Forecasting Corpus-Sampling Disagreement in Masked Text Reconstruction

Hare Krishna

If a second text corpus were collected from the same population, how much would a predictor's outputs change? We study this question for masked-text reconstruction using a smoothed…