Ergative case is said to mark transitive subjects, and it is widely assumed that this is true under the ordinary definition of transitive; however, Bittner and Hale (1996) propose that ergative languages fall into two types, neither of which is based on the ordinary notion of transitivity. In one, a direct object is not necessary for ergative case: any verb with an external argument counts as transitive, following Hale and Keyser 1993 (e.g., Warlpiri). In the other, a direct object is necessary, but not sufficient: the subject gets ergative case only if the object moves out of the VP (e.g., Inuit). This article argues that Niuean, Dyirbal, and Nez Perce are also of this object shift type. A search yielded no language where ergative case is clearly governed only by ordinary transitivity; languages that do fit the stereotype have only ergative agreement. A formal account of the correlation between object shift and ergative case is proposed, under which ergative case can be used as a ‘‘last resort,’’ as one of three ways to avoid the locality violation that object shift creates.
Transitivity is central to the definition of an ergative pattern in typology (e.g., Comrie 1978, Dixon 1979): transitive subjects are marked differently from intransitive subjects and transitive objects. It is often assumed that the word transitive in the typological definition of an ergative pattern has its ordinary/dictionary meaning wherein a clause is transitive if and only if a direct object is present in syntax. However, Bittner and Hale (1996) argue that languages with ergative case fall into two types, neither of which conforms to this stereotypic pattern.
In Bittner and Hale’s first type, illustrated with Warlpiri, ergative case marks the subjects of all verbs that are transitive under the definition in Hale and Keyser 1993: namely, where all verbs with an external argument count as transitive, because they are assumed to always have at least a semantically incorporated object.
‘The man is singing.’
(Levin 1983:149, (4.23b))
The presence of a direct object in syntax is not required in order for ergative case to mark the subject in this first type of ergative language, a type known in the literature and often called active or active ergative.
In contrast, Bittner and Hale’s second type is new and surprising. In this second type, illustrated with Inuit, the presence of a direct object in syntax is necessary, but not sufficient for ergative case to be used. The direct object must move out of the VP before ergative case can mark the subject. In Inuit, specific objects move out of the VP (Bittner 1994), undergoing object shift just as in Icelandic (Diesing 1992), and ergative subjects occur only in clauses where object shift has occurred. In contrast, nonspecific objects remain in situ in the VP, and such clauses take a nominative subject. This contrast is shown in (2) with a specific object and an ergative subject, versus (3) with a nonspecific object and a nominative subject.
Juuna-ERG child-PL(NOM) look.after-IND-[+TRANS]-3SG.3PL
‘Juuna is looking after the children.’
(Bittner and Hale 1996:544, (22b))
Juuna(NOM) book-PL.INSTR two- paid.for-get-IND-[−TRANS]-3SG
‘Juuna bought two books.’
(Bittner 1994:72, (45b))
Evidence that the object moves out of the VP in Inuit clauses with an ergative subject involves scope; the object is necessarily outside the scope of negation in the ergative construction, indicating that it is higher than the negative (Bittner 1994:35).
Juuna-ERG book one get-PERF-NEG-IND-3SG.SG
‘There is one book which Juuna hasn’t got ( yet).’
(Bittner 1994:35, (71a))
Bittner then notes, ‘‘The corresponding oblique object can only take narrow scope, indicating that it remains below negation, inside the VP’’ (1994:35).1
Juuna book-INSTR one-INSTR get-ANTIPASS-PERF-NEG-IND-3SG
(Bittner 1994:35, (71b))
Bittner and Hale’s proposal does not include a third type, corresponding to what many linguists assume is the stereotypic ergative language, where any clause with a syntactic direct object takes an ergative subject, regardless of whether that object is inside or outside the VP. If Bittner and Hale’s two types of languages with ergative case are all that exist, then every ergative language that has been assumed to follow that stereotypic pattern must actually be either an active ergative language, like Warlpiri, or what I will call an object shift ergative language, like Inuit. Investigating this hypothesis is the topic of this article.
Now, despite Bittner’s (1994) evidence that specific objects undergo object shift in Inuit, readers could still doubt that anything more than ordinary transitivity is required for the subject to be marked ergative in Inuit, given that nonspecific objects in Inuit get a case that Bittner labels as instrumental. Under a narrow definition of transitivity, where only clauses with a direct object with structural case count as transitive, the Inuit examples with a nominative subject and an object marked with instrumental case would be classified as intransitive. The first goal of this article is to provide additional, clearer evidence that the object shift type of ergative language does exist (section 2).
Strong evidence for the existence of the object shift type of ergative language comes from Niuean (section 2.1). Niuean is like Inuit in that ergative case marks the subject only when the object moves out of the VP (Massam 2000). On the basis of evidence from word order in Niuean, Massam shows that specific objects move out of the VP, while nonspecific objects remain inside the VP, just as in Inuit. Unlike nonspecific objects in Inuit, however, nonspecific objects in Niuean are not marked with an oblique case; instead, clauses with a nonspecific direct object look much like transitive clauses in English and would likely be analyzed as ordinary nominative-accusative constructions, except for the fact that the language also has clauses with an ergative subject. Seiter (1980), working under the assumption that all ergative languages must fit the stereotypic ergative pattern where all and only transitive clauses have an ergative subject, forces Niuean into that mold by claiming that nonspecific objects are actually incorporated. However, Massam (2000, 2006) shows that, given the length and complexity of many nonspecific objects, the incorporation account cannot be correct for Niuean. In contrast, the Niuean pattern fits Bittner and Hale’s (1996) object shift type of ergative pattern perfectly; the subject is ergative only when the object moves out of the VP, and there is no need to claim that apparently transitive clauses with a nonspecific object are really intransitive.
Other languages in which ergative case marks the subject only when the object moves out of the VP include the related language Tagalog (Aldridge 2012) (section 2.2) and Dyirbal (Dixon 1972) (section 2.3). However, like Inuit, those two languages could also fit the stereotypic ergative pattern, because clauses with a nominative subject have objects marked with what looks like an oblique case. In contrast, Nez Perce (section 2.4) only fits the object shift type of ergative language. As in Niuean, in Nez Perce the subject is ergative only when object shift has occurred. Clauses without object shift have a nominative-accusative case pattern (Woolford 1997, Carnie and Cash 2006). Nez Perce has ditransitive examples with a nominative-accusative-accusative pattern that cannot be analyzed as intransitive in order to force the language to fit the stereotypic ergative pattern. Instead, Nez Perce conforms only to Bittner and Hale’s object shift type of ergative language.
The second goal of this article is to contrast the behavior of this object shift type of ergative language with examples of the active ergative type. Active ergative languages discussed here include several Caucasian languages (Georgian, Batsbi, Ingush, and Udi), plus Basque and Kashmiri (section 3).
The third goal of this article is to begin to address an important question raised by Bittner and Hale’s (1996) proposal: do all languages with ergative case fit into either the active ergative or the object shift ergative type? That is, is there any clear evidence that the stereotypic ergative language even exists, wherein ergative case marks the subject only when there is a direct object in syntax, even if that object remains inside the VP? Or do all languages with ergative case belong to one of the two types in (6)? In the research for this article, no clear counterexample to Bittner and Hale’s claim was found, although there remain a number of languages with ergative case whose type cannot yet be determined. Some of these unclassified languages are discussed in section 4: Archi, Kabardian, Tsez, and Kalkatungu. These languages clearly do not belong to the active type, because they require the presence of a syntactic object in order for the subject to be ergative. What remains to be determined is whether the object in clauses with an ergative subject is always outside the VP in these languages. That is not immediately obvious in SOV languages, nor in languages with very free word order.
Languages where only the agreement pattern is ergative are discussed in section 5, because these languages are commonly assumed to have covert ergative case. In languages of the Selayarese family, the agreement pattern is ergative only when the direct object is definite; thus, if these languages have covert ergative case, they may belong to the object shift type. However, there is reason to believe that the ergative agreement pattern in these languages is more superficial, caused by spell-out decisions at PF. If the Mayan languages have covert ergative case, however, they are a clear counterexample to Bittner and Hale’s hypothesis. Agreement in the Mayan languages follows an almost perfect stereotypic ergative pattern. The Mayan languages are not of the active ergative type because a direct object in syntax is required before the ergative agreement pattern can be used. Moreover, the Mayan languages are also clearly not of the object shift type. Although there is clear evidence that object shift does occur in some of the Mayan languages, the agreement pattern is ergative whether or not object shift occurs. The problem is that the only motivation for positing covert ergative case in the Mayan languages is the assumption that their agreement pattern is like that of languages with overt ergative case. However, if there are no languages with overt ergative case that have this pattern, this motivation disappears. Moreover, the Mayan languages do not behave like ergative languages with overt ergative case in other ways discussed here. Two possible alternative ways to derive the Mayan agreement pattern without positing covert ergative case are presented at the end of section 5.
