CATE: A Contrastive Pre-trained Model for Metaphor Detection with Semi-supervised Learning

Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 3888–3898 November 7–11, 2021. c⃝2021 Association for Computational Linguistics 3888 CATE: A Contrastive Pre-trained Model for Metaphor Detection with Semi-supervised Learning Zhenxi Lin1,2, Qianli Ma1*, Jiangyue Yan1, Jieyu Chen3 1School of Computer Science and Engineering, South China University of Technology, Guangzhou, China 2Tencent Jarvis Lab, Shenzhen, China 3Department of English and Communication, The Hong Kong Polytechnic University zhenxi_lin@foxmail.com, qianlima@scut.edu.cn jiangyue9606@gmail.com, 18043507r@connect.polyu.hk Abstract Metaphors are ubiquitous in natural language, and detecting them requires contextual reason- ing about whether a semantic incongruence ac- tually exists. Most existing work addresses this problem using pre-trained contextualized models. Despite their success, these mod- els require a large amount of labeled data and are not linguistically-based. In this pa- per, we proposed a ContrAstive pre-Trained modEl (CATE) for metaphor detection with semi-supervised learning. Our model first uses a pre-trained model to obtain a contex- tual representation of target words and em- ploys a contrastive objective to promote an increased distance between target words’ lit- eral and metaphorical senses based on linguis- tic theories. Furthermore, we propose a sim- ple strategy to collect large-scale candidate in- stances from the general corpus and general- ize the model via self-training. Extensive ex- periments show that CATE achieves better per- formance against state-of-the-art baselines on several benchmark datasets. 1 Introduction Conceptual metaphors are figurative languages widely used in our daily communication, im- plying a mapping between two conceptual do- mains (Lakoff and Johnson, 2008). At a linguistic level, metaphor is defined as a linguistic expres- sion representing other concepts rather than taking literal meanings of words in context (Lagerwerf and Meijers, 2008). For instance, in the sentence “I have digested all this information,” the word di- gested does not literally mean converting food into absorbable substances. Instead, this word means “arrange and integrate in the mind” in the con- text.1 This metaphor conceptualizes the concept of ideas in terms of the properties of food. Metaphor- ical associations as such are broad generalizations *Corresponding author 1http://wordnetweb.princeton.edu/perl/ webwn?s=digest that allow us to project knowledge and inferences across domains and are beneficial for various down- stream NLP applications, such as machine transla- tion (Shi et al., 2014), sentiment analysis (Cambria et al., 2017; Dankers et al., 2019), and dialogue systems (Dybala and Sayama, 2012). Given the prevalence of metaphors in hu- man communication, the effective detection of metaphors plays an essential role in natural lan- guage understanding. Hence, many efforts have been devoted to metaphor detection (MD), which aims to identify metaphorical expressions in a text automatically. Most previous methods (Mason, 2004; Turney et al., 2011; Tsvetkov et al., 2014; Shutova et al., 2016) for MD are based on var- ious hand-crafted linguistic features and rely on manually annotated resources to extract them. Re- cently, significant progress has been made in ap- plying deep learning techniques for MD (Wu et al., 2018; Gao et al., 2018; Mao et al., 2019; Rohanian et al., 2020; Le et al., 2020). These methods …

Content truncated to 500 words