MelBERT: Metaphor Detection via Contextualized Late Interaction using Metaphorical Identification Theories

Abstract

Automated metaphor detection is a challenging task to identify the metaphorical expression of words in a sentence. To tackle this problem, we adopt pre-trained contextualized models, e.g., BERT and RoBERTa. To this end, we propose a novel metaphor detection model, namely metaphor-aware late interaction over BERT (MelBERT). Our model not only leverages contextualized word representation but also benefits from linguistic metaphor identification theories to detect whether the target word is metaphorical. Our empirical results demonstrate that MelBERT outperforms several strong baselines on four benchmark datasets, i.e., VUA-18, VUA-20, MOH-X, and TroFi.

FrameBERT: Conceptual Metaphor Detection with Frame Embedding Learning

“…Similar to the prior work (Choi et al, 2021;Song et al, 2021), we develop the sentence encoder to produce the sentence encoding v S , the contextualised encoding for the target word v S,t , as well as isolated encoding for the target word v t . Formally, given an input sequence S = (w 0 , ..., w n ), RoBERTa (Liu et al, 2019) encodes each word into a set of contextualised embedding vectors H = (h cls , h 0 , ..., h n ):…”

“…Recent studies in this field have demonstrated its potential to positively impact various Natural Language Processing (NLP) applications, including sentiment analysis (Cambria et al, 2017;Li et al, 2022a), metaphor generation (Tang et al, 2022;Li et al, 2022b,c), and mental health care (Abd Yusof et al, 2017;Gutiérrez et al, 2017). Different strategies have been proposed for modeling relevant context, including employing limited * Corresponding author linguistic context such as subject-verb and verbdirect object word pairs (Gutiérrez et al, 2016), incorporating a wider context encompassing a fixed window surrounding the target word (Do Dinh and Gurevych, 2016;Mao et al, 2018), and considering the complete sentential context (Gao et al, 2018;Choi et al, 2021). Some recent efforts attempt to improve context modelling by explicitly leveraging the syntactic structure (e.g., dependency tree) of a sentence in order to capture important context words, where the parse trees are typically encoded with graph convolutional neural networks (Le et al, 2020;Song et al, 2021).…”

An In-depth Analysis of Implicit and Subtle Hate Speech Messages

“…In our extensive study on hate speech datasets, we identified 18 typical properties of implicitness relying on linguistics features (listed in Table 2). Among the most significant ones we have irony (Frenda et al, 2022), sarcasm (Potamias et al, 2020, black humor, metaphor (Choi et al, 2021;Gao et al, 2018), exaggeration (Troiano et al, 2018), rhetorical question, sentiment (Li et al, 2021), inference, lack of context (Dadvar et al, 2013), and lack of extralinguistic knowledge. Most of the implicit HS messages contain combinations of several properties, as in: 2.…”

“…Fallacya false or mistaken idea; an often plausible argument using false or invalid inference (Merriam-Webster, 2022) Humiliationthe embarrassment and shame a person feels when someone makes them appear stupid or when they make a mistake in public (Dictionary, 2022) Inferencesomething that is inferred. The premises and conclusion of a process of inferring (Merriam-Webster, 2022) Ironythe use of words to express something other than and especially the opposite of the literal meaning; incongruity between the actual result of a sequence of events and the normal or expected result (Potamias et al, 2020) Metaphora figure of speech in which a word or phrase literally denoting one kind of object or idea is used in place of another to suggest a likeness or analogy between them (Choi et al, 2021;Gao et al, 2018) Metonymya figure of speech consisting of the use of the name of one thing for that of another of which it is an attribute or with which it is associated (such as "crown" in "lands belonging to the crown") (Merriam-Webster, 2022) Rhetorical questiona question not intended to require an answer, used mainly for dramatic effect (Frank, 1990) Sarcasma mode of satirical wit depending on its effect on bitter, caustic, and often ironic language usually directed against an individual. Sarcasm differs from irony with one distinct characteristic: negativity.…”

FrameBERT: Conceptual Metaphor Detection with Frame Embedding Learning

“…Recent studies in this field have demonstrated its potential to positively impact various Natural Language Processing (NLP) applications, including sentiment analysis (Cambria et al, 2017;Li et al, 2022a), metaphor generation (Tang et al, 2022;Li et al, 2022b,c), and mental health care (Abd Yusof et al, 2017;Gutiérrez et al, 2017). Different strategies have been proposed for modeling relevant context, including employing limited linguistic context such as subject-verb and verbdirect object word pairs (Gutiérrez et al, 2016), incorporating a wider context encompassing a fixed window surrounding the target word (Do Dinh and Gurevych, 2016;Mao et al, 2018), and considering the complete sentential context (Gao et al, 2018;Choi et al, 2021). Some recent efforts attempt to improve context modelling by explicitly leveraging the syntactic structure (e.g., dependency tree) of a sentence in order to capture important context words, where the parse trees are typically encoded with graph convolutional neural networks (Le et al, 2020;Song et al, 2021).…”