The final goal of this article is to answer the theoretical question of why object shift should correlate with the use of ergative case. Two possible formal approaches to answering this question are presented in section 6. Both are case locality approaches, formalizing the idea that moving the object out of the VP leads to a change in the conditions for subject case licensing. Bittner and Hale (1996) propose that in languages of the object shift type such as Inuit, VP is a barrier so that objects that remain in the VP are not case competitors with respect to the subject, and thus ergative case cannot be licensed. However, that hypothesis is not compatible with theoretical approaches where case licensing involves only a relationship between a licensing head and a DP. I present an alternative account here that builds on the fact noted in Chomsky 1995 that object shift brings the object into a position where it should interfere with nominative licensing of the subject, creating a violation of the Minimal Link Condition (MLC), unless the verb raises. I show that using ergative case (as a ‘‘last resort’’) is a way to avoid the MLC violation that object shift would otherwise cause. Another ‘‘last resort’’ strategy to avoid an MLC violation in object shift constructions, used in Hawaiian, is to encase the shifted object inside a PP by inserting a preposition. The latter strategy may also be used in some nominative-accusative languages with object shift.
2 Object Shift Ergative Languages
The goal of this section is to provide additional evidence for the existence of a type of ergative language where ergative case is only used when the object moves out of the VP, as proposed by Bittner and Hale (1996).
Niuean fits the object shift type of ergative language perfectly. The subject gets ergative case only when the object moves out of the VP. Moreover, unlike in Inuit there are transitive clauses with a nominative subject, where the object remains in situ and is not marked with oblique case.
2.1.1 Morphological Case in Niuean
Before we examine the relevant Niuean examples, we need to note that the case morphemes in Niuean have a different morphological form for proper and common nouns.
Although Massam (2006) uses the label absolutive to gloss the nominative/accusative cases in her Niuean examples, a focus of her paper is to determine whether the cases labeled absolutive are actually all accusative, as proposed by Bobaljik (1993), or all nominative, as argued by Bittner (1994). Massam concludes that neither proposal is correct, because some of the cases labeled absolutive in Niuean behave like nominatives while others behave like accusatives. This conclusion is consistent with the view of Legate (2008) and Broekhuis and Woolford (2013) that there are ergative languages where some of the cases labeled absolutive have abstract nominative case, while others have abstract accusative case, although the case morpheme used to spell out these two abstract cases at PF looks/is the same. It seems clear that prepositions license accusative case in Niuean, while intransitive subjects get nominative case, as labeled in (8) and (9).
PERF rosy NOMPL cheek
‘The cheeks are rosy.’
(Massam 2006:28, (1b))
The possible identification of the case of objects is discussed below.
2.1.2 Word Order and Case in Niuean
In Niuean, clauses with a specific object have VSO word order, while clauses with a nonspecific object have VOS word order (Massam 2000, 2001).4 This word order difference correlates with a difference in the case of the subject: in clauses with VSO order, the subject is ergative, but in clauses with VOS order, the subject is nominative. The following minimal pair illustrates:
PAST drink coffee NOM Sione
‘Sione drank coffee.’
(Massam 2000:98, (2b))
Massam argues that the nonspecific object remains in its base position inside the VP. The VOS order is produced when the VP containing that object fronts.
In contrast, the specific object moves out of the VP before the VP fronts, producing the VSO word order.
Thus, Niuean is like Inuit and Icelandic (Diesing 1992) in that specific objects undergo object shift (movement out of the VP to a position below Spec,IP). The additional factor affecting word order in Niuean is that the VP fronts in both types of clauses, as has been argued for other verb-initial languages such as Malagasy (e.g., Pearson 1997). Because of this, we can detect the occurrence of object shift by the change in the surface word order in Niuean, unlike in Inuit, and thus we can see the correlation between object shift and ergative case more easily in Niuean than in Inuit.
It is thus clear that Niuean fits the object shift type of ergative language proposed by Bittner and Hale (1996), where ergative case marks the subject only in clauses where the object moves out of the VP.
2.1.3 Seiter 1980 and the Stereotypic Ergative Pattern
Since nonspecific objects are not marked with an oblique case in Niuean, clauses with a nominative subject and a nonspecific object in Niuean are/appear to be transitive, just as clauses with an ergative subject are. We see this in the following minimal pair, where for clarity I have bracketed the VP following Massam’s (2000) analysis:
yesterday PAST [plant shoot taro] ABS I at plantation
‘Yesterday, I planted taro shoots at the plantation.’
(Seiter 1980:69, (184b))
yesterday PAST [plant] ERG I ABSPL shoot taro at plantation
‘Yesterday, I planted taro shoots at the plantation.’
(Seiter 1980:69, (183b))
Seiter (1980) assumes that all ergative languages must fit the stereotypic ergative pattern where only clauses with an ergative subject are transitive, and thus he reasons that the example in (16) with a nominative subject ‘‘is syntactically intransitive at some levels of structure, as shown by the fact that its subject appears in the absolutive case’’ ( p. 70). To make his analysis conform to this assumption, he proposes that the object is incorporated into the verb in examples such as (16), so that the construction with a nominative subject is actually intransitive.
However, Massam (2000) rejects Seiter’s proposal that nonspecific objects are incorporated in Niuean, pointing out that true noun incorporation involves unmodified nouns, whereas the nonspecific object in VOS clauses with a nominative/absolutive subject can be long and complex.
PAST eat chip COMACC fish good NOM Sione
‘Sione ate good fish and chips.’
(Massam 2000:106, (16a))
PAST wash dish dirty slowly NOM Sione
‘Sione washed dirty dishes slowly.’
(Massam 2000:106, (16b))
2.1.4 Massam 2001 and the Type of Niuean
Under standard assumptions, it seems clear at this point that Niuean does not fit the stereotypic ergative pattern: there are transitive clauses such as (18) and (19) that lack an ergative subject. However, as a reviewer points out, under the analysis presented in Massam 2001, Niuean could actually fit the stereotypic ergative pattern. Massam (2001) hypothesizes that nonspecific objects such as those in (18) and (19) have no case at all. If so, and if a transitive clause is narrowly defined as a clause with an object with structural case, examples such as (18) and (19) would not count as transitive because the object has no abstract case, structural or otherwise.
One motive for Massam’s claim that nonspecific objects that remain in situ are caseless is that it is crucial for her explanation of why only specific objects move out of the VP in Niuean. Working under the assumption from Chomsky 1986 that case is licensed only in Spec-head configurations and that objects must move to get case, Massam proposes that specific objects move out of the VP in order to get case, while nonspecific objects remain inside the VP because they do not and cannot get case. However, this account is not available under the more recent version of case theory (e.g., Chomsky 2000), where case is licensed under c-command and objects do not need to move to get case.
Another motive for Massam’s claim that nonspecific objects have no abstract case in Niuean is that they lack an overt case morpheme, whereas all other arguments have one. Massam’s analysis of why the lack of a case morpheme indicates a lack of abstract case is as follows. Since nonspecific objects have no determiner in Niuean, Massam argues that these project only to the NP level, and do not project a DP. This could well be true. Under the assumption that the case morpheme and the abstract case feature are contained in a KP projection above the DP, Massam proposes that without a DP projection, there can be no KP projection above that DP, and without a KP projection, there can be no abstract case. However, it is not clear that these last two assumptions are motivated crosslinguistically. Instead, it may be that nonspecific objects in Niuean do have abstract accusative case, but the accusative case morpheme is not spelled out when the object is adjacent to the verb. Why this should be is not clear, but we do have independent evidence that this happens in other languages. Broadwell (2006) describes a similar phenomenon in Choctaw where the accusative morpheme is optionally dropped when the object immediately precedes the verb, as in (20), but obligatorily spelled out when the object is fronted and thus not adjacent to the verb, as in (21).
In double accusative constructions in Choctaw, the accusative case morpheme on the object adjacent to the verb is optionally spelled out, but the accusative case morpheme on the first object that is not adjacent to the verb must be spelled out.
The one difference between Choctaw and Niuean is that the accusative case morpheme is only optionally missing in the verb-adjacent context in Choctaw, but obligatorily missing in this context in Niuean.
2.1.5 Section Summary
In this section, we have seen that Niuean fits the object shift type of ergative language perfectly: ergative case is used only when the direct object moves out of the VP, not when the direct object remains in situ. Whether Niuean also fits the stereotypic ergative pattern, where all transitive clauses take an ergative subject, depends on whether or not Massam’s (2001) claim that nonspecific objects in Niuean have no abstract case is warranted. If those nonspecific objects merely lack a case morpheme, but have abstract structural case, then Niuean provides clear evidence that Bittner and Hale (1996) are correct in postulating that there is a type of ergative language where ordinary transitivity (the mere presence of a direct object) is not sufficient for ergative case to be used; instead, the position of that object inside or outside the VP also matters.
