Polish WordNet on a shoestring

Abstract

A project to create a Polish WordNet is under way. Rather than localise the English WordNet, we are constructing the lexical network from scratch, in two phases. First, we have established the linguistic principles, among them a list of semantic relations with detailed diagnostic tests. We have also implemented a client software tool that records the lexicographers ’ decisions in a central database. A core WordNet, populated with around 10,000 most frequent lexemes in the IPI PAN Corpus, will be a fully functional resource for Natural Language Processing in Polish. In the second phase, the enhanced software tool will detect candidate semantic relations in a much larger corpus, based on statistical methods of grouping words by semantic similarity. Lexicographers will review and approve such candidate relations. 1