Using Thesaurus Data to Improve Coreference Resolution for Russian

Semantic information about entities, specifically, how close in meaning two mentions are to each other, can become very useful for the task of co-reference resolution. One of the most well-researched and widely used forms of presenting this information are measures of semantic similarity and semantic relatedness. These metrics are often computed, relying upon the structure of a thesaurus, but it is also possible to use alternative resources. One such source is Wikipedia, which possesses the category structure similar to that of a thesaurus. In this work we describe an attempt to use semantic relatedness measures, calculated on thesaurus and Wikipedia data, to improve the quality of a co-reference resolution system for Russian language. The results show that this is a viable solution and that combining the two sources yields the most gain in quality.