Visually-Augmented Language Modeling