CRC 1475 at the workshop “Large Language METAPHORS – New Trends in Computational Approaches to Metaphorical Expressions”

The event brought together an interdisciplinary community of researchers working at the intersection of linguistics, computational linguistics, and psycholinguistics to discuss the role of Large Language Models (LLMs) in the analysis of figurative language.

In his talk “A Novel Dataset of Toxic Posts from Christian Subreddits” Sebastian examined the current capabilities and limitations of LLMs in metaphor detection. The work highlighted challenges in reliably identifying metaphorical expressions, particularly in contexts were metaphors were used with harmful intent. The study pointed to systematic difficulties in distinguishing between literal and figurative language and emphasized the need for more linguistically informed evaluation frameworks.

Beyond this contribution, the workshop featured a range of presentations on automatic metaphor identification, cross-linguistic approaches, and the development of new resources for metaphor research. A recurring theme was the reciprocal relationship between metaphor studies and LLMs: while LLMs provide powerful tools for large-scale analysis, metaphor remains a critical test case for evaluating their deeper language understanding.

Overall, the workshop highlighted both the potential and the current limitations of AI-driven approaches to figurative language, underlining the importance of interdisciplinary collaboration.

Here is the recording of Sebastian's talk: https://www.youtube.com/watch?v=U3VYMdrvu54&t=3704s

The full recording of the workshop can be found here: https://youtu.be/U3VYMdrvu54