摘要
事件指代消解根据指代词的不同可以分为代词的事件指代消解和名词短语的事件指代消解。研究了语义角色对名词短语的事件指代消解系统的影响,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。Onto Notes 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。验证了时间和地点元素对事件指代消解的正面影响。
According to the different anaphora,event anaphora resolution can be divided into two parts： event pronoun resolution and event noun phrases resolution. This paper studies the influence of semantic role on noun phrases＇ event anaphora resolution system, and based on SVM machine learning method, carries out English event＇s anaphora resolution, and through adding the elements of time and place in the tuple （semantic role） of computing events＇ semantic similarity, improves the performance of event anaphora resolution system. The experimental results of OntoNotes 4.0 corpus show that both the accuracy rate and F value of the system are somewhat increased after introducing the improved semantic features compared with the benchmark system, and verifies the positive influence of time and place elements on event anaphora resolution.
出处
《山西科技》
2016年第1期109-111,113,共4页
Shanxi Science and Technology
基金
山西省自然科学基金项目"基于框架语义标注的中文篇章指代消解策略研究"(No.2012011011-2)
关键词
事件指代消解
语义角色
特征提取
机器学习
语料库
event anaphora resolution
semantic role
feature extraction
machine learning
corpus
