Generalization by categorical nodes in recurrent neural networks

We experimentally show that, in the grammar learning by recurrent neural networks, the networks can learn a word categorization that is consistent with the grammatical categories of words in exemplar sentences.