ukɤ-n- ‘mumble, read a prayer’ rom ukɤ- ‘read, teach’ [Zinnatullina 1969: 176–181]. Postulating a specific type of semantically incorporated object (‘one’s clothes’ and ‘prayer’ in examples just cited) is use ul or reconstructing the path of semantic development of these verbs, but clearly there is no synchronically transparent syntactic correspondence in such pairs. Similarly, in her discussion of the antipassive marker –e in Wolof (Atlantic), Nouguier-Voisin mentions the transitive verb bëgg ‘love, want’ and its lexicalized derivative bëgg-e ‘be greedy’ and hypothesizes that this derivation is based on the meaning ‘to want money’ [2002: 311]. Again, synchronically the relationship between these two verbs is not transparent. The processes discussed in 6.3 imply that if there is a contrast between a transitive verb and its antipassive counterpart with respect to object-specialization, than it is the antipassive counterpart which is more restrictive. This effect is most clearly seen in transitive verbs that can combine with a wide range of possible objects: their antipassive counterparts often coerce understood objects of certain kinds, which leads to lexicalization and eventually fossilization of erstwhile antipassives (an example of such scenario is found in Tolowa Athabaskan [Givón & Bommelyn 2000: 53]). However, narrowing down the range of possible objects under antipassivization can be attested even with those transitive verbs which themselves can only be combined with a relatively narrow class o objects. In act, even the verb ‘eat’, which is found at the top of the antipassivizability hierarchy and selects objects from a compact semantic domain (FOOD), can have lexicalized antipassive counterparts of this kind. For example, in Quiché Maya the "absolutive voice” (an objectless antipassive construction) counterpart of tix ‘eat’ has the speci ic reading ‘eat people, be carnivorous’ [Næss 2009: 36, with urther re erence to Mondloch 1981: 189]. 21 7. Discussion Certainly there are significant cross-linguistic differences in the ways lexical semantic features interact with antipassives. Moreover, the very hierarchy of verb meanings which reflects their likelihood to participate in the antipassive alternation, which partially served as the point of departure in this study (see Section 2.1), evidently is not a strict implicational hierarchy. In fact, in terms of its lexical profile, the antipassive as a cross-linguistic phenomenon «behaves somewhat multidimensionally but is just on the right side o Guttman’s threshold or unidimensionality» [Wichmann 2016: 437]. This multidimensionality reflects language-specific semantic factors which favour or hinder individual verb’s ability to participate in the antipassive alternation. To give an example, there is an antipassive pattern in Slavic languages which is only possible with verbs that denote aggressive forms of behaviour (see [Janic 2013: 139–142] for a discussion) — a correlation which does not seem to be particularly common cross-linguistically. Although language-specific phenomena of this kind disturb the unidimensional hierarchy and other implicational generalizations, there still are recurrent patterns in the interaction between antipassive and verb’s lexical meaning. In this contribution, I was trying to find these cross-linguistic similarities and disregarded differences. The main empirical claim of this study is identification of natural antipassive properties: those components in the lexical meaning of the verb, which facilitate the verb’s ability to participate in the antipassive alternation. Natural antipassive properties include i) agentivity of the A-argument; ii) specification of the A’s manner; iii) inherent atelicity; iv) narrow class of potential P-arguments; v) affectedness of the A-argument. These five properties recurrently manifest themselves in four types of phenomena in the domain of antipassivization. First, verbs that have more natural antipassive properties are more susceptible to antipassivization in languages where antipassives are lexically restricted. Second, these verbs are more likely to yield the antipassive interpretation when combined with syncretic markers that also cover other valency-related functions. Third, in languages with lexically conditioned distribution between two antipassive constructions, verbs with more natural antipassive properties