Overview of MIRROR at IberLEF 2026: Motivational Interviewing Response Rating via synthetic cOnversational tuRns

Resumen

This paper presents the overview of the MIRROR shared task, organized in the framework of IberLEF 2026. The aim of the task is to promote research on Generative Artificial Intelligence methods for producing synthetic conversation turns that can improve the performance of models trained to recognize Motivational Interviewing Behavior Codes. We formulated the task as data-centric, asking participants to generate datasets for fine-tuning pre-trained models for distinguishing \textit{Behavior Codes} as follows: (1) Simple Reflection vs. Complex Reflection; (2) Open Question vs. Closed Question; and (3) Persuasion vs. Giving Information. A total of six teams submitted their generated datasets for the final evaluation. Two teams outperformed the baseline, and novel sample generation strategies were proposed in the context of this competition. An extensive analysis of the results is presented, proving the potential impact of Generative Artificial Intelligence strategies for robust synthetic data augmentation in the context of Motivational Interviewing analysis.