A pharmaceutical company has millions of unlabeled internal research papers. It wants a foundation model to understand its chemistry terms and research areas before any task-specific training. Which training method fits best?