Published in the Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 6: Industry Track), ACL 2026, San Diego, California, USA.
The paper proposes a self-supervised round-trip pipeline that turns knowledge-graph triples into text, rebuilds the graph from that text, and keeps only verbalizations that allow full recovery of the original graph. The round-trip score correlates strongly with expert judgment, and fine-tuning LLMs on the retained data improves both triple extraction and faithful text generation without manual annotation.
Authors: Kamyar Zeinalipour, Silvia Severini, Alessia Borghini, Sara Cardarelli, Marco Maggini, Marco Gori.
Read the paper | Code | Models and data
All the lab's language releases are listed in Language models and datasets.