Learning Deterministic Finite Automaton with a Recurrent Neural Network

We consider the problem of learning a nite automaton with recurrent n e u r a l networks from positive evidence. We train Elman recurrent neural networks with a set of sentences in a language and extract a nite automaton by clustering the states of the trained network. We observe that the generalizations beyond the training set, in the language recognized by the extracted automaton, are due to the training regime: the network performs a \loose" minimization of the of the training set (the automaton that has a state for each preex of the sentences in the set).