| |
Kustova G., Toldova S.
National Corpus of Russian Language: Semantic Filters for the Verb Sense Disambiguation
This report deals with methods of word sense disambiguation (reduction) using the information about verb argument structure. Most of the systems based on this method require specially designed resources such as WordNet, FrameNet etc. We explore the possibility to extract and use the information available from the standard dictionaries including a Verb-argument dictionary. We used a subcorpus of National corpus of Russian language that has unambiguous morphological annotation as training and testing data. The aim was to reduce the number of tags for verbs in the semantic annotation. The experiment has shown that the information extracted from dictionaries could not be used as it is. However the extracted argument structure can be used as a seed set for future training. It allows to remove rare meanings and can reduce the number of semantic tags for a verb. The further corpus training and enriching the argument structure with general semantic properties of nouns can further improve the method.
Back |
