Reflect, Not Reflex: Inference-Based Common Ground Improves Dialogue Response Quality

TL;DR

This paper introduces Reflect, a dataset with explicit common ground annotations, demonstrating that modeling shared knowledge improves dialogue response quality and that prompting GPT-3 to consider common ground enhances response quality.

Contribution

The paper presents Reflect, a novel dataset with annotated common ground, and shows that explicitly modeling shared knowledge improves dialogue response quality, especially when guided by GPT-3 prompts.

Findings

Less than half of current dialogue responses are high quality

Models trained on existing data produce lower quality responses

Prompting GPT-3 to consider common ground increases response quality by 30%

Abstract

Human communication relies on common ground (CG), the mutual knowledge and beliefs shared by participants, to produce coherent and interesting conversations. In this paper, we demonstrate that current response generation (RG) models produce generic and dull responses in dialogues because they act reflexively, failing to explicitly model CG, both due to the lack of CG in training data and the standard RG training procedure. We introduce Reflect, a dataset that annotates dialogues with explicit CG (materialized as inferences approximating shared knowledge and beliefs) and solicits 9k diverse human-generated responses each following one common ground. Using Reflect, we showcase the limitations of current dialogue data and RG models: less than half of the responses in current data are rated as high quality (sensible, specific, and interesting) and models trained using this data have even lower quality, while most Reflect responses are judged high quality. Next, we analyze whether CG can help models produce better-quality responses by using Reflect CG to guide RG models. Surprisingly, we find that simply prompting GPT3 to "think" about CG generates 30% more quality responses, showing promising benefits to integrating CG into the RG process.