Dataset

For this edition, the corpus consists of simulated yet realistic therapeutic conversations in Spanish between patients and professional therapists. The conversations were created using a combination of human-authored and synthetically generated dialogues, all of which were reviewed by experts to ensure clinical plausibility.

To generate the data, we asked five licensed psychologists to interact through a custom application with simulated patients, played by psychology students. Each student was assigned a patient profile, including specific characteristics and key symptoms derived from standardized questionnaires such as the PHQ-9, GADS-7, and CompACT. During the chats, the application provided AI-generated response suggestions for the therapist, which they could select, modify, or replace entirely with their own input. This process ensured that the conversations reflected both realistic patient behavior and professional therapeutic responses.

Each conversation consists of an ordered sequence of alternating turns between patient and therapist, designed to capture the progressive revelation of emotional states, concerns, and symptoms, as well as evidence-based therapeutic interventions. All dialogues were carefully reviewed by clinical experts to verify the quality, coherence, and clinical validity of both the patient and therapist turns. This step ensures that the simulated interactions remain consistent with real-world therapeutic principles.

All data have been anonymized and underwent ethical review to ensure privacy protection and compliance with responsible research standards.

The same set of conversations is used across both subtasks. To support participation, the organizers will release:

• A limited set of trial conversations for format familiarization.

• Test conversations, delivered incrementally during the evaluation phase. No labeled training data will be released.