Whether or not Tagalog is an ergative language has been controversial. However, Aldridge (2012) argues convincingly that Tagalog is indeed ergative, and moreover that object shift occurs in constructions with an ergative subject.
Under Aldridge’s analysis, Tagalog is much like Inuit. The subject is ergative only when there is a definite object, as in (24). In contrast, the subject is absolutive/nominative when there is only an indefinite object, as in (23), and that indefinite object takes an oblique case, just as in Inuit.6
buy ABS woman OBL fish
‘The woman bought a fish.’ (*the fish)
(Aldridge 2012:194, (7a))
buy ERG woman ABS fish
‘The woman bought the fish.’ (*a fish)
(Aldridge 2012:194, (7b))
Aldridge also shows that Tagalog manifests scope differences in the two types of clauses, paralleling what Bittner (1994) describes for Inuit.
read ERG all GEN child ABS many-LINKER book
‘All the children read many books.’
MANY > ALL
(Aldridge 2012:195, (9a))
read ABS all GEN child OBL many-LINKER book
‘All the children read many books.’
ALL > MANY
(Aldridge 2012:195, (9b))
Aldridge (2012:197) argues that the definite object in the ergative subject construction in Tagalog undergoes object shift, while the indefinite object remains in situ. Thus, under Aldridge’s account Tagalog fits the object shift type of ergative language, like Inuit. However, like Inuit, Tagalog can be viewed as also conforming to the stereotypic ergative pattern because the object in the absolutive/nominative subject construction gets an oblique case. If we use the narrowest definition of transitivity, where only a direct object with structural case counts, then all and only transitive clauses have ergative case in Tagalog.
In this section, we will see that under Dixon’s (1972) analysis of the structure of Dyirbal, this language conforms to the object shift type, although it also conforms to the stereotypic ergative pattern.
In contrast, the typical word order in clauses with a nominative/absolutive subject is SVO.
Dixon (1972:150) gives a structure for OSV clauses, which have an ergative subject, in which the object is outside the VP. In contrast, the object remains inside the VP in SVO clauses with a nominative/absolutive subject. Unlike in Inuit and Niuean, however, it is not specific objects that move out of the VP in Dyirbal; instead, it is objects that are topics. According to Dixon (1972:137), ‘‘in Dyirbal, every sentence must contain a topic NP.’’ The object is the topic in clauses with an ergative subject, whereas in clauses with a nominative/absolutive subject, the subject is the topic.
If the above scenario is right, Dyirbal conforms to the type of ergative language proposed by Bittner and Hale (1996), where only clauses in which the object moves out of the VP have an ergative subject. However, because Dyirbal is like Inuit in marking the object that remains in situ with an oblique case, Dyirbal also fits the stereotype of an ergative language where ergative case marks all and only transitive subjects under the ordinary definition of transitivity. So Dyirbal would fit into either type, just as Inuit would.
The point of this section has been to establish that because Dyirbal does conform to one of Bittner and Hale’s two types, it cannot be used to prove the existence of the stereotypic type based on ordinary transitivity.
2.4 Nez Perce
Nez Perce is like Dyirbal in that the topicality of the object determines whether or not the subject gets ergative case. When the object is topical, the subject is ergative. When the object is nontopical, the subject is nominative (Rude 1985, 1988, Deal 2010), as in the following minimal pair:
Both examples in this minimal pair have a direct object. Thus, ordinary transitivity (the mere presence of a direct object) is not sufficient for an ergative subject to be used in Nez Perce. Deal (2010) notes this fact and cites the following additional minimal pair from Crook 1999:
3SG-ERG 3/3-dig-IMPF edible.root-OBJ
‘He digs qeqíit roots.’
(Deal 2010:74, (2a))
3SG 3SUBJ-dig-IMPF edible.root
‘He digs qeqíit roots.’
(Deal 2010:75, (3a))
Although it is not obvious from the rather free word order of Nez Perce that the topical object moves out of the VP, object shift has been proposed in the literature on Nez Perce to explain the difference in the agreement pattern in the topical object construction. As indicated in the glosses of the above examples, topical objects in the ergative subject construction contribute to the verbal agreement ((31)/(33)), but nontopical objects in the nominative subject construction do not ((32)/(34)). This suggests that the object moves out of the VP in the ergative subject construction, as in (31) and (33), but remains in situ in the nominative subject construction, as in (32) and (34) (Woolford 1997, Carnie and Cash 2006), paralleling the situation in French participle agreement where objects that move out of the VP agree, while in-situ objects do not (Kayne 1989).
Another possible indication that the topical object shifts out of the VP in Nez Perce is the fact that the shifted object changes case. The nontopical object in (32) and (34) has a morphologically unmarked case, while the topical object in (31) and (33) takes a morphologically marked case, -ne. This is an example of what is called a marked object or differential object marking (DOM) in the typological literature. Building on Diesing’s (1992) work on specificity and object shift in Icelandic, it has been argued that marked objects have moved out of the VP, whereas unmarked objects remain in situ (e.g., Hindi; Bhatt and Anagnostopoulou 1996).
If the object has moved out of the VP in the Nez Perce examples with an ergative subject, Nez Perce can be added to the list of object shift ergative languages. What we can conclude with confidence at this point is that the presence of a syntactic object, while necessary, is not sufficient for ergative case to be used in Nez Perce. Thus, Nez Perce is not an example of the stereotypic ergative type where ergative case is governed only by ordinary transitivity.
2.5 Section Summary
In this section, we have seen that several ergative languages conform to the object shift type of ergative language proposed by Bittner and Hale (1996), where object shift is necessary before ergative case can be used, although some of these also conform to the stereotypic ergative pattern where ordinary transitivity (the mere presence of a syntactic object, regardless of its position) governs the distribution of ergative case. Thus, so far we have not found an ergative language that is only consistent with the stereotypic ergative pattern.
In the next section, we turn to languages of the active ergative type.
3 Active Ergative Languages
In active ergative languages, ergative case marks all external arguments (aside from splits). Neither the features, position, or even the presence of an object in syntax is relevant to whether or not the subject gets ergative case.8 In this section, we will see evidence that Georgian, Batsbi, Ingush, Udi, Basque, and Kashmiri are of the active ergative type.
Harris (1981:237) notes that the Caucasian language Georgian does not fit the classic typological definition of an ergative language because ergative case does not depend on transitivity in Georgian; she describes the Georgian case pattern as active/inactive. Harris shows that, as under the unaccusative hypothesis, Georgian has two classes of verbs: one has an initial subject (an external argument), and the other does not. Only verbs in the first class can take an ergative subject.9
(Harris 1981:40, (3a))
Vano-ERG he.thought.II.3 Marika.on
‘Vano thought about Marika.’
(Harris 1981:40, (3c))
Verbs in the other class take only a nominative subject, even in the presence of a dative object.
‘Vano began to croon/sing in a low voice.’
(Harris 1981:251, (31a))
child-NOM he.greeted.her.II.2 aunt-DAT
‘The child greeted (his) aunt.’
(Harris 1981:76, (20a))
Unlike the situation in object shift ergative languages, if an object is present in a Georgian clause, its features play no role in determining whether the subject will take ergative case. Indefinite, nonspecific, and property-type objects can occur with ergative subjects in Georgian if the verb is in the first class.
farmer-ERG he.sowed.it.II.1 corn-NOM
‘The farmer sowed corn.’
(Harris 1981:147, (4b))
In addition, although most objects that occur with ergative subjects take nominative/absolutive (unmarked) case, there are a few Georgian verbs that take an ergative-dative case pattern.
To conclude, in Georgian the presence of an object is neither necessary nor sufficient for an ergative subject.
Batsbi is another Northeast Caucasian language that clearly belongs to the active ergative type, although in this language the same verb may or may not take an external argument depending on factors such as volitionality. Holisky (1987) provides minimal pairs showing that volitional subjects (which are presumably external arguments) take ergative case in intransitives, while nonvolitional subjects of the same verbs (which are presumably internal arguments) take a nominative subject.
‘I fell down, on purpose.’
(Holisky 1987:105, (5a))
‘I fell down, by accident.’
(Holisky 1987:105, (5b))
Because no syntactic object is required for an ergative subject, Batsbi is of the active ergative type. As in any ergative language, the presence of an object is not sufficient to cause ergative case to be used on the subject if that subject is not an external argument. Holisky gives the minimal pair (44)–(45), where the subject alternates between dative and ergative case, with a change of meaning suggesting an alternation between an experiencer subject and an external argument. Batsbi marks the subject of the perception verb ‘see’ with dative case, unless the meaning is such that the subject makes a special effort to perform the action; then, the subject is ergative (Holisky 1984:187).
Ingush (Nichols 2008, 2011), a Dagestani language spoken in the Northeast Caucasus, is another example of an active ergative language. It marks all external arguments with ergative case, regardless of whether or not an object is present in the syntax.10
If an external argument is added, it takes ergative case.
