基于统计的汉英法律文献亚句子级对齐

Sub-Sentence Alignment of Chinese-English Law Literature Based on Statistical Approach

基于统计的句子对齐是根据双语句子的长度在所有可能的对齐中找出概率最大的一个·提出两种对齐汉英语料的参数计算方法,使对齐模型中的评价函数满足标准正态分布·参数s2的值是对训练语料中的所有点(l1,(l2-cl1)2)进行线性回归分析所得直线的斜率,s2的另一种求法是直接计算方差·实验结果表明汉英法律文献亚句子级对齐的正确率为98 8%,召回率为99 2%·

Sentence alignment based on statistical approach is the choice of alignment with maximum probability from all candidates according to the length of bilingual sentences. ChineseEnglish law literature is translated literally, so it is suitable to be aligned with statistical approach. But the method used to compute the parameters in processing IndoEuropean languages cannot be applied to ChineseEnglish corpora. Two parameter computation methods for aligning ChineseEnglish corpora were presented. The method make the evaluation function satisfy the standard normal distribution. One method to get the parameter s2 is to compute slope of the line generated by linear regression analysis to all point (l1,(l2-cl1)2) in the training corpora. The other is to compute the variance. Test results show that the precision rate and recall rate of alignment are 98.8% and 99.2 % respectively.