A web-based unsupervised algorithm for learning transliteration model to improve translation of low-frequency proper names

In machine translation, cross-language information retrieval, and cross-language question answering, the problems of unknown term translation are difficult to be solved. Although we have proposed several effective Web-based term translation extraction methods exploring Web resources to deal with translation of frequent Web query terms. However, many low-frequency unknown terms are still difficult to be translated by using our previous Web-based term translation extraction methods. Therefore, in this paper we propose a two-stage hybrid translation extraction method, which is composed of our pervious Web-based term translation extraction method and a new Web-based transliteration method to improve translation of low-frequency unknown proper names. Additionally, to construct a good quality transliteration model, we also present a Web-based unsupervised learning algorithm to automatically collect diverse English-Chinese transliteration pairs from the Web. Experimental results showed that our new method can make great improvements for translation of unknown proper names.