When an object is present in an Ingush clause, it need not be definite or specific in order for the subject to be ergative.
As noted at the beginning of this section, languages differ with regard to the exact range of verbs that take an external argument. In Ingush, some verbs that express an action take an external argument even if the subject is inanimate and the action is not purposeful.
The Caucasian language Udi also belongs to the active ergative type, where ergative case marks external arguments regardless of whether an object is present in syntax.
Merab-ERG factory-DAT work-3SG-DO-PRES
‘Merab works in a factory.’
(Harris 2002:81, (23a))
When a syntactic object is present, its properties have no effect on whether or not the subject gets ergative case. Clauses with cognate, indefinite, or property-type objects can take an ergative subject in Udi.
guest-ERG wine.ABS-3SG drink.PRES
‘The guest is drinking wine.’
(Harris 2002:67, (3a))
peasant-ERG here-3SGSOW.PRES seed-OBL-DAT
‘The peasant sows seeds here.’
(Harris 2002:95, (3a))
Unlike in Inuit, there is no contrast in subject marking in minimal pairs with indefinite and definite objects in Udi.
girl-ERG bread.ABS bake
‘The girl is baking bread.’
(Harris 2002:7, (5a))
girl-ERG bread-DAT bake
‘The girl is baking the bread.’
(Harris 2002:7, (5b))
What is interesting about Udi is that there is a contrast in object marking in the above pair of examples, a fact that suggests that there may be object shift in Udi, although it does not affect subject marking. Specific objects are DOM marked with (what looks like) dative case, while nonspecific objects are not. So Udi has marked objects, just like Hindi, and if marked objects move outside the VP, as claimed for Hindi (Bhatt and Anagnostopoulou 1996), then Udi shows us something important: a language can have object shift without belonging to the object shift type of ergative language. In Udi, external arguments are marked with ergative case independent of whether or not object shift takes place.
Basque is another clear example of an active ergative language. Intransitives in Basque divide into two classes, one without an external argument (unaccusatives) and one with an external argument (Levin 1989).11 Verbs that do not take an external argument mark their subjects with nominative case.
storm-DET.NOM arise-PERF is
‘A storm arose.’
(Laka 2006:376, (4c))
Verbs that take an external argument mark their subjects with ergative case.12
sun-DET-ERG shine-IMPF has
‘The sun shines.’
(Laka 2006:380, (9b))
Verbs that take an external argument always mark that argument with ergative case in Basque, regardless of whether or not an object is present in syntax.
Klara-ERG well ski-IMPF has
‘Klara skis well.’
(Laka 2006:379, (8a))
As in other active ergative languages, the only notion of transitivity that is relevant for ergativity in Basque is that of Hale and Keyser 1993, where any verb with an external argument counts as transitive. The presence of an object in syntax is not necessary.
Kashmiri is an active ergative language, although it has an aspect split so that ergative case is restricted to the perfective aspect (Wali and Koul 1997). Within the perfective aspect, Kashmiri marks all external arguments with ergative case, regardless of transitivity in the ordinary sense of the term. No object need be present in syntax for a subject to be ergative in Kashmiri.
(Wali and Koul 1997:250, (9a))
I.ERG wrote board.DAT on chalk.ABL.MASC.SG with
‘I wrote on the board with a piece of chalk.’
(Wali and Koul 1997:162, (4b))
Clauses with an external argument and only an indefinite or property-type object also take ergative subjects.
I.ERG learn wrestling
‘I learned wrestling.’
(Wali and Koul 1997:216, (10a))
I.ERG broke.MASC.SG walnut stone-ABL without
‘I broke a walnut without using a stone.’
(Wali and Koul 1997:162, (5a))
In contrast, clauses with no external argument, where the subject is a theme/patient (internal argument), are not marked with ergative case in Kashmiri.
‘The glass broke.’
(Wali and Koul 1997:152, (3c))
Although I will not discuss them specifically in this article, many of the languages related to Kashmiri also fall into the active ergative type.14
3.7 Section Summary
In this section, we have seen various languages that are examples of the active ergative type. In these languages, all external arguments are marked with ergative case (except in languages with an aspect split, where this holds true only in the perfective aspect). Although a syntactic object is not necessary for ergative case in these languages, clauses with an ergative subject nevertheless qualify as transitive under Hale and Keyser’s (1993) proposal that all verbs with an external argument have at least a semantically incorporated object. Given that an object in syntax is not necessary for ergative case to be used in the active ergative type, it is not surprising that when an object is present, neither its position nor its features are relevant to whether or not the subject gets ergative case.
4 As Yet Unclassified Ergative Languages
In this section, we look at various ergative languages whose type is not yet entirely clear, but for which a syntactic object is necessary before ergative case can be used. What is not yet clear is whether that object is always outside the VP in ergative subject constructions in these languages (so that the language would belong to the object shift ergative type) or whether the position of the object is irrelevant (so that the ordinary notion of transitivity is all that governs the distribution of ergative case). The general problem in classifying the languages discussed in this section is that the word order alone, often SOV or free order, does not tell us where the object is located. Further syntactic research is needed before we can determine which type these languages belong to. Although readers may be tempted to simply assume that these languages use only ordinary transitivity to govern ergativity, that is not one of the two types of ergative languages posited by Bittner and Hale (1996), nor did the research for this article identify even one clear example of that commonly assumed type.
Archi is a Caucasian language, but unlike the Caucasian languages discussed in section 3, Archi does not belong to the active ergative type. External arguments do not get ergative case in clauses without a syntactic object; instead, they take nominative/absolutive case.
(Kibrik 1979:67, (2a))
father.NOM lying AUX
‘Father is lying.’
(Kibrik 1979:67, (2b))
In contrast, the subject is ergative in examples such as (74), where a syntactic object is present in the clause.
father-ERG bread.NOM bake AUX
‘Father is baking the bread.’
(Kibrik 1979:67, (2c))
However, the presence of an object is not sufficient for the subject to be ergative in Archi. In the following example, an object is present, and yet the subject is not ergative:
What remains to be determined is whether the object in ergative subject constructions in Archi has moved out of the VP. It is not universal that only definite or specific objects undergo object shift; for example, we saw above that topicality in some sense is associated with object shift in Nez Perce and Dyirbal. If future research should show that the object in examples such as (74) has moved outside the VP, then Archi would belong to the object shift type of ergative language.
Kabardian is another Caucasian language that clearly does not belong to the active ergative type. External arguments do not get ergative case when no object is present in syntax, even with verbs such as ‘dance’ that usually take an external argument.
‘The girl danced on the table.’
(Colarusso 1992:54, (90g))
‘The men are looking.’
(Colarusso 1992:75, (135a))
Nevertheless, Kabardian is an ergative language (Colarusso 1992:52), although Colarusso glosses the ergative case as ‘oblique’ because the ergative and dative morphological cases look alike in Kabardian.15
horse+man-OBL man-ABS 3-3-PRES-kill
‘The horseman is killing the man.’
(Colarusso 1992:85, (158c))
man-OBL girl-ABS 3(ABS)-3(OBL)-NONPRES-see-PAST-AFF
‘The man saw the girl.’
(Colarusso 1992:53, (90a))
The question now is whether ordinary transitivity is sufficient for ergative case to be used in Kabardian, or whether Kabardian belongs to the object shift type of ergative language. The direct objects in (79) and (80) are definite/specific, and specific objects would undergo object shift in Inuit. However, we cannot tell from the SOV word order of Kabardian whether the object has moved out of the VP. As in Inuit, there are clauses in Kabardian with a nominative-oblique case pattern.
girl-ABS shirt-OBL 3-3-DAT-sew-INTR-PAST-AFF
‘The girl was sewing at the shirt.’ (action incomplete)
(Colarusso 1992:54, (90h))
horse+man-ABS horn-OBL 3-3-DIR-DAT-drink+DAT-FUT-DEF-AFF
‘The horseman will drink the cup (drinking horn).’
(Colarusso 1992:178, (294e))
In Inuit, this case pattern occurs when the direct object remains inside the VP; however, we cannot simply assume that Kabardian is like Inuit in this respect. Thus, at this point which notion of ‘‘transitivity’’ is relevant for the distribution of ergative case in Kabardian remains an open question.
In Tsez, all intransitive constructions mark the subject with nominative/absolutive case, even those with external arguments. Verbs with an external argument such as ‘eat’ do not mark their subjects with ergative case unless a syntactic object is present, as the following contrast shows:16
However, it appears that the object need not be specific in order for there to be an ergative subject in Tsez.
girl.ERG water.NOM bring.FUT
‘The girl will bring water.’
Nevertheless, an oblique object is not sufficient for ergative case to be used; clauses with only an oblique object take a nominative subject.
horse.NOM fence.TRANS (on) jump.PAST/EVID
‘The horse jumped over the fence.’
