Multiple layers of visual (and vocal) signals, plus their different onsets and offsets, represent a significant semantic and temporal binding problem during face-to-face conversation.
Despite this complex unification process, multimodal messages appear to be processed faster than unimodal messages.

Multimodal gestalt recognition and multilevel prediction are proposed to play a crucial role in facilitating multimodal language processing.

The basis of the processing mechanisms involved in multimodal language comprehension is hypothesized to be domain general, coopted for communication, and refined with domain-specific characteristics.
A new, situated framework for understanding human language processing is called for that takes into consideration the multilayered, multimodal nature of language and its production and comprehension in conversational interaction requiring fast processing.