Few-shot Dialogue Strategy Learning for Motivational Interviewing via Inductive Reasoning

TL;DR

This work tackles the challenge of building MI-focused dialogue agents by introducing DIIR, a framework that learns natural-language dialogue strategies from expert demonstrations through inductive reasoning. DIIR uses a generator–discriminator–executor loop to create and verify strategy descriptions and then reuses these strategies at inference time to guide LLMs toward collaborative, non-authoritative MI behavior. Evaluations on the AnnoMI dataset with GPT-3.5 and GPT-4 show DIIR outperforms various in-context learning baselines and open-source models in MI-specific alignment metrics, with both automatic and human assessments supporting increased active listening and reduced unsolicited advice. The approach enables few-shot alignment of LLMs to MI techniques, offering a practical path toward more effective and empathetic MI dialogue systems without requiring extensive retraining.

Abstract

We consider the task of building a dialogue system that can motivate users to adopt positive lifestyle changes: Motivational Interviewing. Addressing such a task requires a system that can infer \textit{how} to motivate a user effectively. We propose DIIT, a framework that is capable of learning and applying conversation strategies in the form of natural language inductive rules from expert demonstrations. Automatic and human evaluation on instruction-following large language models show natural language strategy descriptions discovered by DIIR can improve active listening skills, reduce unsolicited advice, and promote more collaborative and less authoritative responses, outperforming various demonstration utilization methods.

Figures (1)

• Figure 1: DIIR retrieves learned dialogue strategy descriptions to generate a response that leads to positive outcomes. See \ref{['tab:learned_strategy_descriptions']} in the Appendix for example strategies.