Latviešu valodas verbu valences marķēšana

LTReikšminiai žodžiai: Latvių kalba; Semantiniai vaidmenys; Semantinis valentingumas; Sintaksinis valentingumas; Valentingumo modeliai; Veiksmažodis; Veiksmažodžiai; Latvian language; Semantic roles; Semantic valence; Semantic valency; Syntactic valence; Syntactic valency; Valence models; Valency models; Verb; Verbs.

ENDevelopment of valency resource for Latvian was started recently. The chosen approach takes into account the experience of other projects dealing with verbal valency (such as FrameNet for English, the valency dictionary of Lithuanian by N. Sližienė and Vallex for Czech) and is supplemented with our own ideas. The article describes our approach to the verbal valency annotation – the valency types (syntactic, semantic valency, selectional restrictions), the set of the semantic roles, and presents the tool developed for the valency annotation. Initially two annotation layers are chosen – semantic valency (semantic roles) and syntactic valency – the morphological description of the dependents of the verbs. The valency models (combinations of semantic roles) are generated from the annotated core semantic roles only. Non-core semantic roles are not used in this process because they are not specific enough to particular verbs. In future the annotation of one more valency layer is planned – very general ontological categories (animate, inanimate, abstract object, physical object, human) will be added. An online tool is developed for manual annotation of valences within a pre-selected corpus. The valency models, as well as morphological forms corresponding to each semantic role are suggested automatically. Summarized information of the valency of each verb, as well as a corpus of annotated sentences will be available as a result of the project. Currently valency of approx. 180 verbs (more than 25 000 utterances) is annotated using data from the Balanced Corpus of Modern Latvian. [From the publication]