摘要
隐喻理解算法层出不穷,但缺乏统一的评价标准与自动评价机制.名词性隐喻理解多角度的特点使常用的自动评价方法不能很好地植入名词性隐喻理解的评价中.为了适应名词性隐喻理解有别于其他自然语言处理领域的特点,亟需研究一种新的方法评价名词性隐喻理解结果.首先介绍名词性隐喻的多角度与结构相似的特点,然后提出符合人类认知的隐喻理解评价标准与可解释的动态评价方法,最后在500句无上下文的名词性隐喻理解结果上进行实验.实验结果表明该方法合理有效,具备与人类相仿的评价能力.
Although there are various algorithms for metaphor interpretation,no unified evaluation criteria or automatic evaluation systems exist.The interpretation of nominal metaphor is multi-angle,which renders common automatic evaluation methods difficult to be implemented in this field.In order to m eet the unique requests of nominal metaphor interpretation,we need a new method for evaluation.First,the characteristics multi-angles and cross-domain structural similarities are introduced.Then an evaluation criterion and an explainable dynamic evaluation method are proposed.After we evaluate 500 context-free Chinese nominal metaphors and their interpretations,experimental results show that the method shares similar scores with human judges,and thus is reasonable and effective.
出处
《厦门大学学报（自然科学版）》
CAS
CSCD
北大核心
2017年第5期718-723,共6页
Journal of Xiamen University：Natural Science
基金
国家自然科学基金(61075058)
关键词
隐喻理解
动态评价
多角度
结构相似
解释
metaphor interpretation
dynamic evaluation
multi-angle
structural similarity
explanation
