The Distributional Hypothesis Does Not Fully Explain the Benefits of Masked Language Model Pretraining.

Ting-Rui Chiang, Dani Yogatama

Browse the full EMNLP paper archive.