Contrastive Learning Reduces Hallucination in Conversations
This addresses the issue of unreliable responses in conversational AI systems, which is critical for applications like chatbots and virtual assistants, though it is an incremental improvement over existing methods.
The paper tackles the problem of hallucination in pre-trained language models used for conversational systems by proposing MixCL, a contrastive learning scheme that reduces irrelevant or factually incorrect statements. It achieves the highest performance among LM-based dialogue agents on the Wizard-of-Wikipedia benchmark, with comparable results to state-of-the-art KB-based approaches while being more efficient and scalable.
Pre-trained language models (LMs) store knowledge in their parameters and can generate informative responses when used in conversational systems. However, LMs suffer from the problem of "hallucination:" they may generate plausible-looking statements that are irrelevant or factually incorrect. To address this problem, we propose a contrastive learning scheme, named MixCL. A novel mixed contrastive objective is proposed to explicitly optimize the implicit knowledge elicitation process of LMs, and thus reduce their hallucination in conversations. We also examine negative sampling strategies of retrieved hard negatives and model-generated negatives. We conduct experiments on Wizard-of-Wikipedia, a public, open-domain knowledge-grounded dialogue benchmark, and assess the effectiveness of MixCL. MixCL effectively reduces the hallucination of LMs in conversations and achieves the highest performance among LM-based dialogue agents in terms of relevancy and factuality. We show that MixCL achieves comparable performance to state-of-the-art KB-based approaches while enjoying notable advantages in terms of efficiency and scalability.