Papers and Workshops at LREC 2026

02/20/2026

Six papers and two workshops at the Language Resources and Evaluation Conference 2026 in May in Palma de Mallorca with BamNLP involvement.

The BamNLP Group at the University of Bamberg has the following papers and contributions at The 15th Language Resources and Evaluation Conference in Palma de Mallorca in May 2026 (https://lrec2026.info/).

Lynn Greschner, Sabine Weber, Roman Klinger:
Trust Me, I Can Convince You: The Contextualized Argument Appraisal Framework and the ContArgA Corpus.
_{This paper introduces the Contextualized Argument Appraisal Framework and the ContArgA corpus to show that people’s emotional appraisals of arguments systematically shape perceived convincingness, with positive emotions and appraisals (e.g., pleasantness, familiarity, positive consequences) increasing convincingness and negative ones decreasing it.}

Yarik Menchaca Resendiz, Roman Klinger:
PARL: Prompt-based Agents for Reinforcement Learning.
_{We introduce PARL, a prompt-based reinforcement learning method that uses frozen large language models as in-context RL agents by encoding states, actions, and rewards in the prompt. It can match or outperform standard RL algorithms in knowledge-driven tasks (e.g., Blackjack).}

Lynn Greschner, Meike Bauer, Sabine Weber, Roman Klinger:
Categorical Emotions or Appraisals – Which Emotion Model Explains Argument Convincingness Better?
_{Zero-shot LLM experiments on the ContArgA corpus show that appraisal-based emotion representations explain and improve argument convincingness prediction more strongly than categorical emotion labels, although joint prediction underperforms pipeline approaches.}

Egil R?nningstad, Roman Klinger, Lilja ?vrelid, Erik Velldal:
Sentence Relevance Detection for Entity-Level Sentiment Analysis.
_{We show that extracting entity-relevant text spans by fine-tuning models with pairwise comparison yields better results for entity-level sentiment analysis than purely fine-tuning for the classification task.}

Johannes Sch?fer and Roman Klinger:
Disambiguation of Emotion Annotations by Contextualizing Events in Plausible Narratives.
_{We introduce the Emotional BackStories (EBS) dataset and a story-planning–based generation framework to show that automatically generated contextual backstories systematically disambiguate otherwise ambiguous events.}

Sabine Weber, Lynn Greschner, and Roman Klinger.
Less is More? The Role of Demographic Author Information in Emotion Classification of Ambiguous Text
_{Showing author information alongside emotionally ambiguous text does not improve inter-annotator agreement in emotion classification, but can even harm it. Zero-shot prompting experiments with LLMs do resemble the human annotation experimental results.}

In addition, BamNLP is involved with two workshop organizations:

Christopher Bagdon and Roman Klinger co-organize the Computational Affective Science Workshop together with Krishnapriya Vishnubhotla, Kristen A. Lindquist, Lyle Ungar, Saif M. Mohammad. (https://casworkshop.github.io/)
Roman Klinger and Sabine Weber co-organize the Workshop on Social Context (SoCon) and Integrating NLP and Psychology to Study Social Interactions (NLPSI) together with Aswathy Velutharambath, Sofie Labat, Neele Falk, Flor Miriam Plaza del Arco, Véronique Hoste, Marco Antonio Stranisci, Soda Marem Lo, Rossana Damiano, Simona Frenda, Viviana Patti, Maarten Sap, and Seid Muhie Yimam. (https://socon-nlpsi.github.io/)