Grasping the Finer Point: A Supervised Similarity Network for Metaphor Detection

Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing, pages 1537–1546 Copenhagen, Denmark, September 7–11, 2017. c⃝2017 Association for Computational Linguistics Grasping the Finer Point: A Supervised Similarity Network for Metaphor Detection Marek Rei♣♠ Luana Bulat♣ Douwe Kiela♦ Ekaterina Shutova♣ ♣Computer Laboratory, University of Cambridge, United Kingdom ♠The ALTA Institute, University of Cambridge, United Kingdom ♦Facebook AI Research, New York, USA {marek.rei,luana.bulat,ekaterina.shutova}@cl.cam.ac.uk, dkiela@fb.com Abstract The ubiquity of metaphor in our every- day communication makes it an impor- tant problem for natural language under- standing. Yet, the majority of metaphor processing systems to date rely on hand- engineered features and there is still no consensus in the field as to which features are optimal for this task. In this paper, we present the first deep learning archi- tecture designed to capture metaphorical composition. Our results demonstrate that it outperforms the existing approaches in the metaphor identification task. 1 Introduction Metaphor is pervasive in our everyday commu- nication, enriching it with sophisticated imagery and helping us to reconcile our experience in the world with our conceptual system (Lakoff and Johnson, 1980). In the most influential account of metaphor to date, Lakoff and Johnson explain the phenomenon through the presence of system- atic metaphorical associations between two dis- tinct concepts or domains. For instance, when we talk about “curing juvenile delinquency” or “corruption transmitting through the government ranks”, we view the general concept of crime (the target concept) in terms of the properties of a dis- ease (the source concept). Such metaphorical as- sociations are broad generalisations that allow us to project knowledge and inferences across do- mains; and our metaphorical use of language is a reflection of this process. Given its ubiquity, metaphorical language poses an important problem for natural language un- derstanding (Cameron, 2003; Shutova and Teufel, 2010). A number of approaches to metaphor pro- cessing have thus been proposed, focusing pre- dominantly on classifying linguistic expressions as literal or metaphorical. They experimented with a range of features, including lexical and syntac- tic information (Hovy et al., 2013; Beigman Kle- banov et al., 2016) and higher-level features such as semantic roles (Gedigian et al., 2006), domain types (Dunn, 2013), concreteness (Turney et al., 2011), imageability (Strzalkowski et al., 2013) and WordNet supersenses (Tsvetkov et al., 2014). While reporting promising results, all of these ap- proaches used hand-engineered features and re- lied on manually-annotated resources to extract them. In order to reduce the reliance on manual annotation, other researchers experimented with sparse distributional features (Shutova et al., 2010; Shutova and Sun, 2013) and dense neural word embeddings (Bracewell et al., 2014; Shutova et al., 2016). Their experiments have demonstrated that corpus-driven lexical representations already en- code information about semantic domains needed to learn the patterns of metaphor usage from lin- guistic data. We take this intuition a step further and present the first deep learning architecture designed to capture metaphorical composition. Deep learn- ing methods have already been shown success- ful in many other semantic tasks (e.g. Hermann et …

Content truncated to 500 words