名词短语事件指代消解研究

随着篇章理解、信息融合要求的日益迫切，提高事件指代消解系统的性能逐渐成为一个研究热点。事件指代的指代词指代的是事件、事实或者命题等实践性、抽象性对象。该文对名词短语的事件指代消解进行研究，使用平面特征、结构化句法特征和语义特征等，根据SVM机器学习的方法进行英文事件的指代消解，通过在计算事件语义相似度的元组（语义角色）中加入时间和地点元素改进语义特征来提高事件指代消解系统的性能；并且单独使用每种特征对语料进行实验，分析每种特征单独使用时对系统的影响；OntoNotes 4.0语料库上的实验结果显示，引入改进的语义特征后，与基准系统相比，系统的准确率和F值均有所提高。由此来看，在语义特征中加入时间和地点元素对事件指代消解具有的正向作用。

Research On Noun Phrase Event Anaphora Resolution

As the requirement of discourse understanding and information fusion is becoming more and more urgent, improving the performance of event anaphora resolution system is becoming a hot topic. Event anaphor is reference to an event, fact, and proposition which is representative of eventually and abstract objects. This paper focuses on noun phrases event resolution. It improved the semantic features by adding the elements of time and address when computing semantic similarity of tuples (semantic roles information). Experiments on the English portion of OntoNotes 4.0 show that the semantic roles information (Argm-Log, Argm-Tmp) can significantly boost the performance of the baseline for Event noun phrases resolution. It outperforms the baseline system by 0.49% in Precision and 0.2% in F-measure. Consequently, it proved Argm-Log and Argm-Tmp can improve the event noun phrases resolution.