Coherence boosting: When your pretrained language model is not paying enough attention

TL;DR

This paper introduces coherence boosting, an inference method that enhances large language models' focus on long contexts, improving their performance in language generation and understanding without additional training.

Contribution

The paper proposes a novel inference procedure called coherence boosting that increases LMs' attention to distant words, improving long-range coherence in generated text.

Findings

Coherence boosting improves text coherence in generated language.

It yields performance gains in zero-shot NLP tasks.

No additional training is required for the method.

Abstract

Long-range semantic coherence remains a challenge in automatic language generation and understanding. We demonstrate that large language models have insufficiently learned the effect of distant words on next-token prediction. We present coherence boosting, an inference procedure that increases a LM's focus on a long context. We show the benefits of coherence boosting with pretrained models by distributional analyses of generated ordinary text and dialog responses. It is also found that coherence boosting with state-of-the-art models for various zero-shot NLP tasks yields performance gains with no additional training.