Neural Metaphor Detection with Visibility Embeddings

Proceedings of the 10th Conference on Lexical and Computational Semantics, pages 222–228 August 5–6, 2021, Bangkok, Thailand (online) ©2021 Association for Computational Linguistics 222 Neural Metaphor Detection with Visibility Embeddings Gitit Kehat and James Pustejovsky Department of Computer Science Brandeis University Waltham, MA 02453 USA {gititkeh,jamesp}@brandeis.edu Abstract We present new results for the problem of se- quence metaphor labeling, using the recently developed Visibility Embeddings. We show that concatenating such embeddings to the in- put of a BiLSTM obtains consistent and signif- icant improvements at almost no cost, and we present further improved results when visibil- ity embeddings are combined with BERT. 1 Introduction When browsing through vision-language datasets, one can make the intuitive observation that their textual parts (“visual corpora”) contain more phys- ical language, mostly descriptive, which tends to be non-metaphorical by nature (See, for example, typical images from the Visual Genome dataset in Figure 1). Recently, this property was used to build visibility embeddings, which aim to provide a good estimation of a word’s concreteness, a feature that has been long related to metaphoricity (Lakoff and Johnson, 1980; Turney et al., 2011). Many metaphors indeed involve noticeable dif- ferences between the abstractness of words con- structing them, like “clean conscience” (vs. “clean air”). Metaphors are not created in isolation, com- monly do not stand alone as non-literal expressions, and are highly context-dependent in nature. Even the most concrete and physical text can be consid- ered as metaphorical when mentioned in a different context than its original one, or in proximity to an- other text from the target domain. For example, a single use of a verb like “push” or “leak” can have both literal and metaphorical meanings, in relation to its context (see Figure 1). Technically, the task of metaphor detection at the sentence level is commonly approached as one of the following two tasks: (1) Sequence Labeling, in which each token in the sentence is classified as either “metaphorical” or Figure 1: Images from the Visual Genome (Krishna et al., 2016) along with their literal (“L”) description, and a metaphorical (“M”) sentence with a similar verb from the MOH-X dataset (Mohammad et al., 2016) (concrete words in green and abstract words in red). “literal” (multiple outputs per sentence). (2) Classification of a specific target word, usually the main verb (one output per sentence). This task is sometimes called “verb classification”. Recently, Kehat and Pustejovsky (2020) pre- sented the simply constructed Visibility Embed- dings (VE), which use references to visual/non- visual corpora to estimate word concreteness, and applied it to the task of verb-classification. In this paper we apply VE also to the sequence labeling task, and show how they consistently improve the result of a BiLSTM model with BERT. We also discuss possible problems when reporting results 223 on very small annotated datasets, and the effect on adding GloVe to the model input. 2 Background and Previous Work 2.1 Visibility Embeddings Visibility embeddings (VE) were shown by (Kehat and Pustejovsky, 2020) to be useful for metaphor …

Content truncated to 500 words