AbstractThe limited amount of the sense annotated data is a big challenge for the word sense disambiguation task. As a solution to this problem, we propose an algorithm of automatic generation and labelling of the training collections based on the monosemous relatives concept. In this article we explore the limits of this algorithm: we employ it to harvest training collections for all ambiguous nouns, verbs and adjectives presented in RuWordNet thesaurus and then evaluate the quality of the obtained collections. We demonstrate that our approach can create high-quality labelled collections with almost full-coverage of the RuWordNet polysemous words. Furthermore, we show that our method can be applied to the Word-in-Context task.
Language Modelling Makes Sense: Propagating Representations through WordNet for Full-Coverage Word Sense Disambiguation
