Data-driven Valence Typology
Mary Esther Kropp Dakubu and Lars Hellan
Nov. 7, 2011
Data-driven Valence Typology (DVT) is a project where we seek to represent the characteristic sentence construction types of a language – called its c-profile - in a transparent, detailed and non-theory-biased format, drawing from a common, restricted repertory of analytic-descriptive primitives [1]. By adhering to a common classification system, DVT in principle allows for its data to be searchable in a relational database. DVT has so far been developed with a view to cover significantly different languages (Ga from the Niger-Congo family Kwa, Norwegian from Germanic, and Kistaninya from Ethio-Semitic), while in a current phase the project has a more ‘micro-comparative’ focus, in showing how a profile for one language of a given family can be derived from the c-profile of another language in the same family. In Germanic we envisage such extensions with regard English and German, and in Kwa/Gur with regard to Dangme and Gurene.
In situating DVT among current projects and initiatives, it can perhaps be most directly related to VerbNet [2], its non-computational predecessor in Levin's work [3], and a cross-linguistic development of the latter, the Leipzig Valency Classes Project[4].
In future publications we will show how an inventory of verb classes in the Levin approach can be derived from a DVT c-profile and an accompanying verb construction lexicon, as are available for Ga [5], and for Norwegian [6]. We will also assess the notion of ‘valence alternation’ as a comparison unit, by itself notoriously difficult to define, and show that for the 150 most salient frames in Ga, none of them are interconnected by any of the ‘alternation’ patterns which are commonly applied in the European setting. We will advocate DVT as offering a sounder general basis for valence typology, not being directly dependent on notions like 'alternation'.