You can read an interesting argument for that in On Intelligence:
I can't find the reference anymore, but there is a body of research which takes human motion and extracts a language from it including grammar, dialog acts, words, phonemes.
The possibilities of the video/body movement equivalent of speech recognition would open up a massive, incredibly interesting new dataset to wrangle.
I've found this part most interesting. I wonder if it's just the general pattern matching at work or something more.