汉语隐喻的语言形式特征及其对隐喻机器理解研究的影响

Linguistic Characteristics of Chinese Metaphors and the Impact on Metaphor Computation

语言学界对隐喻的研究集中于从概念层面对隐喻进行认知分析,而淡化了对语言表达层面隐喻的典型特征做形式化的概括。然而,概念层面的隐喻却必须通过语言形式层面的隐喻表达来加以体现,隐喻的语言形式对隐喻的理解也具有极其重要的意义。特别是,当隐喻作为一种思维现象被纳入机器理解的研究范畴之后,隐喻的计算机制研究也开始要求高度地形式化隐喻的语言表达。鉴于此,本文面向隐喻机器理解,采用语料统计分析的方法,从大量实际隐喻语料出发,总结汉语隐喻的语言形式的复杂性特征:语言隐喻主要成分取值的不确定性、语言隐喻各成分出现与否的不确定性、隐喻嵌套和递进复用现象的普遍性,进而指出面向计算的隐喻研究困境以及出路。

Present research on metaphor is focused on the cognitive view and inside the frame of conceptual metaphor analysis. As a result linguistic characteristics of metaphor expressions are sometimes ignored. However conceptual metaphors finally appear as linguistic forms which also play a significant role in metaphor understanding. Especially when metaphor research is brought into computational view it is highly required to formalize linguistic metaphors. At this point this paper investigates various metaphor phenomena by means of statistic corpus analysis. Oriented to machine-understanding of metaphor this paper then proposes three representative complexities in linguistic metaphors: it is uncertain that topics, vehicles and grounds willappear as a word, a phrase, a sentence or other linguistic units; it is uncertain whether topics, grounds or tags will appear in one metaphor at the same time; it is universal that a metaphor nested in another metaphor or inherited in another metaphor. Then troubles and possible solutions in metaphor computation are discussed.