Middle Voice Marking in Dena'ina Iteratives: How Syntax and Semantics Interact
Posted: 20 Nov 2008
Date Written: October 18, 2008
Recent studies of middle voice verbs in the Athabaskan family have been largely successful at finding a unified semantic motivation for their coherence as a class (Kibrik 1996, Rice 2000). However, in Dena'ina Athabaskan (Alaska), iterative verbs are sometimes middles, but as a set they have generally been analyzed as divergent from other middle constructions: marking of iteratives as middles is generally said to be motivated by syntactic transitivity, rather than by semantics. Forms like (1) are said to be the norm: iteratives that are intransitive have the middle voice marker d, while iteratives that are transitive do not.
(1a) intransitive: nu-ghe-d-yul ITER-ASP-MID.VCE-walk 's/he is walking back'
(1b) transitive: nu-yi-l-jeh ITER-4OBJ-CLF-hit 's/he hit him/her again'
However, our corpus-based study of traditional Dena'ina narratives reveals that many forms display the opposite distribution (2), presenting researchers with the puzzle of what ultimately determines d's presence or absence:
(2a) intransitive: nu-ghi-l-ghatl' ITER-CONJ-CLF-become.dark 'it got dark again'
(2b) transitive: qunsha nu-qe-t-d-ghax ground.squirrels ITER-3P.SUB-ASP-MID.VCE-pack 'they kept packing ground squirrels back'
To determine whether the distribution of middle voice d is determined by syntactic or semantic aspects, we coded 280 iteratives from 220 pages of Dena'ina texts for grammatical transitivity and the semantic subcategory of iterativity. Using two multifactorial statistical approaches - binary logistic regression and hierarchical configural frequency analysis - we show that counter to other accounts, semantics not only plays a large role in middle marking, but it also interacts with grammatical transitivity.
As to the former, semantically different subcategories of iterativity can be ranked in terms of their preference for d-marking on a concrete-to-abstract cline (figure 1). We show that middle marking is statistically most likely to occur in iteratives that are semantically spatial reversives, in which the source and goal of the action are undifferentiated. This finding is congruous with Rice's observation that middles in Athabaskan cohere as a class based on the "suppression of differentiation of arguments and ... events" (2000:179). Our analysis supports the assumption of a semantically unified class of middle verbs.
As to the latter, there is an interesting interaction such that intransitives and transitives exhibit different points on that cline as to where d becomes (dis)preferred.
Apart from illustrating the interaction of syntactic and semantic forces determining d-marking, this paper underscores the relevance of quantitative corpus-based methods: we show that even small textual databases from lesser-studied languages are amenable to quantitative corpus-based work, and we exemplify how the findings from a multifactorial approach can help integrate previously conflicting views on middle voice marking by revealing a previously unnoticed interaction of factors.
Keywords: middle voice, cognitive linguistics, semantics, syntax
Suggested Citation: Suggested Citation