To Pay or to Get Paid: Enriching a Valency Lexicon with Diatheses

摘要

Valency lexicons typically describe only unmarked usages of verbs (the active form); however verbs prototypically enter different surface structures. In this paper, we focus on the so-called diatheses, i.e., the relations between different surface syntactic manifestations of verbs that are brought about by changes in the morphological category of voice, e.g., the passive diathesis. The change in voice of a verb is prototypically associated with shifts of some of its valency complementations in the surface structure. These shifts are implied by changes in morphemic forms of the involved valency complementations and are regular enough to be captured by syntactic rules. However, as diatheses are lexically conditioned, their applicability to an individual lexical unit of a verb is not predictable from its valency frame alone. In this work, we propose a representation of this linguistic phenomenon in a valency lexicon of Czech verbs, VALLEX, with the aim to enhance this lexicon with the information on individual types of Czech diatheses. In order to reduce the amount of necessary manual annotation, a semi-automatic method is developed. This method draws evidence from a large morphologically annotated corpus, relying on grammatical constraints on the applicability of individual types of diatheses.