Word Embeddings of Monosemous Words in Dictionary for Word Sense Disambiguation

摘要

In the recent past, word embedding techniques have shown to capture semantic and syntactic information of natural language which could be exploited to solve the Word Sense Disambiguation (WSD) task. Word embeddings are generated using words appearing in context. However, some co-occurrence words in context have multiple meanings and are ambiguous. Therefore, it is sometimes difficult to identify the meaning of a target word by using word embeddings of context words. In this paper, we propose to use word embeddings of monosemous words for the WSD task. We consider that word embeddings of monosemous words can contribute to determining the correct sense of a target word. Also, by using word dependency in a sentence, it is possible to capture the semantic relationship between the target word and the co-occurrence word as a feature. To evaluate the efficiency of the proposed WSD method, we show that it is effective for the WSD task to use both monosemous word information and dependency relation to the target word.