Sentence Alignment of Historical Classics based on Mode Prediction and Term Translation Pairs
Chao Che, Xiaojun Zheng · 2014
Parallel corpora are essential resources for the construction of bilingual term dictionary of historical classics. To obtain large-scale parallel corpora, this paper proposes a sen-tence alignment method based on mode pre-diction and term translation pairs. On one hand, the method rebuilds the sentence alignment process according to characteris-tics of the translation of historical classics, and adds mode prediction into the sentence alignment. On the other hand, due to the lack of bilingual ancient Chinese dictionary, the method exploits the term translation pairs ex-tracted from manually aligned sentence pairs to perform alignment. The method first pre-dicts the alignment mode probability accord-ing to the character number, punctuation number and some characters of Chinese sen-tence, then performs sentence alignment us-ing length alignment probability, term align-ment probability and mode probability. Be-sides, the method selects anchor sentence pairs based on sentence length and predicted mode to prevent the spread of alignment er-rors. The experiment on ”Shi Ji ” demon-strates that mode prediction and term transla-tion pair both enhance the performance of sentence alignment obviously. 1