Connecting the Dots: A Knowledgeable Path Generator for Commonsense Question Answering

Findings of the Association for Computational Linguistics: EMNLP 2020, pages 388–406 November 16 - 20, 2020. c⃝2020 Association for Computational Linguistics 388 Contextual Modulation for Relation-Level Metaphor Identification Omnia Zayed, John P. McCrae, Paul Buitelaar Insight SFI Research Centre for Data Analytics Data Science Institute National University of Ireland Galway IDA Business Park, Lower Dangan, Galway, Ireland {omnia.zayed, john.mccrae, paul.buitelaar}@insight-centre.org Abstract Identifying metaphors in text is very challeng- ing and requires comprehending the under- lying comparison. The automation of this cognitive process has gained wide attention lately. However, the majority of existing ap- proaches concentrate on word-level identifica- tion by treating the task as either single-word classification or sequential labelling without explicitly modelling the interaction between the metaphor components. On the other hand, while existing relation-level approaches im- plicitly model this interaction, they ignore the context where the metaphor occurs. In this work, we address these limitations by in- troducing a novel architecture for identifying relation-level metaphoric expressions of cer- tain grammatical relations based on contex- tual modulation. In a methodology inspired by works in visual reasoning, our approach is based on conditioning the neural network com- putation on the deep contextualised features of the candidate expressions using feature-wise linear modulation. We demonstrate that the proposed architecture achieves state-of-the-art results on benchmark datasets. The proposed methodology is generic and could be applied to other textual classification problems that benefit from contextual interaction. 1 Introduction Despite its fuzziness, metaphor is a fundamental feature of language that defines the relation be- tween how we understand things and how we ex- press them (Cameron and Low, 1999). A metaphor is a figurative device containing an implied map- ping between two conceptual domains. These do- mains are represented by its two main components, namely the tenor (target domain) and the vehicle (source domain) (End, 1986). According to the conceptual metaphor theory (CMT) of Lakoff and Johnson (1980), which we adopt in this work, a concept such as “liquids” (source domain/vehicle) can be borrowed to express another such as “emo- tions” (target domain/tenor) by exploiting single or common properties. Therefore, the conceptual metaphor “Emotions are Liquids” can be mani- fested through the use of linguistic metaphors such as “pure love”, “stir excitement” and “contain your anger”. The interaction between the target and the source concepts of the expression is impor- tant to fully comprehend its metaphoricity. Over the last couple of years, there has been an increasing interest towards metaphor process- ing and its applications, either as part of natural language processing (NLP) tasks such as machine translation (Koglin and Cunha, 2019), text sim- plification (Wolska and Clausen, 2017; Clausen and Nastase, 2019) and sentiment analysis (Ren- toumi et al., 2012) or in more general discourse analysis use cases such as in analysing political discourse (Charteris-Black, 2011), financial report- ing (Ho and Cheng, 2016) and health communica- tion (Semino et al., 2018). Metaphor processing comprises several tasks including identification, interpretation and cross- domain mappings. Metaphor identification is the most studied among these tasks. It is …

Content truncated to 500 words