Using Textual and Lexical Resources in Developing Serbian Wordnet

In this paper we present two techniques for using textual and lexical resources, such as corpora and dictionaries, in validation and refinement of Serbian wordnet. We first describe how the existing monolingual Serbian corpus, the bilingual Serbian/English (S/E) and Serbian/French (S/F) aligned corpora, and the appropriate morphological e-dictionaries can be used in validation and enhancement of Serbian wordnet synsets. Then we present a quantitative technique based on a set of frequency parameters which indicate the coverage of a corpus by wordnet literals, the significance of one sense of a literal relative to the others, as well as the significance of one literal in a synset compared to other literals in the same synset. Experimental results justifying the applied techniques are given.