Filtering or adapting
Lixin Shi, Jian‐Yun Nie · 2006
Noisy parallel corpora have been widely used for Cross-language information retrieval (CLIR). However, the previous studies only focus on truly parallel corpus. In this paper, we examine two possible approaches to exploit noisy corpora: filtering out noise from the corpora or adapting the training process of translation model to the noise corpora. Our experiments show that the second approach is better suited to CLIR.