Automatic Questionnaire Construction for Lexical Typological Research of Verbs with more than 1 Argument

The paper addresses an issue of an automatic creating questionnaires for lexical typological studies in the Frame approach paradigm. A research in this framework is based on the analysis of distributional properties of the lexemes in question. Hence, questionnaires for such studies consist of typical contexts where lexical items from a given semantic domain can potentially occur. This paper aims at creating such questionnaires automatically, particularly for verbs with more than 1 argument. This task can be divided into two different problems and this study struggles with both of them. Firstly, it is the search for contexts which reflect the distributional properties of the relevant verb. Solving this task means finding optimal parameters of context defining such as context window width, semantics of contextual words and morphological properties of lexemes to be considered as a context. Secondly, it’s dividing contexts into what is called frames in the Frame approach paradigm. Therefore, this study deals with the task of word clustering and proposes the most convenient clustering model for the aim of the research. The algorithm suggested for completing of both tasks uses text data from monolingual morphologically annotated corpora. Its final version is tested on the data ‘closing verbs’ semantic domain. Keywords: machine learning, nature language processing, lexical typology, semantics, distributional semantics, corpus study, automatic creating questionnaires, verbs, clustering.