The translation of patents or scientific papers is a key issue that should be helped by the use of statistical machine translation (SMT). In this paper, we propose a method to improve Chinese-Japanese patent SMT by pre-marking the training corpus with aligned bilingual multi-word terms. We automatically extract multi-word terms from monolingual corpora by combining statistical and linguistic filtering methods. We use the sampling-based alignment method to identify aligned terms and set some threshold on translation probabilities to select the most promising bilingual multi-word terms. We pre-mark a Chinese-Japanese training corpus with such selected aligned bilingual multi-word terms. We obtain the performance of over 70% precision in bilingual term extraction and a significant improvement of BLEU scores in our experiments on a Chinese-Japanese patent parallel corpus.
展开▼