It is thus clear that the presence of a direct object in syntax is necessary before ergative case can be used in Tsez, and thus Tsez is not an active ergative language. However, to determine whether it is only ordinary transitivity that determines the distribution of ergative case in Tsez, or whether Tsez belongs to the object shift ergative type, we need to know whether the object is inside or outside the VP in the examples with an ergative subject. Until that is determined, the question of the type of ergative language Tsez belongs to remains open.
We now turn to an Australian language, Kalkatungu (Blake 1979, 1982, 1983, 1994). We saw earlier that, as with the Caucasian language family, some Australian languages belong to the object shift type (Dyirbal) and some to the active ergative type (Warlpiri). Kalkatungu is clearly not an active ergative language, but whether it belongs to the object shift type remains an open question at this point.
We can tell that Kalkatungu is not an active ergative language because external arguments do not take ergative case in Kalkatungu unless an object is present in syntax. We see this in the following examples, where the verb ‘wash’ takes a nominative subject when no object is present in syntax (even though a reflexive object is implied), but an ergative subject when an object is present:
woman.NOM wash-REFL-FUT soap-INSTR
‘The woman will wash with soap.’
(Blake 1994:50, (2b))
woman-ERG wash-FUT child.NOM soap-INSTR
‘The woman will wash the child with soap.’
(Blake 1994:50, (2a))
One indication that Kalkatungu might belong to the object shift type of ergative language is that there are pairs of examples where the subject is ergative when the object is specific, but nominative when the object is nonspecific.
mother.ERG food-NOM cook
‘Mother cooks the food.’
(Blake 1979:7, (1.4))
mother.NOM food-DAT cook
‘Mother cooks food.’
(Blake 1979:7, (1.5))
This pattern is very like what we saw in Inuit, which is an object shift ergative language. However, the word order does not provide much information about the position of the object because word order in Kalkatungu is rather free (Blake 1994:140) and we lack the scope information that enabled Bittner (1994) to determine that the specific object moves out of the VP in Inuit. Thus, at this point we cannot determine for sure whether Kalkatungu is an object shift ergative language, like Inuit.
4.5 Section Summary
In this section, we have seen several ergative languages that require a direct object before ergative case can be used, but whose ergative type is not yet certain. Although it is clear that none of these languages are of the active ergative type, we cannot yet determine whether any or all of them conform to the other type proposed by Bittner and Hale (1996), called here the object shift type. To do that, we would need evidence showing whether the object in examples with an ergative subject is inside or outside the VP in these languages. What we can say at this point is that no conclusive evidence has yet been found to disprove Bittner and Hale’s hypothesis, although some of the languages discussed in this section might turn out to be counterexamples.
5 Ergative Agreement without Ergative Case
Some languages are typologically categorized as ergative only on the basis of their agreement pattern, in the absence of any morphological case on arguments. It is often assumed that such languages must have covert ergative case. If this assumption is correct, then these languages are relevant to the question of what definition(s) of transitivity govern the distribution of ergative case. Even if these languages do not have covert ergative case, we can still ask what notion of transitivity is relevant for their agreement patterns.
In this section, I discuss languages from two families, South Sulawesi and Mayan. Both have ergative agreement patterns but no morphological ergative case on arguments. Neither conforms to the active ergative type. The particular South Sulawesi language discussed here, Selayarese, might belong to the object shift type because the agreement pattern is ergative only when the direct object is definite. In contrast, the Mayan languages do not fit into either of the types of ergative languages postulated by Bittner and Hale (1996). Therefore, if the Mayan languages actually have covert ergative case, they are a counterexample to the hypothesis that all languages with ergative case belong to one of these two types. On the other hand, if Bittner and Hale’s hypothesis turns out to hold for all languages with overt ergative case, then the Mayan languages are unlike any known language with overt ergative case and thus the motivation for positing covert ergative case in Mayan evaporates.
In this section, I will show that these languages differ from languages with ergative case in several ways. The only motivation for positing covert ergative case in these two language families is the assumption that there is no other way to generate the agreement pattern in these languages. This motivation disappears if the theory can generate the Mayan agreement pattern(s) without positing covert ergative case, and in this section, I will discuss two possible alternative means of doing so. In one, the pattern is formed in syntax; in the other, it is formed at PF as a result of what is and is not spelled out.
5.1 South Sulawesi Languages
Selayarese, like the other South Sulawesi languages, has no morphological case marking. Transitive clauses with a definite object have a very ordinary-looking cross-referencing pattern like that found in many nominative-accusative languages: an agreement prefix on the verb cross-references the subject, while the definite object is cross-referenced with a classic second position pronominal clitic.17
2.AGR-cook=3.CL fish-DEF in pan
‘You cooked the fish in the pan.’
(Finer 1999:142, (11a))
in pan=3.CL 2.AGR-cook fish-DEF
‘In the pan you cooked the fish.’
(Finer 1999:142, (11b))
As in many languages, pronominal clitics can only cross-reference definite objects.
What makes the Selayarese pattern qualify as typologically ergative is the fact that intransitive subjects are cross-referenced with a pronominal clitic instead of the agreement prefix.
(Finer 1994:158, (7d))
If Selayarese has covert ergative case, it may fit the object shift type because transitive clauses with an indefinite object have the same cross-referencing pattern as intransitives.
INTR-put=2.CL money in cupboard
‘You put money in a cupboard.’
(Basri and Finer 1987:145, (6a))
Yet none of the South Sulawesi languages have overt ergative case, and moreover, the ergative agreement pattern of Selayarese breaks down somewhat in related languages in a way that suggests that the agreement pattern is determined at PF. Konjo is not as strict as Selayarese about limiting clitics to second position, and when there are legal attachment sites for two pronominal clitics, Konjo can use a pronominal clitic for both the object and the subject. When the verb is in initial position, the Konjo pattern looks like that of Selayarese, as in (95), but when something fronts to which a pronominal clitic can attach, the Konjo cross-referencing pattern changes: a (nominative) clitic attaches to the fronted element, while an (accusative) clitic attaches to the verb, as in (96).18
‘I saw you there.’
(Friberg 1996:148, (37a))
‘There, I saw you.’
(Friberg 1996:148, (37c))
In my view, the ergative agreement pattern of the languages in this family is only superficial and does not reflect the presence of covert ergative case. Instead, it is produced at PF by selectively spelling out cross-referencing elements ‘‘overgenerated’’ in syntax.
I argue that in syntax, languages can generate both the agreement associated with Infl/T (associated with nominatives) and a pronominal clitic cross-referencing the nominative. Independent motivation that nominative arguments can be cross-referenced by both agreement and a nominative clitic in syntax comes from the fact that both of these elements are spelled out at PF in Kashmiri. The agreement affix suffixes to the verb or auxiliary verb, and any pronominal clitics follow it.
you.ERG saw-AGR.MASC.SG-CL.2SG.ERG-CL.1SG.NOM 1.NOM.MASC
‘You saw me(masculine).’
(Wali and Koul 1994:972, (4b))
1.NOM.MASC be-AGR.MASC.SG-CL.1SG.NOM go.PRES.PART
‘I(masculine) am going.’
(Wali and Koul 1997:152, (3a))
Normally languages do not spell out both agreement and a nominative pronominal clitic. In familiar languages, it is usually the agreement affix that is spelled out. However, the potential exists for a language to make the opposite choice. Spelling out a nominative clitic instead of the agreement affix can produce a surface ergative agreement pattern, but only if an additional factor is present: ‘‘slot’’ competition. In this situation, a nominative clitic can be spelled out in intransitives, but if pronominal clitics compete for a single ‘‘slot’’ at PF, transitive clauses cannot spell out two pronominal clitics. The alternative is to spell out the agreement affix instead of the nominative pronominal clitic. Essentially, spelling out agreement is a ‘‘last resort’’ in these languages. This spell-out pattern is shown in (99)–(101), where the overstriking indicates an element present in syntax that is not spelled out at PF.
The spell-out pattern can change if there are two different attachment sites for pronominal clitics so that ‘‘slot’’ competition among the pronominal clitics is removed, as we saw in the Konjo example (96), repeated here.
‘There, I saw you.’
(Friberg 1996:148, (37c))
Thus, we see that there is no need to posit covert ergative case to generate this kind of surface ergative agreement pattern. In the next section, we turn to the Mayan languages, which have a similar ergative agreement pattern.
Because the Mayan languages do not mark case on arguments, descriptive work on these languages is traditionally neutral concerning the identity of the cross-referencing morphemes, labeling them simply Set A and Set B. However, because the agreement pattern is typologically ergative, these series are often labeled ‘‘ergative’’ and ‘‘absolutive,’’ respectively. Set A cross-references only subjects, and usually only transitive subjects. Set B cross-references objects, and usually intransitive subjects.
‘I see you.’
