Verbal Multiword Expressions for Identification of Metaphor

Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 2890–2895 July 5 - 10, 2020. c⃝2020 Association for Computational Linguistics 2890 Verbal Multiword Expressions for Identification of Metaphor Omid Rohanian†, Marek Rei§ ‡, Shiva Taslimipoor‡, Le An Ha† †RGCL, University of Wolverhampton, United Kingdom ‡ALTA Institute, University of Cambridge, United Kingdom §Department of Computing, Imperial College London, United Kingdom {omid.rohanian, l.a.ha}@wlv.ac.uk marek.rei@imperial.ac.uk, st797@cl.cam.ac.uk Abstract Metaphor is a linguistic device in which a con- cept is expressed by mentioning another. Iden- tifying metaphorical expressions, therefore, re- quires a non-compositional understanding of semantics. Multiword Expressions (MWEs), on the other hand, are linguistic phenomena with varying degrees of semantic opacity and their identification poses a challenge to com- putational models. This work is the first at- tempt at analysing the interplay of metaphor and MWEs processing through the design of a neural architecture whereby classification of metaphors is enhanced by informing the model of the presence of MWEs. To the best of our knowledge, this is the first “MWE-aware” metaphor identification system paving the way for further experiments on the complex inter- actions of these phenomena. The results and analyses show that this proposed architecture reach state-of-the-art on two different estab- lished metaphor datasets. 1 Introduction Human language is rife with a wide range of tech- niques that facilitate communication and expand the capacities of thinking and argumentation. One phenomenon of such kind is metaphor. Metaphor is defined as a figure of speech in which the speaker makes an implicit comparison between seemingly unrelated things which nonetheless have certain common characteristics (Shutova, 2010). This is done to convey an idea which is otherwise difficult to express succinctly or simply for rhetorical effect. As an example, in the sentence she devoured his novels, the verb devour is used in a metaphorical sense that implies reading quickly and eagerly. The literal and metaphorical senses share the element of intense desire which in turn helps to decode the meaning of the word in its context. It is clear that a mere literal understanding of se- mantics would not result in proper understanding of a metaphorical expression and a non-compositional approach would be required (Shutova et al., 2013; Vulchanova et al., 2019). The human brain is equipped with the necessary machinery to decode the intended message behind a metaphorical utter- ance. This involves mentally linking the seemingly unrelated concepts based on their similarities (Rapp et al., 2004). Verbal MWEs (VMWEs) are another example of non-literal language in which multiple words form a single unit of meaning. These two phenomena share some common ground. Expressions like take the bull by the horns, go places, kick the bucket, or break someone’s heart can be categorised as metaphorical VMWEs. Based on this observa- tion we hypothesise that a metaphor classification model can be bolstered by knowledge of VMWEs. In this work we focus on how identification of verbal metaphors can be helped by verbal MWEs. We devise a deep learning model based on attention-guided graph convolutional …

Content truncated to 500 words