基于树形条件随机场的跨语言时态标注

Cross-Lingual Tense Tagging Based on Tree Conditional Random Fields

提出时态树的概念和构造方法,从而将汉英时态转换问题转换为时态树标注的问题.而后,使用树形条件随机场为未标注时态树的结点标注英语时态.提出的特征函数的模板较好地满足了模型推断的需要.实验结果表明:与基于线性条件随机场模型的时态标注方法相比,基于时态树方法的准确率有大幅度的提高,说明使用时态树能够更好地表达子句间时态的依赖关系.

In this paper, the concept of a tense tree and its construction method are proposed such that the issue of Chinese-English tense conversion is transformed into the issue of tagging a tense tree. Futher, tree-CRF is used to tag nodes of the untagged tense tree with English tenses. Templates of feature functions are suitable for the need of model inference. Experimental results show that the method of tree-based tense tagging is much better than linear-based tense tagging, and that tense trees can better express tense dependencies between clauses.