tend to choose the construction which is formally less marked (sometimes, unmarked), whereas other verbs participate in the more marked antipassive alternation. And fourth, if there is a difference in the lexical meaning of a transitive verb and its lexicalized antipassive counterpart, then the antipassivized alternant has natural antipassive properties to the same or larger extent than its transitive counterpart. In other words, antipassivization can result in coercion of natural antipassive properties. All of the generalizations above are relative rather than absolute. For example, there are languages where antipassives are fully productive and semantically transparent. Such languages do not show lexical effects discussed here; however, they irrelevant for the findings above rather than discard them. Similarities in the lexical machinery of antipassivization were often observed in languages that drastically differ in many other respects, including the formal marking of antipassive construction and alignment phenomena. In particular, although no 22 attempt was made here to directly compare languages with ergative and accusative alignment features (and certainly there can be statistical differences between them), lexical phenomena discussed here are systematically observed in languages with discrepant alignment patterns. Now that the empirical findings of the study are summed up, we are in a position to (somewhat speculatively) situate these findings in a wider theoretical context. Antipassives are commonly believed to be similar to passives in that they do not change numerical valency; moreover, according to many definitions, including the one adopted in this volume, the lexical meaning of the verb should not be affected by antipassivization. Thus, in principle antipassives — as well passives — are expected to be inflectional. In this respect both antipassives and passives contrast with e.g. causatives and anticausatives, which, by definition, do change the verb’s numerical valency — the cornerstone o the verb’s lexical meaning — and hence are expected to be derivational (see e.g. [Haspelmath 2002: 218]). However, in reality there seems to be a problem. Antipassive constructions, that is, construction in which the P-argument is either demoted or left unexpressed, are typically motivated by semantic or pragmatic actors, which re lect “a certain degree of difficulty with which an effect stemming from an activity by A on an identifiable O can be recognized” [Cooreman 1994: 51]. However, P-arguments are very tightly integrated into the meaning of transitive verbs; typically they are more relevant for the verb’s meaning than A-arguments. This is the likeliest reason why antipassive alternations tend to strongly interact with the lexical meaning of the verb. Many of the phenomena discussed above are explained by the tight integration o the P’s semantic properties into the verb’s meaning: the difficulty of forming antipassives from causative change-of-state verbs li e ‘open’ or ‘brea ’; antipassive’s selectivity with respect to the P’s animacy / inanimacy; the tendency to infer an object of a particular kind in constructions without overtly expressed P-argument. In short, antipassivization can easily lead to semantic developments which ultimately a ect the verb’s meaning. Antipassives which are used to rearrange the arguments syntactically without affecting the lexical meaning of the verb are not typologically common. It may be further hypothesized that antipassives of this kind are diachronically unstable: limitations on productivity, aspectual shifts, semantic absorption of specific types of implied objects and other lexicalization patterns can contribute to the erosion of productive and semantically regular antipassive alternations; potentially this leads to fossilization of erstwhile antipassives. These developments are then likely to foster grammaticalization of newer antipassive constructions, which results in grammatical systems with multiple antipassive constructions — typological occurrence of such systems is remarkably high [Heaton 2017: 265–285]. 23 References Arkadiev, Peter & Alexandr Letuchiy. 2008. Derivacii antipassivnoj zony v adygejskom jazyke. Vladmir Plungian & Sergey Tatevosov (eds). Issledovanija po glagol’noj derivacii. Moscow: JaSK. 77–102. Austin, Peter K. 2013. A Grammar of Diyari, South Australia. Second edition. London: SOAS, University of London. Basilico, David. 