(Craig 1977:90n1, (c))
(Craig 1977:90n1, (a))
We can tell that the Mayan languages are not of the active ergative type because all verbs, even those that take an external argument, such as ‘eat’, cross-reference intransitive subjects with the Set B ‘‘absolutive’’ marker.
COMPL-SETB.2 eat-ANTIPASS you DEM
(Mateo-Toledo 2008:48n2, (ii))
Yet the Mayan languages do not behave like ergative languages of the object shift type either. In the Mayan languages that do have object shift, there is no associated change in the agreement pattern. We see this in Chol, where there is a VSO/VOS word order alternation that correlates with the definiteness of the object (Coon 2010b). The word order is VSO with a definite object, but VOS with an indefinite object, and Coon (2010b) argues that the syntax is just as Massam (2000) describes for Niuean (see section 2.1 above), where definite objects shift out of the VP, but indefinite objects remain in situ. However, there is no associated change in the agreement pattern: if the Set A agreement marks arguments with covert ergative case, then the subject is ergative in Chol regardless of whether or not object shift has occurred.
PERF SETA-carry-TRANSDET-Maria DET wood
‘Maria carried the wood.’
(Coon 2010b:355, (2a))
PERF SETA-carry-TRANS wood DET-Maria
‘Maria carried wood.’
(Coon 2010b:355, (1a))
Thus, Chol would fit into neither of Bittner and Hale’s two types, and would be a counterexample to the hypothesis that all languages with ergative case must fall into one of these two types. The problem is that if there is no language with overt ergative case that has the Chol pattern, the motivation for positing covert ergative case in Chol in order to account for its agreement pattern evaporates.
5.2.1 Reasons to Doubt That Mayan Languages Have Covert Ergative Case
The only motivation for positing covert ergative case in the Mayan languages is the assumption that this is necessary in order to generate the observed agreement pattern. None of the Mayan languages have overt ergative case; moreover, these languages are unlike languages with overt ergative case in two ways, involving association of cross-referencing morphemes and case, and aspect splits.
188.8.131.52 Backward Association of Cross-Referencing Morphemes and Case
In languages with overt ergative case, the association between the kind of cross-referencing element and the case of an argument is typically as in (108). Nominative arguments are cross-referenced with agreement, while arguments with ergative case, if they are cross-referenced at all, are typically cross-referenced with pronominal clitics.19 We saw this pattern above in Kashmiri and it is characteristic of Basque as well (Arregi and Nevins 2012).
In Mayan, the Set A forms are agreement while the Set B forms are pronominal clitics (Coon 2010b:373).20 If the Mayan languages have covert ergative case, we would expect to see the association pattern in (108), as shown in (109) with reference to what Set A and Set B forms would be expected to cross-reference.
What we actually see in Mayan, however, is the reverse of this pattern.
184.108.40.206 Aspect Splits
Another way in which the Mayan languages are unlike languages with overt ergative case involves the nature of its aspect splits. In languages with ergative case that have an aspect split, the ergative case is used in the perfective aspect, but not in the imperfective aspect (DeLancey 1981:630, Dixon 1994:99). We see this typical ergative split in the following Hindi/Urdu examples:
Thus, if the Mayan languages had covert ergative case, we would expect any ergative aspect split to match the above pattern. Some of the Mayan languages do have an aspect split, but as noted by Krämer and Wunderlich (1999) with respect to Yucatec Maya, if these languages have covert ergative case, the pattern of their aspect splits is the reverse of what occurs in languages with overt ergative case. The Set A series is used in the imperfective aspect, as in (113), while the Set B series is used in the perfective aspect, as in (114).
Thus, if the Set A series is identified as ergative, then ergative is restricted to the imperfective, rather than the perfective as in other ergative languages with an aspect split.
There is a second way in which the aspect split in Mayan languages is not like aspect splits found in ergative languages. Aspect splits in Mayan are limited to intransitive clauses, as in the examples above. There is no aspect split in transitive clauses, as we see in the Yucatec Maya examples (115) and (116); the agreement pattern is the same in both the perfective and the imperfective aspect.
I would argue that the Mayan pattern is not an ergative split at all; rather, it is a superficial agreement ‘‘split,’’ made possible by the fact that pronominal clitics are generated in syntax for both subject and object in Mayan languages, just as posited for Selayarese above. We do not see this ‘‘split’’ in transitives because, as in Selayarese, only one pronominal clitic can be spelled out. If the object is to be cross-referenced, it must be with that one pronominal clitic ‘‘slot,’’ because objects cannot control the agreement. Since the subject clitic cannot also be spelled out (given the limit of one clitic), the subject agreement is spelled out instead. It is only in intransitives that there is a choice of whether to spell out agreement or the nominative pronominal clitic for the intransitive subject. This choice can be manipulated by the grammar to produce an aspect split, wherein each form parasitically/redundantly marks the aspect of the clause.21
The idea that there is a choice of which subject-cross-referencing element generated in syntax to spell out at PF, agreement or a (nominative) pronominal clitic, is further supported by the fact that in the related Mayan language Chontal, there is a cross-cutting split between positive and negative intransitives (Knowles-Berry 1987). Set A is spelled out only in positive imperfective clauses, while Set B is spelled out in any negative or perfective clause.22
To summarize this section: the only reason to posit covert ergative case in the Mayan languages is the assumption that this is necessary to generate the surface agreement pattern. In this section, we have seen two ways in which the Mayan languages would not conform to known ergative languages if they do have covert ergative case. In the next section, we will consider how the surface agreement pattern in Mayan languages could be generated without positing covert ergative case.
5.2.2 Deriving the Mayan Agreement Pattern without Positing Covert Ergative Case
If the Mayan languages do not have covert ergative case, how is the surface ergative agreement pattern generated? I would argue that a PF approach like that described above for Selayarese may be right for the Mayan languages as well. However, I will also discuss a possible alternative, syntactic approach, using a modified version of Bobaljik’s (1993) proposal that was designed to generate an ergative pattern using only elements present in a nominative-accusative language.
220.127.116.11 A PF Approach to the Mayan Agreement Pattern
As described above for the South Sulawesi language family, there is some evidence in the Mayan languages for the idea that syntax overgenerates cross-referencing elements but PF spells out only a subset of them. Specifically, the idea is that syntax generates two cross-referencing elements for subjects—agreement (Set A) and a (nominative) pronominal clitic (Set B)—and which of these to spell out in any given construction is determined at PF. There is a general preference for spelling out the nominative pronominal clitic (Set B), rather than the agreement affix (Set A), but PF limits the number of pronominal clitics (Set B forms) to one per clause. Thus, in transitive clauses where the object pronominal clitic is spelled out, the subject pronominal clitic cannot also be spelled out, so agreement is spelled out instead.
Although we do not usually see a Set B form (a pronominal clitic) used for a transitive subject in Mayan (for the reason just stated), there is evidence that it is nonetheless generated in syntax, because it can be spelled out in at least one circumstance. Many languages manifest what has been described as an ‘‘antiagreement’’ effect whereby agreement is blocked or altered when the subject extracts. Richards (2001) discusses a wide range of languages with this antiagreement effect, including some Mayan languages. He gives the examples in (120) and (121) from Berber. When the subject is extracted in Berber, the normal subject agreement is blocked/ungrammatical, as in (121), and the grammatical form (120) has no subject agreement.
Richards lists K’ichee’ as one of the Mayan languages that manifests this antiagreement effect. Hale and Storto (1997) discuss the relevant construction in K’ichee’, which they refer to as the agent focus construction. In this construction, the Set A agreement that normally cross-references transitive subjects is ‘‘suppressed,’’ and the Set B form can be used (spelled out) to cross-reference the transitive subject.
Q you.SGASP-2SG.SETB-cure-AF the man
‘Was it you who cured the man?’
(Hale and Storto 1997:3, (3b))
Now, under the hypothesis that syntax generates both a nominative pronominal clitic (Set B form) and agreement (Set A), both cross-referencing the nominative subject, it is not too surprising to see that when the agreement is blocked (by whatever causes the antiagreement effect), the option is there to spell out the nominative pronominal clitic instead, as in (122). Yet the object is not demoted in this construction, and an accusative pronominal clitic is generated for it in syntax.23 The two pronominal clitics generated in syntax compete for the one ‘‘slot’’ at PF. In (122), the second person subject won the one available slot, while in (123), the second person object wins.
QFOC the man ASP-2SG.SETB-cure-AF ( you.SG)
‘Was it the man who cured you?’
(Hale and Storto 1997:3, (3a))
Now some readers may be of the opinion that agreement is determined entirely in syntax, and the next section outlines a possible syntactic approach along the lines proposed by Bobaljik (1993).
18.104.22.168 A Possible Syntactic Approach Building on Bobaljik 1993
Before it was known that ergative is the inherent case associated with external arguments, Bobaljik (1993) put forth a clever proposal for generating a case pattern that qualifies as typologically ergative, yet uses only nominative and accusative case. The idea was that in some languages, the lower head that licenses accusative case is always ‘‘active’’ (in that it must assign its case), and the upper head that licenses accusative case is used only if there are two arguments. As a result, intransitive subjects would be marked with accusative case instead of nominative (and thus marked like objects), but transitive subjects would be marked differently, with nominative case.
