Whamit!

Most sentences are unique … except the ones that aren’t. The frequency distribution in any corpus of natural language has a famously `long tail’ filled with unique phrases only ever used once.  Conversely, the other end of the distribution is filled with endless repetitions of familiar and formulaic phrases used to manage conversation (yeah, mhm, oh, y’know, right, okay, well), express reactions (aw man, wow, cool, whoa, yuck, yikes, phew), and serve social scripts (good morning, thanks, fuck off, I’m sorry).    

We observe three quirky properties that seem to be peculiar to the frequent and formulaic phrases at the fat end of the distribution. 
    1) phones — phones/phonotactics/phonation outside the productive phonology 
    2) tones — phrase-specific intonational contours and floating intonational contours
    3) emblems — manual gestures with highly conventionalized meaning that occur with (and substitute for) verbal equivalents 

These properties tend to occur together and seem to be restricted to the elements at the fat end of the distribution. We propose that this be explained by positing a co-grammar for English that operates over frequent and formulaic phrases. In addition to accounting for the presence and distribution of properties (1-3) a co-grammar makes it possible to account for an over-abundance of surface regularities in the fat end of the distribution (e.g., prosodic doubling: hear hear, there there, come come, now now, aye aye, nix nix, etc.) and the presence of proto-morphology in patterns which seem compositional but not productive ({welp, nope, yup, yep}, {wowza, yowza, wowzers, yowzers, yeppers, yuppers, right-o, neat-o}, {whatevs, natch, obvs, obvi}).