Abstract
We introduce a method which automatically aligns bilingual pages on the Web by using a machine translation system. In our approach, a source page is translated using a MT system and aligned with most similar candidate pages. The method involves collecting bilingual pages on the Web, translating each source page, aligning and evaluating each source page with the possible candidate pages based on some similarity measures. The similarity measure is based on a simple approach of counting the ratio of shared characters between a machine translated page and an actual target pages. In order to achieve optimal performance, shared characters are checked for their positions in both pages. These position pairs deviated from the norm are excluded from consideration in calculating page to page similarity. We describe an implementation of the method by using three MT systems to align online English and Chinese news pages at Voice of America (VOA) website. Experimental results indicate that English and Chinese Web news pages can be aligned with the average precision rate of 99%. The results indicate that the method can be applied to collect online parallel corpora easily and efficiently, therefore, meeting the increasing demand for parallel corpora to be used as a linguistic resource in various natural language processing tasks.