Leveraging Explicit Reasoning for Inference Integration in Commonsense-Augmented Dialogue Models

TL;DR

This paper investigates whether making commonsense reasoning explicit in dialogue models improves response quality. It introduces ConvoSense-E, a generate–select–respond pipeline that explicitly generates, selects, and integrates commonsense inferences, and contrasts it with ConvoSense-I (implicit reasoning) and baselines GPT and Doctor. Across human evaluations on Reflect dialogues, explicit reasoning yields clearer gains in engagingness, specificity, and overall quality, with naturalness remaining similar. The work demonstrates the value of modular, explicit reasoning over commonsense for open-domain dialogue and provides insights into which commonsense types most benefit response generation, marking a new state-of-the-art in commonsense-augmented dialogue modeling.

Abstract

Open-domain dialogue systems need to grasp social commonsense to understand and respond effectively to human users. Commonsense-augmented dialogue models have been proposed that aim to infer commonsense knowledge from dialogue contexts in order to improve response quality. However, existing approaches to commonsense-augmented dialogue rely on implicit reasoning to integrate commonsense inferences during response generation. In this study, we explore the impact of explicit reasoning against implicit reasoning over commonsense for dialogue response generation. Our findings demonstrate that separating commonsense reasoning into explicit steps for generating, selecting, and integrating commonsense into responses leads to better dialogue interactions, improving naturalness, engagement, specificity, and overall quality. Subsequent analyses of these findings unveil insights into the effectiveness of various types of commonsense in generating responses and the particular response traits enhanced through explicit reasoning for commonsense integration. Our work advances research in open-domain dialogue by achieving a new state-of-the-art in commonsense-augmented response generation.

Figures (5)

• Figure 1: Example dialogue demonstrating the reasoning flow for integrating commonsense.
• Figure 2: Overview of the explicit and implicit reasoning approaches (ConvoSense-E/I).
• Figure 3: Evaluation interface.
• Figure 4: Percentage that ConvoSense-E wins against other models on Quality, split by the type of commonsense selected for response integration by ConvoSense-E.
• Figure 5: Proportion of explanations that include each response characteristic for each model.