Authors: H Ullrich, T Mlynář, J Drchal
Venue: Proceedings of the Seventh Fact Extraction and VERification Workshop (FEVER)
Year: 2024
Citations: 10
Links
Abstract
We recast automated fact-checking as a retrieval-augmented generation task: retrieve relevant evidence, then generate a verdict and justification directly from that evidence. The paper details a lightweight but effective pipeline and discusses where simplification helps versus where specialized modules are still needed.
Resources
- Video: Underline lecture recording
- Slides: Underline presentation slides
- Code: TODO
- Dataset: TODO
Notes
This was an important baseline paper for our team. It proved that a strong, well-tuned RAG approach can be surprisingly competitive, and it directly informed the system ideas we later pushed further in subsequent shared tasks.