A corpus-based approach to reevaluation of Croatian verb classification

Abstract

Summary Croatian grammar textbooks have a long tradition of classifying verbs based on their morphosyntactic characteristics. Conclusions, such as the frequency or productiveness of a class, were drawn without having the insight into a big corpus. Corpora used in such descriptions were not described and were presumably made of literary works which is, in our opinion, describing a form of the Croatian language distant from its everyday use. The corpus used for analyzing verbs in this paper is hrWaC which contains 1.9 billion tokens and about 90,000 verbs. This corpus was selected with the intention of describing and analyzing a less formal and less standardized language This paper offers a corpus-based approach to the problem of verb classification and emphasizes the importance of NLP methods in the process of classification as they fasten and simplify it. The paper gives a brief introduction to verbs, their morphological characteristics and their classification. By extracting verbs from the Croatian web corpus hrWaC and processing them computationally, the paper gives an insight into the verb distribution in the Croatian language and points out some difficulties that were encountered during this study. Even though this paper aimed to reevaluate the existing data data, the present findings mostly confirm the claims of previous researches. A number of recommendations for future research are given, foremost, the need of the extension of the language material.