1amirkabir university of technology, computer and information engineering, hlt lab
Parallel corpus in different domains from domain of parallel training corpus, may have some sentences that are useful to use in training of system. These sentences have words and phrases that are occur in in-domain corpora too. The context of these sentences is more similar to context of parallel training corpora and is far from context of out-of-domain parallel corpora that are contain them. In these paper we suggest a method based on topic models to extract some sentences from out-of-do-main parallel corpora that their context are similar. We used these extracted sentences in training of system. We will show BLEU of the system will in-crease about 4.69 BLEU.