Decision-making and opinion-forming are everyday tasks that involve weighing pro and con arguments for or against different options. Our goal is to foster the development of technologies that support people in decision-making and opinion-forming and to improve our understanding of these processes. We invite you to participate in the 6th Touché lab on argumentation at CLEF 2025 featuring four tasks.
University of the Basque Country (UPV/EHU), HiTZ center
Automatic Argumentation faces new challenges in the LLMs era, including truthfulness, integration of external knowledge for counterargumentation and the fostering of critical reasoning. Central to our approach is the automatic generation of Critical Questions that systematically evaluate LLMs' argumentative reasoning by identifying logical weaknesses and challenging argument validity. This critical questioning framework serves as both an evaluation mechanism and a tool for enhancing automatic argumentation systems. Our findings reveal that truthful counterargumentation requires factual accuracy, logical reasoning, and critical evaluation capabilities, which may be facilitated by automated Critical Questions Generation systems.
Significant challenges remain in evaluation methodology and cultural sensitivity, highlighting the essential role of critical argumentation frameworks in developing more reliable and logically rigorous language models across diverse linguistic contexts.
Program
Touché is part of the CLEF 2025 conference program. All session times below are given in Madrid local time (CEST). Touché is also featured in the CLEF Lab Overviews session September 9, 11:30-13:15, in Salón de Actos - Facultad de Educación.
Tuesday, September 9, in Florentino Sanz - Facultad de Educación
14:15-15:45
Touché Session 1
14:15-14:30
Welcome
14:30-15:45 Keynote
Truthfulness and Critical Reasoning in Automatic Argumentation with LLMs [slides] Rodrigo Agerri and Blanca Calvo Figueras
15:45-16:30
Coffee Break and Poster Session
15:45-16:30
SINAI at Touché: From Generation to Evaluation through Multistep and Comparative Prompting for Retrieval-Augmented Debate [poster] María Estrella Vallecillo-Rodríguez, María Teresa Martín-Valdivia and Arturo Montejo-Ráez
15:45-16:30
Git Gud at Touché: Unified RAG Pipeline for Native Ad Generation and Detection [poster] Sameer Kamani, Muhammad Taqi, Ansab Chaudhry, Ahmed Hanif, Abdul Samad and Faisal Alvi
16:30-18:00
Touché Session 2
16:30-16:40
Overview of the Image Retrieval/Generation for Arguments Task [paper] [slides]
16:40-16:55
Infotec+CentroGEO at Touché: MCIP, CLIP and SBERT as Retrieval Score [slides] Tania Ramirez-Delreal, Daniela Moctezuma, Guillermo Ruiz, Mario Graff and Eric Tellez
16:55-17:05
Overview of the Advertisement in Retrieval-Augmented Generation Task [paper] [slides]
17:05-17:20
Git Gud at Touché: Unified RAG Pipeline for Native Ad Generation and Detection [slides] Sameer Kamani, Muhammad Taqi, Ansab Chaudhry, Ahmed Hanif, Abdul Samad and Faisal Alvi
17:20-17:35
TeamCMU at Touché: Adversarial Co-Evolution for Advertisement Integration and Detection in Conversational Search [slides] To Eun Kim, João Coelho, Gbemileke Onilude and Jai Singh
17:35-17:50
JU-NLP at Touché: Covert Advertisement in Conversational AI-Generation and Detection Strategies [slides] [video] Arka Dutta, Agrik Majumdar, Sombrata Biswas, Dipankar Das and Sivaji Bandhopadhay
17:50-18:00
Open Discussion
Wednesday, September 10, in Florentino Sanz - Facultad de Educación
14:15-15:45
Touché Session 3
14:15-14:25
Overview of the Retrieval-Augmented Debating Task [paper] [slides]
14:25-14:40
DS@GT at Touché: Large Language Models for Retrieval-Augmented Debate [slides] Anthony Miyaguchi, Conor Johnston and Aaryan Potdar
14:40-14:55
SINAI at Touché: From Generation to Evaluation through Multistep and Comparative Prompting for Retrieval-Augmented Debate [slides] María Estrella Vallecillo-Rodríguez, María Teresa Martín-Valdivia and Arturo Montejo-Ráez
14:55-15:05
Overview of the Ideology and Power Identification in Parliamentary Debates Task [paper] [slides]
15:05-15:20
GIL_UNAM_Iztacala at Touché: Benchmarking Classical Models for Multilingual Political Stance and Power Classification [slides] Jesús Vázquez-Osorio, Luis A. H. Miranda, Adrián Juárez-Pérez, Gerardo Sierra and Gemma Bel-Enguix
15:20-15:35
Munibuc at Touché: Generalist Embeddings for Orientation and Populism Detection [slides] [video] Marius Marogel and Silviu Gheorghe
