Testing the performance of spoken dialogue systems by means of an artificially simulated user