Do LLMs fail in bridging generation?

Natalia Skachkova; Simon Ostermann; Josef van Genabith; Bernd Kiefer

doi:10.21248/jlcl.38.2025.286

Do LLMs fail in bridging generation?

Authors

Natalia Skachkova Deutsches Forschungszentrum für Künstliche Intelligenz (DFKI) https://orcid.org/0000-0002-1752-3452
Simon Ostermann Deutsches Forschungszentrum für Künstliche Intelligenz (DFKI) https://orcid.org/0000-0002-0899-0657
Josef van Genabith Deutsches Forschungszentrum für Künstliche Intelligenz (DFKI) https://orcid.org/0000-0003-1322-7944
Bernd Kiefer Deutsches Forschungszentrum für Künstliche Intelligenz (DFKI) https://orcid.org/0000-0003-2323-091X

DOI:

https://doi.org/10.21248/jlcl.38.2025.286

Keywords:

bridging, LLMs, text generation, artificial data generation, semantic similarity

Abstract

In this work we investigate whether large language models (LLMs) ‘understand’ bridging relations and can use this knowledge effectively. We present the results obtained from two tasks: generation of texts containing bridging and filling in missing bridging spans. We show that in most of the cases LLMs fail to generate bridging in a reliable way.

Downloads

Published

2025-07-08

How to Cite

Skachkova, N., Ostermann, S., van Genabith, J., & Kiefer, B. (2025). Do LLMs fail in bridging generation?. Journal for Language Technology and Computational Linguistics, 38(2), 77–95. https://doi.org/10.21248/jlcl.38.2025.286

Download Citation

Issue

Vol. 38 No. 2 (2025): LLM fails – Failed experiments with generative AI and what we can learn from them

Section

Research articles

License

This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.