We might modify this proposal so that it would only affect the agreement pattern and not the case by postulating that in some languages, although Infl/T still values nominative case on the subject, the φ-features of Infl/T are only valued by the nominative argument as a ‘‘last resort.’’ The preference would be to use a pronominal clitic to match features with the nominative argument. Thus, intransitive subjects would be cross-referenced with a pronominal clitic rather than with agreement.
We could then add to this a syntactic means of prohibiting more than one pronominal clitic per clause. One idea would essentially involve a defective intervention effect wherein a closer pronominal clitic blocks a further pronominal clitic from establishing a chain with a lower argument. In the configuration in (124), the lower clitic (Clitic 2) would intervene between the higher clitic (Clitic 1) and its argument (Argument 1).
(124) Clitic 1 Clitic 2 Argument 1 Argument 2
A nested arrangement would not improve things. Here, the lower clitic (Clitic 1) would block the higher clitic (Clitic 2) from reaching its argument (Argument 2), as in (125).
(125) Clitic 2 Clitic 1 Argument 1 Argument 2
With the resulting restriction of one pronominal clitic generated per clause, one would have to use agreement to cross-reference the nominative subject (as a ‘‘last resort’’).
(126) agreement clitic nominative DP accusative DP
This approach could be extended to the agent focus construction in Mayan languages described in Hale and Storto 1997. In antiagreement situations where agreement is suppressed, only one of the two arguments in a transitive clause could be cross-referenced with the one allowed pronominal clitic. Assuming that the syntax could generate either a nominative or an accusative clitic, then the K’ichee’ patterns discussed above could be generated.
In the next section, I return to the discussion of languages with overt ergative case, and the question of why there should be a type of ergative language wherein ergative case is not used unless the object moves out of the VP.
6 Theories of the Association between Object Shift and Ergative Case
In this section, I turn to the interesting theoretical question of why there is an association between object shift and ergative case in the object shift type of ergative language. I will describe two possible approaches to this question. One is from Bittner and Hale 1996. The other is an updated version of a proposal in Woolford 2007. Both approaches begin with the same basic idea:
(127) Moving an object outside the VP puts that object in a syntactic position that potentially changes the conditions for case-licensing the subject.
These approaches described below develop this same leading idea in two very different ways.
6.1 Bittner and Hale 1996
Bittner and Hale (1996) assume that ergative case is a dependent case that can only be licensed in the presence of another argument that serves as a case competitor. To maintain this assumption even in active ergative languages, where there is not necessarily an object present in syntax, they adopt the idea in Hale and Keyser 1993 that verbs with an external argument always have at least a semantically incorporated object, and that this kind of object can serve as a case competitor. To account for why ergative case is restricted to object shift constructions in languages such as Inuit, Bittner and Hale postulate that in these languages, VP is a barrier blocking an object inside the VP from serving as a case competitor for a subject outside the VP. Without a local case competitor, ergative case cannot be licensed under their approach, and thus clauses without object shift cannot have an ergative subject in languages where VP is a barrier. The object would need to move out of the VP to get close enough to the subject to be a case competitor.
In this approach, the problem is with the licensing of ergative case, and object shift mitigates the problem. In contrast, in my proposal below, the problem is with the licensing of nominative case, and object shift causes the problem.
6.2 Object Shift, Ergative Case, and the Minimal Link Condition
In this section, I propose a rather different kind of approach to the question of why ergative case is used only in object shift constructions in some languages. This approach does not assume that ergative is a dependent case; rather, it takes the now more standard view that all case is licensed by heads and that ergative case is licensed by little v. The leading idea is that object shift places the object in a position where it can interfere with licensing nominative case on that external argument. The solution to this problem is not to license nominative case on the external argument, but instead to use ergative case as a ‘‘last resort’’ to avoid the problem. Below, I present a formalization of this leading idea, but I would not want the merits of the basic idea to be judged by the details of this particular formalization.24
The beginning point of this approach is a prediction of the Minimal Link Condition (MLC) for clauses with object shift in Chomsky 1995. Because object shift places the object in a position between Infl and the base position of the subject, that closer object should block the probegoal relation between Infl and the base position of the subject/external argument, thus blocking nominative case licensing on that external argument.25
(128) [Infl Obj [ExtArg V V
To account for the fact that clauses with object shift are nonetheless grammatical in Icelandic, Chomsky (1995) proposes that the fact that the verb raises above the shifted object in Icelandic extends the lower phase so that the shifted object remains in the lower phase. This makes the subject and the object equidistant from Infl, and there is no violation of the MLC if Infl probes the external argument and licenses it with nominative case.26 In contrast, Holmberg (1999) proposes a very different reason for the fact that object shift does not cause the expected MLC violation: namely, that object shift does not occur until PF. These two proposals make radically different predictions for crosslinguistic variation: if object shift occurs in syntax, we expect to find that it does block nominative licensing of the external argument in some languages; in contrast, if object shift occurs at PF, we should not find any language where it blocks nominative licensing of the subject/external argument. I argue here that the first prediction is correct: there are many languages where object shift blocks the ability of Infl to license nominative case on the subject, and this is why we so often see deviations from the nominative-accusative case pattern in object shift constructions crosslinguistically.
In the configuration in (128) where the shifted object is closer to Infl than the external argument is, when Infl probes down, it can only value nominative case on that closest argument, the shifted object.27
(129) [Infl Obj-nom [ExtArg V V t]]
If Infl probed past the closer object to license nominative case on the external argument, it would produce an MLC violation. The configuration in (129) is grammatical if the external argument gets ergative case in its base position from little v, as in (130).
(130) [Infl Obj-nom [ExtArg-ergV V t]]
This is, in essence, a ‘‘last resort’’ use of ergative case in languages that normally assign nominative case to external arguments, whenever that can be done without incurring an MLC violation.28
There is also a second way that languages can avoid an MLC violation in object shift constructions. If the shifted object is encased inside a PP, by means of P-insertion, this PP will not block Infl from probing past it to license nominative case on the external argument.
(131) [Infl [PP P Obj] [ExtArg-nom V V t]]
A language that uses this P-insertion strategy is Hawaiian. Medeiros (2013) shows that Hawaiian has a syntax much like that proposed for Niuean by Massam (2000), discussed above in section 2.1. As in Niuean, clauses with a nonspecific object have a VOS word order, produced by fronting the VP with the object remaining inside it, as in (132).
drink coffee cold Noelani
‘Noelani is drinking cold coffee.’
(Medeiros 2013:77, (10b))
In contrast, specific objects move out of the VP before it fronts, producing a VSO word order, just as in Niuean. However, unlike the subject in Niuean, the subject does not get ergative case in the object shift construction in Hawaiian; instead, a preposition is inserted preceding the shifted object, as in (133).29
drink DIR Noelani PREP the coffee cold
‘Noelani is drinking the cold coffee.’
(Medeiros 2013:77, (10a))
With the shifted object safely encased in a PP and thus rendered inert as an intervener, Infl can probe the subject/external argument in its base position and license nominative case on it, as in (131), without incurring an MLC violation.
This raises the question of whether this P-insertion solution might also be used in other nominative-accusative languages with object shift. The answer appears to be yes. It is known that many languages ‘‘mark’’ specific objects, but leave nonspecific objects unmarked, a phenomenon known as differential object marking (DOM) (e.g., Aissen 2003). It is also known that in many such languages, the ‘‘marked’’ objects have undergone object shift, as argued for Turkish by Diesing (1992). If the DOM marker is an inserted preposition/postposition,30 the fact that it appears on shifted objects in some languages makes sense. As in Hawaiian, encasing the shifted object in a PP would render it ‘‘inert’’ as an intervener between Infl and the base position of the subject, thus allowing nominative licensing of the subject in object shift constructions without violating the MLC.31
This article has provided additional support for Bittner and Hale’s (1996) claim that there is a type of ergative language where ergative case cannot be used unless the object moves out of the VP. Strong evidence comes from Massam’s (2000) work on Niuean, where word order confirms the association of object shift and ergative case, and Aldridge’s (2012) similar analysis of the related language Tagalog. Other languages discussed here that belong to this object shift type are Dyirbal and Nez Perce.
