Preliminary experiments for automatic speech understanding through simple recurrent networks