Language Models can Evaluate Themselves via Probability Discrepancy.

Tingyu Xia, Bowen Yu, Yuan Wu, Yi Chang, Chang Zhou

Browse the full ACL paper archive.