Blog CRC1646

The 19^th Conference of the European Chapter of the Association for Computational Linguistics (EACL 2026), held in Rabat, Morocco from March 24–29, 2026, is one of the flagship venues in the fields of Computational Linguistics and Natural Language Processing (NLP), bringing together leading researchers from both academia and industry worldwide. As a major international conference under the ACL umbrella, EACL plays a central role in shaping research directions and fostering collaboration across a broad spectrum of topics, including large language models (LLMs), multilinguality, interpretability, and ethical considerations in NLP. The conference program was structured to provide a comprehensive and dynamic experience, combining plenary sessions and keynote talks with parallel oral presentations, poster sessions, system demonstrations, and industry tracks, alongside affinity meetings and Birds-of-a-Feather discussions that encourage community interaction. In addition to the main conference, EACL 2026 featured co-located workshops and tutorials covering emerging and niche topics, making it a rich platform for both knowledge exchange and professional networking.

The CRC 1646 was represented with two contributions in the conference: Omar Momen, PhD student in project A05, had the honour of presenting the project's research in a paper in the main conference and Maryam Mohammadi, postdoctoral researcher in project INF, presented her research in a poster on the behavior of LLMs in complex pragmatic structures in Farsi in the first workshop on NLP and LLMs for the Iranian language family Silkroad.

In the paper titled “Surprisal and Metaphor Novelty Judgments: Moderate Correlations and Divergent Scaling Effects Revealed by Corpus-Based and Synthetic Datasets”, the project A05 investigated whether surprisal, a probabilistic measure of predictability in LLMs, correlates with annotations of metaphor novelty in corpus-based and synthetic metaphor datasets. The results showed that LLM surprisal yields significant positive correlations with scores/labels of metaphor novelty. They further identified a divergent scaling pattern: on corpus-based data, correlation strength decreases with model size, whereas on synthetic data it increases. Omar’s oral talk on the paper was followed by a brief Q&A session. During the discussion, one attendee raised a broader concern regarding the evaluation of linguistic creativity, emphasizing that such assessments should account for cultural differences in how texts are perceived and interpreted. Another attendee noted that while surprisal shows a measurable correlation with metaphor novelty, it should not be considered a comprehensive or standalone metric for capturing the full complexity of metaphorical meaning.

The SilkRoad workshop, in which Maryam presented her poster entitled “Do Large Language Models Understand Double Mismatches? Evidence from Farsi”, highlighted the critical need to include low-resource languages in the rapidly evolving LLM landscape. Maryam’s work on the behavior of LLMs in complex pragmatic structures in Farsi contributed to this goal by examining how statistically driven models handle less frequent linguistic forms in an underrepresented language.

Participating in this conference offered a valuable opportunity to present the research by the CRC at the intersection of multilingual NLP and pragmatics, engage with researchers working on Farsi and related languages, and receive insightful feedback. This experience strengthens the projects A05 and INF through new perspectives and potential collaborations. But last not least, the EACL 2026 also contributes to the broader community by advancing understanding of LLMs in underrepresented languages, while enhancing the visibility, networking, and interdisciplinary knowledge exchange.

Maryam Mohammadi (INF) & Omar Momen (A05) © EACL

Presentation by Omar Momen (A05) © Omar Momen

Maryam Mohammadi (INF) at her poster © Omar Momen

Project A05

From left to right: Sina Zarrieß, Omar Momen, Emilie Sitter, Berenike Herrmann

Photo: Sascha Hermannski © SFB 1646

Project INF

Back row from left to right: Paul T. Schrader, Hendrik Buschmeier, Annett Jorschick & Marc Joppek

Front row from left to right: Katja Politt & Maryam Mohammadi

Photo: Sascha Hermannski © SFB 1646