Reconstruction of Implied Semantic Relations in Russian Wiktionary
С.В. Клименков, Evgenij Tsopa, Alexey Pismak, Alexander Yarkeev · 2016
There were several attempts to retrieve semantic relations from free, online Wiktionary for Russian language. Previous works combine automatic parsing of wiki snapshot with experts’ assistance. Our main goal is to create machine readable lexical ontology from Russian Wiktionary, maximally close to its online state. This article provides approach to automatic creation of explicit and implicit semantic relations between words (lexemes) and meanings (senses) to provide exact relations from sense to sense. Explicit semantic relations are constructed comparatively easy. For example, if the lexeme contains single sense, then all relations that point to the lexeme will point to this single sense. Reconstruction of implicit relations relies on logical conclusions from already created explicit ones. Several algorithms for implicit semantic links were developed and tested on Russian Wiktionary. There were parsed more than 550000 online pages, containing about 250000 Russian lexemes with about 500000 senses in them, but only about 20% of these senses were linked with at least one external lexeme. About 47% of explicitly existing links were resolved as “sense-to-sense” relations and about 28% of new implicit “sense-to-sense” links were reconstructed. 53% of lexemes’ references could not be resolved to exact sense.