Contextual Information Based Technical Synonym Extraction

Popular methods for acquiring synonymous word pairs from a corpus usually require a similarity metric based on contextual information between two words, such as cosine similarity. This metric enables us to retrieve words similar to a query word, and we identify true synonyms from the list of synonym candidates. Instead of stopping at this point, we propose to go further by analyzing word similarity network induced by the similarity metric and re-ranking the synonym candidates -- a mutual re-ranking method (MRM). We apply our method to a specific domain: technical synonym extraction from aviation reports in Japanese. Even though the Technical Corpus is small and the contextual information is sparse, the experimental result shows the effectiveness of applying the contextual information on extracting technical synonyms in Japanese, and that MRM boosts the quality of acquired synonyms.