
Leveraging a smaller but cleaner training dataset, the authors propose LIM-RA, an improved factual consistency evaluation model that outperforms the current state-of-the-art AlignScore across multiple benchmarks.


coremsg

improving-automatic-evaluation-of-factual-consistency-in-generated-text-by-leveraging-smaller-but-cleaner-training-data


Improving Automatic Evaluation of Factual Consistency in Generated Text by Leveraging Smaller but Cleaner Training Data
