【作者】 李剑锋；

【导师】 周昌乐；

【作者基本信息】 厦门大学 ， 计算机应用技术， 2008， 硕士

【摘要】 隐喻是思维及语言的中心问题,近几年隐喻计算研究开始受到人工智能研究者的关注。在隐喻现象中,有一大类现象表现出来的特征是搭配关系的超常性,如“My car drinks gasoline”(我的汽车喝汽油)中“喝汽油”就是一种异于常规的超常搭配,而这样一种超常搭配现象构成了隐喻表达的一个重要组成部分。本文对隐喻语言中的异常搭配现象进行了研究,提出了基于实例的汉语语义超常搭配识别的计算模型,并完成了系统实现。在此基础之上将汉语语义超常搭配识别模型应用到汉语隐喻识别系统当中,取得了较好的效果。本文的研究工作主要包括以下几个方面:1.隐喻及汉语语义超常搭配计算研究背景分析及相关研究综述。介绍了汉语语义搭配及隐喻识别现有计算思想和方法,并在此基础上分析了几种方法之间的差异和优劣。2.超常搭配的语言学分析。从语言学的角度对汉语超常搭配进行全面的分析和总结,给出了超常搭配的定义,对常见的超常搭配现象进行了分类,并详细讨论了超常搭配的语言与认知特征。3.汉语语义超常搭配计算模型的提出与实现。通过基于统计的生成方法得到汉语常规搭配实例库,将其作为已知的搭配实例。在此基础之上,定义了语义搭配超常度量及其计算方法,并完成了相关实验及结果分析。4.超常搭配自动发现算法在隐喻计算模型中的应用。设计并构建了面向计算的汉语隐喻语料库,提出了面向隐喻识别的嵌入式隐喻依存结构模式匹配算法,结合语义超常搭配度量的计算,可以完成相关的隐喻搭配的识别。本文首次从计算角度研究汉语中语义超常搭配现象,创造性地提出基于实例的汉语语义超常搭配识别计算模型,并将其应用到汉语隐喻的计算模型当中,实验表明我们的研究和所提出的方法取得了比较好的效果,为汉语语义超常搭配的计算研究和汉语隐喻计算模型的基础研究都作出了积极的贡献。更多还原

【Abstract】 Metaphor, an ordinary everyday phenomenon of language use, has become the focus of mind and language mechanism. The comprehension of metaphor by machine will be a bottleneck problem in natural language understanding and machine translation. Therefore, great attention has been paid to computational mechanisms of metaphor within the last few decades.This paper researches on classified recognition of unconventional semantic collocations, which is one of most important types of metaphor recognition. Proposed an example-based classified recognition model of unconventional semantic collocations, and achieved a Chinese Metaphor Recognition System.Our research includes the following aspects:1. Introduced the background and significance of our research. And made a summarization of existing research results in this field, including semantic collocation and metaphor recognition.2. Made the semantic analysis and introduced the definition, classification and characteristics of unconventional collocations.3. Proposed the example-based classified recognition model of unconventional semantic collocations. Introduced the basic idea of this model, and the method of constructing the semantic example model bank. As were as given the experimental results of precision and recall.4. The application of semantic unconventional collocations used in Chinese metaphor computation. Introduced the Metaphor Role Dependency Schema, the construction of Chinese Metaphor Corpus, the embedded tree matching algorithm based on metaphorical dependency structure, and the design of Chinese Metaphor Recognition System.Our research of semantic unconventional collocations is a good innovation, and according to the experimental results, shows that the method is feasible.更多还原