¹School of Software, Harbin University of Science and Technology, Harbin, China
²College of Information and Communication Engineering, Harbin Engineering University, Harbin, China
³School of Computer Engineering, Vocational University of Suzhou City, Suzhou, China
⁴School of Computer Engineering, Vocational University of Suzhou City, Suzhou, China
⁵School of Computer Science and Technology, Heilongjiang University, Harbin, China

---

Abstract—Phrase translation pairs are very useful for bilingual lexicography, machine translation system, crosslingual information retrieval and many applications in natural language processing. Phrase translation pairs are always extracted from bilingual sentence pairs. In this paper, we extract phrase translation pairs based on word alignment results of Chinese-English bilingual sentence pairs and parsing trees of Chinese sentences, in order to decrease the influence of the grammar disagreement between Chinese and English. Discriminative features for phrase translation pairs are proposed to evaluate extracted ones in this paper, including translation literality, phrase alignment probability and phrase length difference. Multiple linear regression model combined with N-best strategy will be employed to filter phrase translation pairs, in order to improve the evaluating and filtering performance. Experimental results indicate that the filtering performance of phrase alignment probability is best in three kinds of discriminative features for evaluating Chinese- English phrase translation pairs. After multiple linear regression model combined with N-best strategy is used, its F1 achieves 86.24%.

Index Terms—Phrase translation pairs, natural language processing, bilingual sentence pairs, parsing trees, discriminative features, multiple linear regression

[PDF]

Cite: Chun-Xiang Zhang, Ming-Yuan Ren, Zhi-Mao Lu, Ying-Hong Liang, Da-Song Sun, Yong Liu, "Multiple Linear Regression for Extracting Phrase Translation Pairs," Journal of Computers vol. 6, no. 5, pp. 905-912, 2011.