Eliciting Knowledge from Large Pre-Trained Models for Unsupervised Knowledge-Grounded Conversation
This addresses the challenge of generating informed conversations without supervision, though it appears incremental in improving existing techniques.
The paper tackled the problem of leveraging large pre-trained models as knowledge bases for unsupervised knowledge-grounded conversation, proposing methods to elicit and exploit generated knowledge, with empirical results showing advantages over state-of-the-art methods on two benchmarks.
Recent advances in large-scale pre-training provide large models with the potential to learn knowledge from the raw text. It is thus natural to ask whether it is possible to leverage these large models as knowledge bases for downstream tasks. In this work, we answer the aforementioned question in unsupervised knowledge-grounded conversation. We explore various methods that best elicit knowledge from large models. Our human study indicates that, though hallucinations exist, large models post the unique advantage of being able to output common sense and summarize facts that cannot be directly retrieved from the search engine. To better exploit such generated knowledge in dialogue generation, we treat the generated knowledge as a noisy knowledge source and propose the posterior-based reweighing as well as the noisy training strategy. Empirical results on two benchmarks show advantages over the state-of-the-art methods.