EVALUATING SEMANTIC SIMILARITY MEASURES FOR ENGLISH VERBS IN WORDNET AND THESAURI
Iryna Dilai, Mykhailo Bilynskyi · Inozenma Philologia · 2016
Semantic similarity measures the distance between concepts and is based on their likeness. WordNet-based similarity metrics summarized by Pedersen can serve to compare both the distances between separate concepts and the metrics themselves. Establishing and comparing WordNet-based verb similarity can be applicable for a number of NLP tasks. However, the measures, being predominantly of a non-linear character, fail to account for synonymy of words, as well as convey the principles of mental lexicon structuring, in particular the assymetry of associations. The combination of concept-based and word-based similarity measures can be leveraged to solve this problem. Attention will be paid to the applicability of the synonymous distance metric to the reversal of multiple (over twenty) thesauri of English verbs and the study of versatile issues of the geometry of semantic spaces based on vast numbers of semantic proximity values in the (near-)synonymy of verbs. Keywords: semantic similarity measure, asymmetry of associations, taxonomic relations, synset, corpus, WordNet, thesauri of English verbs, reverse synonymous strings of verbs, uneven distance-from-the-dominant scales, (non-)Euclidean geometry, vector analysis, angular geometry in a thesaurus