We present some results from an experiment in multimodal human-computer interaction, done under the "Wizard of Oz" paradigm. This experience is a first glimpse at the use of future multimodal interfaces by average users. The means of communication available to the subjects were natural language and 2D gesture through a touch panel. Two operators simulated this intelligent multimodal interface. The subjects tended to make use of the flexibility of the machine in order to choose a comfortable type of expression. Further results should influence the design of vocal recognition systems based on spontaneous speech combined with gesture.

Keywords: spoken dialogue, human factors, natural language processing, multimodal interfaces.