What do Language Models know about word senses? Zero-Shot WSD with Language Models and Domain Inventories

TL;DR

This paper investigates how well language models like BERT and RoBERTa can perform zero-shot word sense disambiguation by framing it as a textual entailment task based on domain relations, showing promising results.

Contribution

It introduces a novel zero-shot WSD method using domain-based textual entailment prompts, demonstrating effectiveness comparable to supervised approaches.

Findings

Language models can effectively perform zero-shot WSD.

Framing WSD as a textual entailment task improves disambiguation accuracy.

The approach approaches the performance of supervised systems.

Abstract

Language Models are the core for almost any Natural Language Processing system nowadays. One of their particularities is their contextualized representations, a game changer feature when a disambiguation between word senses is necessary. In this paper we aim to explore to what extent language models are capable of discerning among senses at inference time. We performed this analysis by prompting commonly used Languages Models such as BERT or RoBERTa to perform the task of Word Sense Disambiguation (WSD). We leverage the relation between word senses and domains, and cast WSD as a textual entailment problem, where the different hypothesis refer to the domains of the word senses. Our results show that this approach is indeed effective, close to supervised systems.