Penn Tree Bank

The Penn Tree Bank dataset is a corpus split into a training, validation and testing set of 929k words, a validation set of 73k words, and a test set of 82k words. The vocabulary has 10k words. The dataset is used for word-level language modeling.