2012. The antipassive and its relation to scalar structure. In: Cuervo, María Cristina & Yves Roberge (eds.). Syntax and Semantics, 38. The end of argument structure? 75–103. Bingley: Emerald Press. Bugaeva, Anna. 2013. Ainu Valency Patterns. In: Hartmann, Iren & Haspelmath, Martin & Taylor, Bradley (eds.) 2013. Valency Patterns Leipzig. Leipzig: Max Planck Institute for Evolutionary Anthropology. (Available online at http://valpal.info/languages/ainu, Accessed on 2017-06-24) Bugaeva, Anna. 2015. Valency classes in Ainu. In Malchukov & Comrie (eds.) 2015. 807–854. Campbell, Lyle. 1985. The Pipil language of El Salvador. Berlin et al.: Mouton. Cooreman, Ann. 1994. A functional typology of antipassives. In: B. Fox & P. J. Hopper (eds.). Voice: form and function. Amsterdam: Benjamins. 49–88. Creissels, D. 1991. La voix en sonin é. Mandenkan 23. 1-24. Creissels, Denis. 2012. The origin of antipassive markers in West Mande languages. Paper presented at the 45th Annual Meeting of the Societas Linguistica Europaea. Stockholm, 29 August — 1 September 2012. Creissels, Denis. 2014. P-lability and radical P-alignment. Linguistics, 52(4). 911–944. Dayley, Jon P. 1985. Tzutujil Grammar [= Publications in linguistics, 107]. Berkeley: University of California Press. Dixon, Robert Malcolm Ward. 1972. The Dyirbal language of North Queensland. Cambridge: Cambridge University Press. Dixon, Robert Malcolm Ward. 1977. A grammar of Yidiɲ. Cambridge: Cambridge University Press. Dixon, Robert Malcolm Ward & Alexandra Yu. Aikhenvald. 2000. Introduction. In: Dixon, R. M. Ward & A. Yu. Aikhenvald (eds.). Changing valency: case studies in transitivity. Cambridge: Cambridge University Press. P. 3–29. Estigarribia, Bruno. 2017. A grammar s etch o Paraguayan Guaraní. In: Bruno Estigarribia and Justin Pinta (eds.). Guarani Linguistics in the 21st Century. Leiden: Brill. P. 7–85. Fillmore, Charles J. 1986. Pragmatically controlled zero anaphora. In: Vassiliki Nikiforidou, Mary VanClay, Mary Niepokuj & Deborah Feder (eds.). 24 Proceedings of the twelfth annual meeting of the Berkeley Linguistics Society. Berkeley: University of California. 95–107. Fleck, David W. 2006. Antipassive in Matses. Studies in language 30(3). 541–573. Givón, Talmy & Loren Bommelyn. 2000. The evolution of de-transitive voice in Tolowa Athabaskan. Studies in language 24(1). 41–76. Hartmann, Iren 2013. ooc Valency Patterns. In: artmann et al. (eds.). (Available online at http://valpal.info/languages/hoocak/examples/430, Accessed on 201706-24) Hartmann, Iren & Haspelmath, Martin & Taylor, Bradley (eds.) 2013. Valency Patterns Leipzig. Leipzig: Max Planck Institute for Evolutionary Anthropology. (Available online at http://valpal.info, Accessed on 2017-06-28) Haspelmath, Martin. 1993. More on the typology of inchoative/causative verb alternations. In Bernard Comrie & Maria Polinsky (eds.). Causatives and transitivity. Amsterdam, Philadelphia: John Benjamins.87–120. Haspelmath, Martin. 2002. Understanding Morphology. London: Arnold. Haspelmath, Martin. 2015. Transitivity prominence. In Malchukov & Comrie (eds.) 2015. 131–147. Heath, Jeffrey. 1976. Antipassivization: a functional typology. In: H. Thompson, K. Whistler et al. (eds.). Proceedings of the Second Annual Meeting of Berkeley Linguistics Society. Berkeley: Berkeley Linguistics Society. 202–211. Heaton, Raina. 2017. A typology of antipassives, with special reference to Mayan. Ph.D. dissertation. University of Hawaii. Heath, Jeffrey. 1999. A grammar of Koyraboro (Koroboro) Senni. The Songhay of Gao, Mali. Köln: Rüdiger Köppe. Hopper Paul J. & Thompson Sandra A. 1980. Transitivity in grammar and discourse. Language 56 (2). 251–350. Janic, Katarzyna. 2013. L’antipassif dans les langues accusatives. Ph.D. dissertation. Université Lumière Lyon2. Kazenin, Konstantin I. 1994. On the lexical distribution of agent-preserving and object-preserving transitivity alternation. Nordic Journal of Linguistics 17(2). 141–154. Kemmer, Suzanne. 1993. The middle voice. Amsterdam / Philadelphia: John Benjamins. Kibrik, Aleksander E. & Jakov G. Testelec. 2004. Bezhta. In: Michael Job (ed.). The indigenous languages of the Caucasus. Vol. 3. The North East Caucasian languages. Part 1. Ann Arbor: Caravan. 217–298. Levin, Beth. 1993. English verb classes and alternation. A preliminary investigation. Chicago: University of Chicago Press. 