17th International Conference on Digital Preservation, iPRES2020
Montag, 16. Dezember 2019
10:00 – 11:30 Uhr
FB Linguistik/AG Butt
Vered Shwartz (Institute for Artificial Intelligence - AI2 and Paul G. Allen School of Computer Science & Engineering, University of Washington.)
With the availability of massive training data for the task, natural language inference (NLI) has become the go-to task for testing natural language understanding. But a recent line of work showed that the current benchmarks over-estimate the actual performance on the task, and that models often perform well even when shown incomplete data, due to overfitting to benchmark artifacts.