1. Introduction

An antipassive construction is a derived detransitivized construction with a two-place predicate, related to a corresponding transitive construction whose predicate is the same lexical item. In the basic transitive construction, the patient-like argument is realized as a direct object; in the antipassive construction, that argument is either suppressed (left implicit) or realized as an oblique complement. The term antipassive (Silverstein 1972) was coined to indicate that the construction is a mirror image of the passive: in the passive, the suppressed or demoted argument is the agent-like argument, in the antipassive, the patient-like argument. An example of a transitive/antipassive alternation is given in (1a-b).

(1) Chukchi (Kozinsky et al. 1988: 652)

a.	ʔaaček-a	kimitʔ-ən	ne-nlʔetet-ən
	youth-erg	load-abs	3pl.subj-carry-aor.3sg.obj
	‘The young men carried away the/a load.’ (trans)
b.	ʔaaček-ət	ine-nlʔetet-gʔe-t	kimitʔ-e
	youth-abs	antip-carry-aor.3.subj-pl	load-instr
	‘The young men carried away the/a load.’ (anti)

In (1a), the transitive verb ‘carry’ agrees with the ergative subject and absolutive object. In (1b), the verb is marked with the antipassive prefix ine- and no longer agrees with the object; the object is now expressed by an oblique case (instrumental).

A verb in the antipassive is derived from the corresponding transitive verb, often with the help of overt morphology. For example, in (1b), the verb nlʔetet- bears the dedicated antipassive prefix ine-. Languages also make use of antipassive markers that are syncretic with other categories: detransitivizers and aspect/modality markers.

Some languages show syncretism between the morphology of the antipassive and the morphology of other detransitivizing operations, most commonly reflexivization (e.g., in Pama-Nyungan languages of Australia). The antipassive verb can take on other formal characteristics of intransitive verbs in a given language; one such common characteristic is the change in the agreement pattern, as illustrated by the Chukchi examples above. In some languages (for example, in Mayan: England 1983b; 1988), a change in verbal agreement may be the sole indication of the antipassive. Finally, there is a correlation between the antipassive and the imperfective/irrealis (see §4.3).

2. Definition of values

Given that our knowledge of variation in antipassive constructions is far from complete, it is possible that some languages which are identified on Map 108A as having no antipassive will ultimately be reanalyzed as having an antipassive construction.

Among the languages which have the antipassive construction, the major division is between the type that requires the patient-like argument to be unexpressed (the implicit argument type) and the type where the patient-like argument can be expressed by an oblique complement. The following values are represented on the map:

Values of Map 108A. Antipassive Constructions

Go to map
	Value	Representation
	Antipassive with patient-like argument left implicit	18
	Antipassive with patient-like argument expressed as oblique complement	30
	No antipassive	146
Total:		194

As shown on Map 108A, the antipassive is found across different language families and geographical areas; the antipassive with an oblique complement seems to be more widespread than the antipassive with an implicit argument.

3. The antipassive construction and ergativity

One of the vexing questions concerning the distribution of the antipassive is its correlation with ergativity (see Chapters 98, 99, and 100 for ergative constructions). Some researchers insist on the link between the antipassive and ergativity (Silverstein 1976; Dixon 1979; Spencer 1991: 24), while others propose that the antipassive is not limited to ergative languages (Heath 1976; Postal 1977; Davies 1984; Givón 1984; Lidz 1996). Table 1 lists all the languages of the sample with antipassives; it shows no principled correlation between ergativity and the antipassive. The transitive/antipassive alternation is simply more visible in an ergative language, where it typically involves a change in subject case marking from ergative to absolutive, as in (1a-b) above.

4. Theoretical issues

4.1. Productivity of the antipassive construction

Antipassives often apply only to a certain subset of transitive predicates (Gibson 1980; Chung 1998: 39). Productive antipassives show genealogical and areal clustering, as indicated on Map 108B: they are found in some Mayan, Salishan, Nakh-Daghestanian, Austronesian, and Australian languages.

Values of Map 108B. Productivity of the Antipassive Construction

Go to map
	Value	Representation
	productive	24
	partially productive	14
	not productive	2
	no antipassive	146
Total:		186

4.2. Morphological marking

The variation observed with respect to the marking of antipassives has to do with whether or not the language has a dedicated antipassive marker. A dedicated marker can often be associated diachronically with other functions (e.g., middle, reflexive), so what matters is its nonsyncretic synchronic status. In our sample, languages with a dedicated antipassive marker include: Acoma, Chamorro, Chukchi, Comanche, Diyari, Dyirbal, Godoberi, Gooniyandi, West Greenlandic, Halkomelem, Hunzib, Jakaltek, Kiowa (“Type 2” antipassive, see Watkins 1984: 140-141, 149-150), Koyraboro Senni, Krongo, Lango, Mam, Ojibwa, Tzutujil, Warrungu, Zoque (Copainalá).