25 Levin, Beth. 2015. Verb Classes Within and Across Languages. In: Malchukov & Comrie (eds.) 2015. 1627–1670. Lichtenber , Františe . 1999. Reciprocals without reflexives. In: Zygmunt Frajzingier & Tracy S. Curl (eds.). Reflexives: forms and functions. Amsterdam, Philadelphia: John Benjamins. 31–63. Lichtenber , Františe . 2008. A grammar of Toqabaqita. Berlin — New York: Mouton de Gruyter. Malchukov, Andrej. 2005. Case pattern splits, verb types, and construction competition. In: Mengistu Amberber & Helen de Hoop (eds.). Competition and Variation in Natural Languages. London: Elsevier. 73–117. Malchukov, Andrej. 2015. Valency classes and alternations: parameters of variation. In: Malchukov & Comrie (eds.) 2015. 73–130. Malchukov, Andrej & Comrie, Bernard (eds.). 2015. Valency classes in the world’s languages. Vols. 1–2. Berlin, Boston: De Gruyter Mouton. Mondloch, James Lorin. 1981. Voice in Quiche-Maya. Ph.D. dissertation. State University of New York at Albany. Næss, Åshild. 2007. Prototypical Transitivity [= Typological studies in language, 72]. Amsterdam / Philadelphia: John Benjamins. Næss, Åshild. 2009. How transitive are EAT and DRINK verbs? In: John Newman (ed.). The linguistics of eating and drinking. Amsterdam, Philadelphia: John Benjamins. 27–43. Nouguier-Voisin, Sylvie. 2002. Relations entre fonctions syntaxiques et fonctions sémantiques en wolof. Thèse de doctorat. Université Lumière Lyon 2. Polinsky, Maria. 2013. Antipassive Constructions. In: Dryer, Matthew S. & Haspelmath, Martin (eds.). The World Atlas of Language Structures Online. Leipzig: Max Planck Institute for Evolutionary Anthropology. (Available online at http://wals.info/chapter/108, Accessed on 2017-06-21.) Polinsky, Maria. 2017. Antipassive. In: Jessica Coon, Diane Massam & Lisa DeMena Travis (eds.). The Oxford handbook of ergativity. Oxford: Oxford University Press. P. 308–331. Pustet, Regina & David S. Rood. 2008. Argument dereferentialization in Lakota. In Mar Donohue & Søren Wichmann (eds.). The typology of semantic alignment. Oxford: Oxford University Press. P. 334-356. Quesada, Diego J. 2007. The Chibchan languages. Cartago: Editorial Tecnólogica de Costa Rica. Rappaport Hovav, Malka & Beth Levin. 1998. Building verb meanings. In: Miriam Butt and Wilhelm Geuder (eds.). The Projection of Arguments: Lexical and Compositional Factors. Stanford: CSLI publications, 97–134. 26 Sansò, Andrea. 2017. Where do antipassive constructions come rom? A study in diachronic typology. Diachronica 34(2). 175–218. Say, Sergey. 2005. Antipassive sja-verbs in Russian: between inflection and derivation. In: Wolfgang U. Dressler, Dieter Kastovsky, Oskar E. Pfeiffer & Franz Rainer (eds). Morphology and its demarcations. [= Current issues in linguistic theory, 264]. Amsterdam, Philadelphia: John Benjamins. 253–275. Say, Sergey. 2008. K tipologii antipassivnyx konstrukcij: semantika, pragmatika, sintaksis. Ph.D. dissertation. St. Petersburg: Institute for linguistic studies, RAS. Sumbatova, Nina & Rasul Mutalov. 2003. A Grammar of Icari Dargwa. München — Newcastle: LINCOM Europa. Tatevosov, Sergei. 2011. Detelicization and argument suppression: Evidence from Godoberi. Linguistics 49. 135-174. Terrill, Angela. 1997. The development of antipassive constructions in Australian languages. Australian journal of linguistics 17. 71–88. Tsunoda, Tasaku. 1985. Remarks on transitivity. Journal of Linguistics 21(2). 385– 396. Tsunoda, Tasaku. 2011. A grammar of Warrongo. Berlin: Mouton. Vakhtin, Nikolai B. 1981. Refleksiv v eskimosskom jazyke. In: Viktor S. Xrakovskij (ed.). Zalogovye konstrukcii v raznostrukturnyx jazykax. Leningrad: Nauka. 266– 272. Wichmann, Søren. 2015. Statistical observations on implicational (verb) hierarchies. In: Malchukov & Comrie (eds.) 2015. 155–181. Wichmann, Søren. 2016. Quantitative tests o implicational verb hierarchies. Kageyama, Taro and Wesley M. Jacobsen (eds.). Transitivity and Valency Alternations: Studies on Japanese and Beyond [= Trends in Linguistics, Studies and Monographs, 297]. Berlin: De Gruyter Mouton. 423–444. Yanko-Trinickaya, Nadija A. 1962. Vozvratnye glagoly v sovremennom russkom jazyke. Moscow: AN SSSR. Yemel’yanova, Nina M. 1982. Klassy glagolov v eskimosskom jazyke. Leningrad: Nauka. Zinnatullina, K. Z. 1969. Zalogi glagola v sovremennom tatarskom jazyke. Kazan: tatrskoe knizhnoe izdatelstvo. 27