An Effective Approach for Searching Closest Sentence Translations from the Web

Abstract

There are large numbers of well-translated sentence pairs on the Web, which can be used for translating sentences in different languages. It is an interesting problem to search the closest sentence translations from the Web for high-quality translation, which has attracted significant attention recently. However, it is not straightforward to develop an effective approach, as this task heavily depends on the effectiveness of the similarity model which is used to quantify the similarity between two sentences. In this paper, we propose several optimization techniques to address this problem. We devise a phrase-based model to quantify the similarity between two sentences. We judiciously select high-quality phrases from sentences, which can capture the key features of sentences and thus can be used to quantify similarity between sentences. Experimental results show that our approach has performance advantages compared with the state-of-the-art sentence matching methods.