Understanding by Understanding Not: Modeling Negation in Language Models

TL;DR

This paper introduces a method to improve language models' handling of negation by augmenting training with an unlikelihood objective based on negated sentences, significantly reducing errors on negation tasks.

Contribution

It proposes a novel training objective combining likelihood and unlikelihood to enhance negation understanding in pre-trained language models.

Findings

Reduced top-1 error rate to 4% on negated LAMA dataset

Improved performance on negated NLI benchmarks

Demonstrated effectiveness of unlikelihood training for negation

Abstract

Negation is a core construction in natural language. Despite being very successful on many tasks, state-of-the-art pre-trained language models often handle negation incorrectly. To improve language models in this regard, we propose to augment the language modeling objective with an unlikelihood objective that is based on negated generic sentences from a raw text corpus. By training BERT with the resulting combined objective we reduce the mean top~1 error rate to 4% on the negated LAMA dataset. We also see some improvements on the negated NLI benchmarks.