融合特征约束模型的纳西-汉语双语词语对齐算法

A Bilingual Word Alignment Algorithm of Naxi-Chinese Based on Feature Constraint Models

针对纳西语、汉语因句法结构差异较大而导致双语词语自动对齐较为困难的问题,提出一种融合特征约束模型的纳西-汉语双语词语对齐算法.首先在语料中统计纳西-汉语词语区间扭曲和位置转换特性,并由此建立2个双语词语对齐的特征约束模型;然后将提出的特征约束模型融入词语对齐的对数线性模型框架,并结合最小错误率算法训练模型参数;最终搜索出最佳的词语对齐结果.实验以IBM Model3为词语对齐比较模型,结果表明,该双语词语对齐算法可以使纳西-汉语词语的对齐准确率提升21.9%.

A bilingual word alignment algorithm of Naxi-Chinese based on feature constraint models is proposed to reduce the difficulty of bilingual word alignment for Naxi-Chinese which has huge difference in syntactic structure. Two feature constraint models- interval distortion model and position transformation model are established by counting the traits of interval distortion and position transformation in corpus, and are integrated into a log-linear framework of word alignment. Then parameters in the models are trained using the minimum error rate algorithm and the best alignment results are eventually searched. Experimental results on IBM Model3 show that the proposed algorithm increases the word alignment accuracy of Naxi-Chinese about 21.9%.