Iterative Mining Translations from the Web

Multilingual translations play a vital role multilingual or cross-lingual information retrieval and extraction. In this paper, we describe a new method mine translations from bilingual web pages based on our former research. Two new features are introduced, one is the iterative mining process in order to increase the number of translation pairs; the other is the filtering step which deletes language-specific prefix and postfix in hyperlinks. Experiments show that the precision has been greatly improved due to the filtering step and the number of translation pairs increased after six iterations.