Research On The Integration Of Metaphor Detection And Interpretation

As a common rhetorical device,metaphors are frequently used to enhance language expression and sensual experience,making receivers resonate with speaker.In addition to metaphor's rhetorical function,metaphor is also a kind of cognitive mechanism and the foundation of human conceptual system.Due to both the rhetorical and cognitive function,metaphors stay active in various application scenarios of natural language processing.Meanwhile a large amount of implicit metaphorical expressions have caused difficulties which researchers of natural language processing have to inevitably face.Metaphor computing consists of two main branches:metaphor detection and metaphor interpretation.The aim at metaphor detection is to distinguish metaphorical utterances from literal ones.Metaphor interpretation aims at transforming metaphorical utterances into equivalent expressions which are friendlier to machine natural language processing.Metaphor computing systems should have demonstrated consistency,however,current research generally focus on one aspect of metaphor computing,leading to the gap between metaphor detection and interpretation.Analyzing the deficiencies in current researches,this paper proposes a framework for the integration of metaphor detection and interpretation.We respectively build a metaphor detection and interpretation model which can be easily implemented into the framework,so as to realize the integration of metaphor computing.We think that existing metaphor detection models based on binary classification can only label sentences or phrases'metaphoricity,lacking the ability to detect metaphor in word level and failing to provide key information necessary for metaphor interpretation.In this paper,metaphor detection is considered as a sentence tagging task,through which we can obtain each word's metaphoricity.Based on the pretraining language model,the detection model can strongly support metaphor interpretation.In aspect of metaphor interpretation,we focus on English sentential verbal and adjective metaphors.Extending theory of interaction and cooperative mechanism in metaphor research,we propose the latent cooperative theory.We model the relation of metaphorical words,candidate substitutes and context based on the latent cooperative relation.By constructing latent association of different semantic domains,our model chooses paraphrase with the strongest cooperative willingness as the substitutive interpretation.Our interpretation model is interpretable.In summary,this paper design and implement metaphor detection and interpretation model respectively.Based on the metaphor detection and interpretation model,we construct the integrated system of metaphor computing.We conduct experiments on metaphor detection and interpretation.This paper fills the gap of related metaphor computing research and offer thinking for future related works.