Valency Lexicon of Czech Verbs

Valency is a property of language units reflecting their combinatorial potential in language utterances. The availability of the information about valency is supposed to be crucial in various Natural Language Processing tasks. In general, valency of language units cannot be automatically predicted, and therefore it has to be stored in a lexicon. The primary goal of the presented work is to create a both humanand machine-readable lexicon capturing valency of the most frequent Czech verbs. For this purpose, valency theory developed within Functional Generative Description (FGD) is used as the theoretical framework. The thesis consists of three major parts. The first part contains a survey of literature and language resources related to valency in Czech and other languages. Basic properties of as many as eighteen different language resources are mentioned in this part. In the second part, we gather the dispersed linguistic knowledge necessary for building valency lexicons. We demonstrate that if manifestations of valency are to be studied in detail, it is necessary to distinguish two levels of valency. We introduce a new terminology for describing such manifestations in dependency trees; special attention is paid to coordination structures. We also preliminarily propose the alternation-based lexicon model, which is novel in the context of FGD and the main goal of which is to reduce the lexicon redundancy. The third part of the thesis deals with the newly created valency lexicon of the most frequent Czech verbs. The lexicon is called VALLEX and its latest version contains around 1600 verb lexemes (corresponding to roughly 1800 morphological lemmas); valency frames of around 4400 lexical units (corresponding to the individual senses of the lexemes) are stored in the lexicon. The main software components of the dictionary production system developed for VALLEX are outlined, and selected quantitative properties of the current version of the lexicon are discussed.