Hierarchical Information Matters: Text Classification via Tree Based Graph Neural Network

Proceedings of the 29th International Conference on Computational Linguistics, pages 65–77 October 12–17, 2022. 65 Metaphorical Polysemy Detection: Conventional Metaphor meets Word Sense Disambiguation Rowan Hall Maudslay1,2 Simone Teufel1 1Dept. of Computer Science & Technology University of Cambridge2Magdalene College {rh635,sht25}@cam.ac.uk Abstract Linguists distinguish between novel and con- ventional metaphor, a distinction which the metaphor detection task in NLP does not take into account. Instead, metaphoricity is formu- lated as a property of a token in a sentence, regardless of metaphor type. In this paper, we investigate the limitations of treating con- ventional metaphors in this way, and advocate for an alternative which we name metaphori- cal polysemy detection (MPD). In MPD, only conventional metaphoricity is treated, and it is formulated as a property of word senses in a lexicon. We develop the first MPD model, which learns to identify conventional metaphors in the English WordNet. To train it, we present a novel training procedure that combines metaphor detection with word sense disambiguation (WSD). For evaluation, we manually annotate metaphor in two subsets of WordNet. Our model significantly outperforms a strong baseline based on a state-of-the-art metaphor detection model, attaining an ROC- AUC score of .78 (compared to .65) on one of the sets. Additionally, when paired with a WSD model, our approach outperforms a state- of-the-art metaphor detection model at identi- fying conventional metaphors in text (.659 F1 compared to .626). 1 Introduction Linguists differentiate between two types of metaphor: novel and conventional. While novel metaphors are creative expressions made in a partic- ular situation by one particular individual, conven- tional metaphors are those which have been widely adopted by a language community. Consider: (1) The attack began at dawn. (2) The government has come under attack. (3) The government torpedoed the housing bill. Example (1) is a literal usage of attack which refers to a military offensive. Example (2) is a conven- tional metaphor referring to intense verbal criticism. "The new task is a bridge between existing ones" Metaphorical MPD WSD SMD bridge noun3 something resembling a bridge in form or function, e.g. "his letters provided a bridge across the centuries" Figure 1: Metaphorical polysemy detection is the miss- ing link between standard metaphor detection and word sense disambiguation Example (3) is a novel metaphor, which reuses the well-established imagery from (2), but extends it with the word torpedoed. Detecting novel metaphors and conventional metaphors are fundamentally different tasks. Novel metaphors are creative usages of words, which vi- olate statistical patterns of language (e.g. Wilks, 1975, 1978), and which extend the meaning of a word into unexpected areas of semantic space. Conventional metaphors, on the other hand, can be considered lexicalised: examples (1) and (2) come from the gloss of two senses of attack in WordNet (Miller, 1995), where the novel sense of torpedoed in example (3) is not yet captured. Despite these inherent differences, in NLP the stan- dard metaphor detection task, abbreviated as SMD here, makes no distinction between novel and con- ventional metaphor. Although recent works …

Content truncated to 500 words