Enhancing Linguistic Competence of Language Models through Pre-training with Language Learning Tasks
arXiv:2601.03448v2 Announce Type: replace
Abstract: Language models (LMs) are pre-trained on raw text datasets to generate text sequences token-by-token. While this approach facilitates the learning of world knowledge and reasoning, it does not explic…