Jul 2026
Evaluating Communicative Belief Updates in Large Language Models via Implicature Recognition and Cancellation
It is found that LLM belief update understanding lags behind that of humans, especially in more naturally-occurring scenarios, and the first expert-annotated implicature cancellation dataset, ImplicatureX, is created, crowdsourced for human judgements of implicatures and their corresponding cancellations.
C. Piano, Verna Dankers, Marius Mosbach et al.
· arXiv.org · 1 citation