Second Call For Participation: MIRROR@IberLEF2026: Platform online

<Apologies for cross-postings>

--------------------------------------
SECOND CALL FOR PARTICIPATION
--------------------------------------

MIRROR@IberLEF20206: Motivational Interviewing Response & Rating via 
Synthetic cOnversational tuRns
https://mirror-iberlef.vercel.app/

Challenge platform now online: https://www.mirror-iberlef.lat/dashboard
Just register, log in, and start making submissions.


-------------------------------
***Task description***
-------------------------------

We invite the community to develop Generative AI (GenAI) methods for 
creating synthetic conversation turns that can substantially improve the 
performance of models trained to recognize behavior codes (BCs) in the 
context of motivational interviews. A BC is a discrete, observable 
clinician action (e.g., asking a question, giving information) that is 
counted during coding of a motivational interviewing session to quantify 
specific techniques used. These codes allow raters to tally how often 
particular clinician behaviours occur, which helps assess adherence to 
MI-consistent versus MI-inconsistent practice. Our ultimate goal is to 
generate valuable data for training models for the automatic assessment 
of clinicians’ motivational-interviewing skills. These skills — crucial 
for promoting behavior change among patients — can be evaluated by using 
the “Motivational Interviewing Treatment Integrity (MITI)” rubric 
(https://tinyurl.com/38byjrwy).

This is a data-centric competition: participants are expected to produce 
high-quality datasets representing a wide range of clinical 
conversations (rather than training a model) to enhance the performance 
of a frozen baseline model used for BC classification. We encourage 
participants to include samples featuring clients from diverse 
backgrounds, varied conversation topics, and conversing with different 
types of health professionals.

Participants in this competition should provide three datasets (one per 
pair of considered BCs) of at most 100 labeled conversation turns that 
will be used to fine-tune pretrained models; the fine-tuned models will 
then be used to make predictions for a hold-out dataset. The performance 
of the fine-tuned model will be used as the leading evaluation metric to 
rank participants.  The considered pairs of BCs are:

(1) Simple reflection vs. Complex reflection;
(2) Open question vs. Closed question;
(3) Persuasion vs. Giving Information.

Sample submissions, and detailed instructions on the formatting, 
evaluation criteria and competition platform will be available at the 
MIRROR website.

-------------------------------
***Important dates***
-------------------------------

* Mar 9th: Start of the development phase (platform starts receiving 
submissions for the validation set)
*   May 1st: Start of the final phase (platform starts receiving 
submissions for the test set)
*   May 11th: End of evaluation campaign (deadline for submission of runs)
*   May 22nd; Publication of official results
*   Jun 8th: Deadline for paper submission
*   Jun 23th: Acceptance notification
*   Jun 30th: Camera-ready submission deadline
*   Sep, TBD: Publication of proceedings
*   Sep, TBD: Workshop with SEPLN 2026

-------------------------------
***Organizing team***
-------------------------------

* Luis J. Arellano INAOE, Mexico
* Carlos Olachea  INAOE, Mexico
* John Piette, University of Michigan, USA
* Hugo Jair Escalante, INAOE, Mexico
* Delia Irazú Hernández, INAOE, Mexico
* Luis Villaseñor, INAOE, Mexico
* Manuel Montes, INAOE, Mexico

Contact: Hugo Jair Escalante ([email protected])
_______________________________________________
Corpora mailing list -- [email protected]
https://list.elra.info/mailman3/postorius/lists/corpora.list.elra.info/
To unsubscribe send an email to [email protected]