Evaluating the Quality of Web-Mined Bilingual Sentence Pairs

We come up with the problem of evaluating the quality of bilingual sentence pairs mined from the web, which is critical for a wide range of applications such as statistical machine translation (SMT) and English as Second Language (ESL) learning. To address this problem, we propose a novel method that integrates multiple linguistic features related to spelling, grammar, alignment, and particularly the sentence type feature that indicates if a sentence can be parsed by the Link Grammar Parser (LGP). Promising results are achieved on a bilingual corpus of about 6 million English-Chinese sentences mined from the web, indicating the effectiveness of our proposed method.