Nominalization and Predicate-Fronting: Two Sources of Ergativity

In this paper we propose a unified account of ergativity in two unrelated languages with very different systems of marking grammatical relations. Both Chol (Mayan, Mexico) and Mẽbengokre (Jê, Brazil) have what appear to be aspect-based ergative splits. However, while in Mẽbengokre ergativity is found in verb stems which are formally nominal, ergativity in Chol is found in what can be shown to be truly verbal verb stems. We propose that despite these differences, the source of ergativity is the same in both languages: Ergativity is the result of a separation of the predicate from T. In Mẽbengokre this separation arises from a nominalization. In Chol, a predicate-initial VOS language, the separation is the result of fronting of the predicate vP to Spec,TP. The analysis presented here provides a unified account not only of ergativity in Chol and Mẽbengokre, but has the potential to be extended to a number of other ergative languages discussed in the literature. The data: According to the typological literature, if a language has an aspect-based ergative split, it associates ergative alignment with perfective aspect (or with the perfect, if it is distinct from perfective), and accusative alignment with imperfective, as in (1). Both Chol and Mẽbengokre have splits that conform to this generalization. Chol’s split opposes perfectives to imperfectives: the perfective clauses in (2) show an ergative-absolutive pattern, while the imperfectives in (3) appear nominative-accusative. In Mẽbengokre, matrix clauses with an ergative pattern receive a perfect interpretation, (4), while clauses with accusative alignment, such as (5), are either perfective or imperfective. Nevertheless, the typological generalization fails to capture the fact that in Mẽbengokre the aspectual values in question come about via a nominal form of the predicate, which is itself argued to be directly responsible for ergativity. The connection between ergativity and nominalization has been discussed in Inuktitut by Johns (1992), and in process nominals in Indo-European languages (Alexiadou 2001). Given the formal identity between matrix ergative clauses in Mẽbengokre and embedded nominalizations, nominalization constitutes a plausible source for the ergativity in the former. Attempting to tie ergativity to nominalizations in general, however, runs into problems in the case of Chol, where the perfective forms—which trigger ergative alignment—are verbal, but the imperfective stem forms—which trigger accusative alignment—are argued to be nominal. Proposal: We argue that despite the differences between Mẽbengokre and Chol, the ergative alignment in both of these languages arises under very similar circumstances: ergativity occurs whenever the predicate cannot be linked directly to T. In Mẽbengokre, this occurs because nominal predicates (in particular, deverbal ones) require an abstract existential copula to be interpreted, shown in (6). In Chol, on the other hand, the link is broken by movement of a phrase containing the predicate to a higher specifier, as in (7). We follow Marantz (1991) in proposing that ergative and accusative are both dependent cases: the appearance of ergative case depends on the presence of an absolutive case competitor within the same case assignment domain, and likewise for accusative versus nominative. However, the choice between ergative and accusative alignment, we argue, is not a parameter that can be set in a construction-specific way. Rather, we contend that while dependent accusative case is assigned in a domain that contains two case-assigning heads (v and T, merged or not), dependent ergative is assigned within domains that contain a single case-assigning head: n, or v which is not in the same case-assigning domain as T (Salanova 2007). The lack of a direct link between the predicate and inflectional morphology has clear semantic consequences in Mẽbengokre: clauses with ergative alignment differ from those with accusative alignment in that the former do not link the event time directly to a topic time, but rather give its occurrence as background information (i.e., ergative clauses translate the English experiential perfect or generic). In Chol, evidence for predicate-fronting can be found from the placement of adverbs and other adjuncts, as well as from the impossibility of definite objects in VOS constructions: definite objects must undergo object shift out of the predicate prior to predicate-fronting, resulting in VSO order. Predicate-initial languages which also exhibit ergativity include other Mayan languages, as well as languages in the Otomanguean, Austronesian and Salish families. If predicate-initiality is the result of fronting of a phrasal predicate to a higher functional projection (as argued, for example by Massam (2001) for Niuean (Austronesian) and Lee (2000) for Zapotec (Otomanguean)), then the fact that these languages are also ergative is predicted under our analysis.