A novel method for parallel resources acquisition from bilingual web pages

A new approach is proposed for acquiring parallel resources by expanding seeds (some high-quality parallel sentence pairs) in the same page. Specifically, given a bilingual web page, the method is composed of three challenging tasks: 1) page's content structure is detected for adaptive viewing on its main data region; 2) some high-quality parallel sentence pairs in the main data region are selected as seeds for the next task; 3) wrappers for mining parallel resources are learnt by expanding the seeds. We identify and extract more parallel resources in the same page using the learnt wrappers. Our method can be applied to web documents written in any markup language and in any human language. The test is based on 500 bilingual web pages randomly selected from 12 popular English learning web sites, which gets a high F-score of 85.68%. The experimental results show that our method is quite promising. Copyright © 2013 Binary Information Press.