Introduction to an Interlanguage Corpus-WECCL 2.0

Authentic and objective material with regard to a certain language is the bare necessity and premise of any kind of linguistic research and then on the basis of the data, corpus-based approach, which is also addressed as a performance-based approach, has been involved in almost every corner of linguistics. Thus far, it is hardly surprising that some typical Corpora are essential for contemporary Chinese EFL's study. WECCL 2.0, as a representative of interlanguage corpus, is introduced in this paper.

基于多特征多注意力模型的中文事件关系识别

当前的关系识别任务多采用基于词或字粒度单一特征进行,忽略了全局信息对于关系识别的重要性,而且现代汉语具有词类和句法成分关系复杂的特点,这使得特征选择成为中文文本处理中的重点与难点。文中所用多特征多注意力模型除事件自身特征外,充分地考虑到位置、事件要素和上下文三类额外特征,利用全局信息以解决模型特征矩阵语义表征不足的问题。该模型结合双向注意力机制、点积注意力机制和双向门控循环神经网络进行事件关系识别,结合注意力机制的神经网络模型来较好地提取文本中的深层语义信息。其中双向注意力从特征矩阵两个方向提取事件自身有效信息,点积注意力提取事件之间的对应关系,双向门控循环神经网络提取矩阵中的上下文特征。在CEC2.0中文突发事件语料库上的实验结果表明,文中方法以及所用模型均有较好的识别效果。