Bittner and Hale (1996) propose that ergative languages fall into (just) two types, the object shift type mentioned just above, and the type where all external arguments get ergative case (known as the active or active ergative type). They do not discuss what is often considered the prototypic ergative language, where the presence of a direct object in syntax (regardless of its position) is both necessary and sufficient for ergative case to be used. This article raised the question of whether this prototypic/stereotypic ergative pattern ever actually occurs in languages with ergative case. A survey of ergative languages discussed here failed to turn up even one really clear example of an ergative language that, given the evidence available so far, cannot belong to one of Bittner and Hale’s two types. However, there remain a number of ergative languages whose type the available literature does not allow us to determine. These languages do limit ergative case to clauses with an object present in syntax, but there are not yet sufficient data to determine whether the object has always shifted out of the VP in ergative constructions in those languages. Thus, whether Bittner and Hale’s two types of ergative languages are the only types that occur remains an open question.
With respect to theory, the active type of ergative language is expected under the view that ergative is an inherent case licensed in connection with the θ-licensing of the external argument by little v (e.g., Laughren 1989, Mahajan 1989, Woolford 1993, 1997, 2006b, Massam 1994, 1998, and additional references cited in Woolford 2006b). However, the association of ergative case and object shift in the object shift type initially seems unexpected. Bittner and Hale (1996) propose an account under which VP is a barrier in some languages, so that an object inside VP is not sufficiently close to serve as a case competitor for the subject outside VP, under the assumption that ergative is a dependent case. An alternative account is presented here under which object shift (without verb raising) creates a Minimal Link Condition violation if Infl tries to probe the external argument across the closer shifted object, as suggested in Chomsky 1995. Ergative case is licensed on the external argument as a ‘‘last resort’’ in object shift constructions in some languages to avoid this violation of the MLC. An alternative means of avoiding an MLC violation discussed in this article is P-insertion on the object, a strategy used in Hawaiian that may be used more generally in many languages with differential object marking.
I would like to thank Eric Potsdam, the audience at University College London, and the LI reviewers for helpful comments on this article. I would also like to express my gratitude to the three University of Massachusetts undergraduate interns who worked with me on the research for the article: Tyler Forni, Glynis MacMillan, and Abril Navarro.
2Bittner and Hale (1996) call the first type transparent and the second type opaque, terms that stem from their particular account of the difference between these two types (summarized in section 6 below). They also refer to the first type as morphologically ergative and the second type as syntactically ergative. While this is accurate in the sense that the second type requires a particular syntax (object shift), it remains to be determined whether the languages they label as syntactically ergative are actually the same set as those labeled syntactically ergative in the typological literature (languages having grammatical processes that single out transitive subjects). Here I use more neutral descriptive terms for these types.
3 For those working in other frameworks, the term external argument is equivalent to what has been called a deep or initial subject (e.g., Lexical-Functional Grammar) or an initial 1 (e.g., Relational Grammar). In all of these frameworks, the verbs of a language fall into two classes, one that takes an external argument/initial subject/initial 1 and one that does not. Although the first class is generally more active/agentive, the membership in these two classes differs somewhat crosslinguistically. In the Minimalist Program, external arguments are generated and θ-marked in the specifier position of the highest little v node (Chomsky 1995, Kratzer 1996). Ergative case is the inherent case associated with the θ-marking of the external argument (Laughren 1989, Mahajan 1989, Woolford 1993, 1997, 2006b, Massam 1994, 1998, and additional references cited in Woolford 2006b).
5 Note that, as shown in (7), the ergative case morpheme for proper nouns, e, looks just like the absolutive case morpheme for common nouns, e. It is difficult to determine what the case of the object in an ergative subject construction is in Niuean, given that it is nominative in some languages, but accusative in other languages (Legate 2006, 2008, Woolford 2007). I will label these nominative here since this pattern is more familiar to readers and the label is not crucial for the present discussion.
6Aldridge (2012) notes that the ergative case written as ng in the examples is actually pronounced as nang.
7Rude (1988) glosses the morphologically marked object case in the ergative subject construction simply as OB.
8 Clauses with ergative case in active ergative languages are classified as transitive by Hale and Keyser (1993) under their view that all verbs with an external argument have at least a semantically incorporated object. In this sense, one can say that even in active ergative languages, all and only transitive clauses have ergative subjects.
It is important to note that although agents are typical external arguments, the external argument category (initial subject or initial 1 in other frameworks) also includes some other kinds of θ-roles in many languages, and the exact range of verbs that take an external argument is known to vary from language to language.
9 There is a split based on tense in Georgian, so that ergative case is used only in Tense II.
10B, J, and V gloss gender categories (Nichols 2011:431).
11 The class of intransitives with an external argument has been called unergative, but this term is confusing since these are exactly the verbs that do mark their subjects with ergative case in active ergative languages.
13 Other verbs of this type include asun ‘to laugh’, ladun ‘to quarrel’, and gidun ‘to play’ (Wali and Koul 1997:153).
14 Additional languages that may be of the active ergative type include the Australian language Pitjantjatjara (Bowe 1990).
15 Ergative case often looks like another inherent or oblique case in the same language, sometimes dative, sometimes instrumental, genitive, or locative.
16 The Class I agreement in Tsez is male, human. The Class IV agreement is inanimate.
17 Note that this pattern of what is cross-referenced with agreement versus pronominal clitics is the reverse of what we find in clearly ergative languages such as Kashmiri (Wali and Koul 1997) and Basque (Arregi and Nevins 2012), where transitive/ergative subjects are cross-referenced by pronominal clitics while nominative objects control the agreement.
18 These pronominal clitics are not morphologically distinguished by case.
19 The rare situation in which an argument with ergative case can solely control the agreement, in those few languages that allow it at all, is in clauses where there is no nominative argument (e.g., Warlpiri; Woolford 2006a).
20Mateo-Toledo (2008) comes very close to identifying the Set A and Set B forms as agreement versus pronominal clitics/incorporated pronouns. He notes that the Set B forms ‘‘cliticize’’ to either a preceding or a following verbal element or function word and ‘‘may not be agreement markers,’’ as the Set B forms are more loosely connected than agreement affixes are and have been called incorporated pronouns (Mateo-Toledo 2008:48, citing Kaufman 1990 and Robertson 1992).
21 The idea that aspect splits function in grammar to mark, or redundantly mark, aspect is presented in Woolford 2013. I refer to this phenomenon as parasitic marking because existing grammatical elements are manipulated so as to correlate with aspect, and thus ‘‘mark’’ aspect in this sense.
22Coon (2010a) also argues that there is no ergative case split in Mayan languages, but for a different reason. Coon proposes that there are no clauses with imperfective aspect in Chol; instead, all the imperfective examples are actually nominalized, so that the apparent use of the Set A ‘‘ergative’’ markers is actually the use of a genitive possessive marker, which looks like the Set A markers in Mayan. Under Coon’s proposal, the imperfective example (i) is more accurately translated as indicated in parentheses, where the imperfective marker is analyzed as the main verb of the clause.
It is not clear whether this analysis can extend to the Chontal data in (118) and (119), however, since one would have to claim that only the positive imperfective clauses are nominalized, while the negative ones are ordinary imperfective clauses.
23 The abstract case of the nominative and accusative pronominal clitics is not morphologically distinguished.
24 The details of the formal proposal here are rather different from those in Woolford 2007.
25 I put aside word order differences among languages here for ease of exposition.
26 Under the view that only a closer DP with an unvalued case feature would block Infl from probing past it (e.g., Chomsky 2000), there is another possible explanation for the fact that verb raising prevents the MLC violation that would otherwise be expected. If the raised verb, which c-commands the shifted object, values accusative case on that object, the shifted object would be rendered inactive in Chomsky’s terms and would not prevent Infl from entering into an Agree relation with the external argument.
27 I assume here that object shift is a type of A-movement and that the object could not have undergone A-movement if it had already received structural accusative case from V in its base position.
28 I assume that ergative case is optionally assigned in these languages, but nominative is preferred unless the clause would crash because of an MLC violation. The preference for nominatives can be built into the Minimalist Program if an economy condition is added to the grammar making it more economical to use the least-marked case that can be licensed on an argument (Broekhuis and Woolford 2013).
29 I would like to thank Eric Potsdam for bringing this fact about Hawaiian to my attention. Medeiros (2013) glosses this i morpheme as accusative case, even though it does not appear on nonshifted objects, but Cook (1999:50) identifies it as a preposition because it is used as a preposition elsewhere in the language, marking among other things goals of motion, temporals, and locative.
Here I follow Cook (1999) in identifying i as a preposition.
30 The DOM marker is often glossed as accusative case, but it seems much more likely that it is the unmarked object in situ that has accusative case. It seems unlikely that the DOM marker is an inherent case (although that would avoid an MLC violation) since it should then also mark objects that remain in their θ-position. P-insertion may be the only option for altering the shifted object so as to avoid an MLC violation in object shift constructions. However, it is difficult to determine the exact identity of DOM markers crosslinguistically, given the problem of distinguishing postpositions that are actually case morphemes from postpositions that are Ps.
31 Some languages add a postposition/DOM marker to shifted objects even in clauses with an ergative subject; one such language is Hindi (Bhatt and Anagnostopoulou 1996). Why this occurs is not yet clear.