Abstract
We introduce a method for aligning words and phrases in a given pair of bilingual sentence using the Web as corpus. In our approach, each Chinese word and all English words are transformed into an query, sent to the search engine to retrieve mixed-code snippets. We use the returned snippets to align words and phrases such that the word and aligned word or phrase are likely to be translation counterparts. The method involves calculating features of alignment candidates, filtering candidates, scoring candidates, and ranking candidates. Finally, we select the most likely word or phrase alignment for each source word. The results show that the method can reach 53.2% recall and 88.6% precision.