In languages that have several different antipassive markers, these markers can co-occur (or “stack”). Stacking is found in Chukchi (the productive antipassive prefix ine- and the semiproductive antipassive suffix –tko-, Kozinsky et al. 1988: 661) and in Halkomelem (the middle antipassive suffix –əm/ -eʔəm- and the activity suffix –els-, Gerdts and Hukari 2005). The semantic contrast between the stacked and the single antipassive is, however, unclear.

4.3. Semantic and discourse functions of the antipassive

The semantic import of the antipassive has to do with the affectedness and individuation of the patient.

The use of a prototypical transitive verb entails that the event denoted by that verb causes a change of state in the object participant (Tsunoda 1981; Hopper and Thompson 1980; Van Valin 1991; Dowty 1991, among many others). The semantic function of the antipassive is to cancel such an entailment; this correlates with the marking of the patient as an oblique complement. Comparable effects can be found in the English conative alternation, where by the referent of a direct object is construed as affected while the referent of a prepositional complement is not (Levin 1993: 5-11):

(2)

a.	The hunter shot the bear (entailment: the bear died)
b.	The hunter shot at the bear (no entailment; the bear may not be affected at all)

If there is no affected participant which allows one to measure out the effects of the event (the so-called incremental theme, Dowty 1991), the event itself is interpreted as incomplete. This accounts for the high correlation between the use of the antipassive and the habitual, durative, iterative, and imperfective (Tchekhoff 1987; Cooreman 1994; Dixon 1994; Dowty 1991; van den Berg 2001: 60, and others). A correlation between the use of antipassive and irrealis is found in Yukulta (Keen 1983).

Compared to the transitive construction, the patient participant in the antipassive construction is lower in individuation (or identifiability). A lower degree of individuation may be manifested by a number of structural possibilities, summarized in Table 2.

A referent with low individuation is less likely to be maintained in discourse; this explains why the antipassive is often used to introduce episodic referents (Polinsky and Nedjalkov 1987; Bittner 1987), whose identification is not crucial for communication.

However, there are languages (e.g., Yukulta) where the antipassive is required when the object is expressed by a first or second person pronoun, which are arguably the highest on the scale of individuation. Such a subtype contrasts with the subtype exemplified by Halkomelem, where the antipassive is impossible with a first or second person object (Gerdts 1988a: 157). This variation supports the conclusion, made by several researchers, that the semantic and discourse functions of antipassives can and do differ across languages (Comrie 1978; Heath 1976; Cooreman 1988, 1994).

4.4. Structural functions of the antipassive

In a number of languages, only a particular grammatical role or case form can serve as the syntactic pivot for extraction, reference tracking across clauses, and other grammatical processes. Antipassivization makes the sole argument of the detransitivized verb accessible to relevant grammatical processes; in many ergative languages, where the absolutive is the pivot for all or most grammatical processes, antipassivization allows the subject to occur in the absolutive case. To illustrate the syntactic function of antipassives, let us turn to more Chukchi examples. In Chukchi, relativization is possible only for absolutive arguments (Polinsky 1994); thus, the subject of intransitives and the direct object can relativize directly, as shown by (3) for the subject absolutive. In order to relativize the subject of a transitive (4), it first needs to be converted into an absolutive, via antipassivization (5a); the subject of the antipassive is then relativized, (5b).

(3)

a.	ŋinqey	pəkir-gʔi
	boy.abs	arrive-aor.3sg
	‘The boy arrived.’
b.	pəkərə-lʔ-ən	ŋinqey
	arrive-ptcp-abs	boy
	‘the boy that arrived’

(4)

a.	tumg-e	ŋinqey	rəyegtetew-nin
	friend-erg	boy.abs	save-aor.3sg.3sg
	‘The friend saved the boy.’
b.	* [ŋinqey	rəyagtala-lʔ-ən]	tumgətum
	boy.abs	save-ptcp-abs	friend
	(‘the friend that saved the boy’)

(5)

a.	tumgətum	ŋinqey-ək	ine-nyegtele-gʔi
	friend.abs	boy-loc	anti-save-aor.3sg
	‘The friend saved the boy.’
b.	[ŋinqey-ək	ine-nyegtelewə-lʔ-ən]	tumgətum
	boy-loc	anti-save-ptcp-abs	friend
	‘the friend that saved the boy’

The antipassive also allows speakers to avoid coreference between the object of a transitive clause and the subject of a following clause; such use of the antipassive is observed in Australian languages (Tsunoda 1988; Cooreman 1994), Chukchi (Kozinsky et al. 1988: 689-697), and Mayan (England 1983b).