Coherence boosting: When your pretrained language model is not paying enough attention
This addresses the challenge of insufficient attention to distant words in language models, which is crucial for improving coherence in automatic language generation and understanding, though it appears incremental as it builds on existing pretrained models.
The authors tackled the problem of long-range semantic coherence in language models by introducing coherence boosting, an inference procedure that enhances focus on distant context, resulting in improved performance in text generation, dialog responses, and zero-shot NLP tasks without extra training.
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.