隐喻是语言中的一种普遍现象，反映了人们思维活动的特征。长期以来，隐喻在语言 研究中处于较为边缘的地位，同时也是制约自然语言理解效果的重要因素。随着语言研究 的深入、认知科学的发展和人工智能研究的再次繁荣，隐喻现象在自然语言中扮演的重要 角色逐渐被人们所重视。按照词性划分，隐喻可分为名词性隐喻、动词性隐喻、形容词性 隐喻等，其中，动词性隐喻的形成多与动词搭配相关，当动词同其逻辑上的主语或宾语在 字面的语义理解上产生矛盾冲突时，则可能产生动词性隐喻。经过长期的使用和演变，一 些经常使用和被人们普遍接受的动词隐喻用法逐渐固定下来，动词在这些语境中的意义逐 渐演化为动词本身的义项，因此通过隐喻进行引申是现代汉语中动词引申义产生的一种重 要方式。动词引申义同基本义的差别对不同语境下的动词搭配有较大影响，搭配词在一定 程度上可以反映动词在当前语境中的意义。 基于动词的引申义与基本义的差异，本文 出一种通过计算动词上下文搭配词的词向 量相似度进行动词性隐喻识别的方法，先从短语层面研究动词性隐喻，再将动词性短语的 隐喻判别结果移植到句子的识别中。首先，根据动词性隐喻短语的产生条件，分析出动词 性隐喻可能出现在主谓短语和动宾短语中。同时本文分析了动词性隐喻与动词义项类别 (基本义、引申义、隐喻义)的关系，认为动词的基本义搭配倾向于形成常规用法，而引 申义的搭配倾向于形成隐喻。本文据此建立了汉语动词常规搭配词表和动词隐喻搭配词表。 而后，在已有的隐喻计算相关理论和神经网络模型工具的基础上，利用大规模网络语料训 练的词向量，将待判别的短语搭配词与词表中的搭配词进行词向量相似度计算和比对，最 后根据相似度与设定的阈值的大小关系判别短语是否为动词性隐喻，从而实现动词性隐喻 的自动识别。 本文抽取了 CCL2018 动词隐喻识别评测任务中的 100 条数据对该方法进行检验，实 验结果中短语的总体识别准确率达到了 86.26%，隐喻短语的识别准确率为 78.00%;句子 的隐喻识别准确率为 90.70%，召回率为 78.00%。实验证明，通过动词上下文搭配词的词 向量相似度进行动词性隐喻识别的方法是有效的，也验证了依据义项类别建立的动词搭配 词表可以有效应用于动词隐喻识别任务。另外，通过依存分析 取句子的主谓关系和动宾 关系，可将短语层面的动词性隐喻识别方法有效地移植到句子层面的动词性隐喻识别中。

Metaphorisa commonphenomenoninlanguage, which reflectsthecharacteristicsofpeople's thinking activities. For a long time, metaphor which is an important factor restricting the effect of natural language understanding has been placed in a relatively marginal position in language research. With the deepening of language research, the developing of cognitive science and the thriving of artificial intelligence research, the important role of metaphor in natural language has gradually been valued. According to parts of speech, metaphor can be divided into nominal metaphor, verbal metaphor, adjective metaphor, and etc. Among them, the formation of verbal metaphor is usually related to collocations of verbs. Verbal metaphor may generate when verb conflicts with its logical subject or object literally. After a long period of use and evolution, the metaphorical usage of some verbs, which are often used and generally accepted by people, has gradually been fixed, and the meaning of verbs in these contexts has gradually evolved into the meaning of verbs themselves. Therefore, metaphorical process is an important way for the generation of verb extended meaning in modern Chinese. The difference between extended meaning and basic meaning of a verb has a great influence on collocations of the verb in different contexts, and collocating words can reflect the meaning of the verb to some extent. Based on the difference between the extended meaning and the basic meaning of verbs, this paper proposes a method of verb metaphor recognition by calculating the similarity of word embedding of collocating words in context, and we study verbal metaphor from the phrase level, and then transplant the metaphorical discrimination results of verbal phrases into sentence recognition. First, according to the generating conditions of verbal metaphorical phrases, it is analyzed thattheymayappear inphraseswithsubject-predicatestructureandverb-objectstructure. At the same time, this paper analyzes the relationship between verbal metaphor and verb meaning categories (basic meaning, extended meaning, and metaphorical meaning), and we believe that basic meaning collocations of a verb tend to form conventional usages, while the collocations of extended meaning tend to form metaphor. Based on this, this paper established a list of Chinese conventional collocations of verbs and a list of metaphorical collocations of verbs. At the same time, word vectors are trained using large-scale network corpus based on existing theories of metaphor calculation and neural network model tools. Finally, we calculate and compare similarity II of word vectors between the word in the phrase to be identified and collocating words of the verb in vocabularies. Then, we judge whether a phrase is a verbal metaphor according to the relationship between similarity and the set threshold, and in this way, verbal metaphor can be recognized automatically. This paper extracts 100 pieces of data of verb metaphor recognition evaluation task in CCL2018 to test the method. The overall recognition accuracy of phrases in the experimental reached 86.26%, and the recognition accuracy of metaphorical phrases is 78.00%. The accuracy of sentences is 90.70%, and the recall rate is 78.00%. According to the experiment, it is effective that identifying verbal metaphor through the similarity of word vectors of collocating words in context, and verb collocation vocabularies based on meaning categories can be effectively applied in recognition of verbal metaphor. In addition, the phrasal verbal metaphor recognition method can be effectively transplanted into verbal metaphor recognition in sentences by extracting the subject-predicate relations and verb-object relations of a sentence through dependency analysis.