VERVE: Template-based ReflectiVE Rewriting for MotiVational IntErviewing

TL;DR

VERVE tackles the challenge of converting non-reflective counselor statements into reflective Motivational Interviewing responses without relying on parallel data. It uses a template-based editing framework with a reflection-discriminator-guided masking step, encoder-decoder filling, paraphrase-augmented training, and an adaptive template updating mechanism to control content preservation while enhancing reflection. Across PAIR and AnnoMI datasets, and via automatic and human evaluations, VERVE achieves stronger reflection improvements while maintaining content and exhibits favorable alignment with MI expert rewrites, outperforming two template-based baselines. The work demonstrates a scalable approach for counselor training, with practical implications for coaching and education, and suggests avenues for integrating larger language models in future work.

Abstract

Reflective listening is a fundamental skill that counselors must acquire to achieve proficiency in motivational interviewing (MI). It involves responding in a manner that acknowledges and explores the meaning of what the client has expressed in the conversation. In this work, we introduce the task of counseling response rewriting, which transforms non-reflective statements into reflective responses. We introduce VERVE, a template-based rewriting system with paraphrase-augmented training and adaptive template updating. VERVE first creates a template by identifying and filtering out tokens that are not relevant to reflections and constructs a reflective response using the template. Paraphrase-augmented training allows the model to learn less-strict fillings of masked spans, and adaptive template updating helps discover effective templates for rewriting without significantly removing the original content. Using both automatic and human evaluations, we compare our method against text rewriting baselines and show that our framework is effective in turning non-reflective statements into more reflective responses while achieving a good content preservation-reflection style trade-off.

Figures (3)

• Figure 1: In this example of counselor response rewriting, a counseling trainee is asked to provide a reflective response given the client prompt and produces a poor response by giving a piece of advice rather than reflecting the client's concerns. Our system generates an improved response that preserves content and increases the use of reflective language.
• Figure 2: Overview of the VERVE framework. During training, we use attribute-masked versions of paraphrases of reflections as templates for the MLE training for generator training. In the inference time, we adjust the content weight iteratively to achieve the desired edit effect.
• Figure 3: Analysis of edit effect by original counselor behavior. The error bars are 95% confidence intervals.