Query translation for CLIR: EWC vs. Google Translate

A new approach to find accurate translation of search engine queries from Japanese into English for the CLIR task is proposed. The Mecab system and online dictionary SPACEALC are utilized to segment Japanese queries and to get all possible English senses for every term detected. To disambiguate terms, the idea of the shortest path on an oriented graph is applied. Nodes of this graph symbolize word senses and edges connect nodes representing neighboring Japanese terms. The EWC semantic relatedness measure is used to select the most related meanings for the translation results. This measure combines the Wikipedia-based Explicit Semantic Analysis measure, the WordNet path measure and the mixed collocation index. The proposed technique is tested on the NTCIR data collection. Queries generated by Google Translate were used to evaluate the quality of translation.