Check out Sihao's internship project @GoogleAI!
Most #NLI datasets label full sentences. Instead, #PropSegmEnt focuses on fine-grained proposition-level textual entailment in the wild!
For example, we identify the specific hallucinated spans out of a generated summary! #NLProc
Paper: https://github.com/google-research-datasets/PropSegmEnt/blob/main/propsegment-preprint.pdf
Dataset: https://github.com/google-research-datasets/PropSegmEnt
RT: https://twitter.com/TalSchuster/status/1605430905107087363?t=SOyNskQ01Pyyzz0ytsUCSQ&s=19