Generating Derivational Morphology with BERT.
Valentin Hofmann, Janet B. Pierrehumbert, Hinrich Schütze · arXiv (Cornell University) · 2020
Can BERT generate derivationally complex words? We present the first study investigating this question. We find that BERT with a derivational classification layer outperforms an LSTM-based model. Furthermore, our experiments show that the input segmentation crucially impacts BERT's derivational knowledge, both during training and inference.