EmbodiedBERT: Cognitively Informed Metaphor Detection Incorporating Sensorimotor Information
Abstract: The identification of metaphor is a crucial prerequisite for many downstream language tasks, such as sentiment analysis, opinion mining, and textual entailment. State-of-the-art systems of metaphor detection require training data annotated based on heuristic principles such as Metaphor Identification Procedure (MIP) (Pragglejaz Group, 2007) and Selection Preference Violation (SPV) (Wilks, 1975; Wilson, 2002). We propose an innovative approach that leverages the cognitive information of embodiment that can be derived from word embeddings, and explicitly models the process of sensorimotor shedding that has been demonstrated as essential for human metaphor processing. We showed that this cognitively motivated module is more effective and can improve the prediction of metaphoricity compared with the heuristic MIP that has been applied previously.
Paper Type: Short
Research Area: Linguistic theories, Cognitive Modeling and Psycholinguistics
Research Area Keywords: linguistic theories; cognitive modeling; computational psycholinguistics
Contribution Types: NLP engineering experiment
Languages Studied: English
Submission Number: 2626
Loading