Multimodal fusion combines the output of speech understanding, gestures and mimic recognition (if available) to an uniform representation of the user intention.