This paper deals with the potential grammaticalization of English (it) could be and (it) might be into epistemic sentence adverbs in analogy to maybe. They can occur in adverb-like position and function in informal language use, e.g. (it) could be something good has begun, often with the pronoun it omitted. But, given that no diachronic development is attested, to what extent does their usage indicate innovation or an emerging convention of adverbial could be / might be? How are they differentiated from maybe?

I present a corpus study and a set of small experiments investigating these questions. The results suggest that grammaticalization mechanisms are at work, such as the connection between formal reduction and functional/structural aspects, even in the absence of diachronic change and irrespective of item frequency. On the other hand, there is little evidence of conventionalization of adverbial could be / might be, and also no clear signs of semantic or pragmatic differentiation. I conclude that weak conventions leave room for variability, and propose that these forms have a proclivity to be continuously re-innovated as micro-steps on a grammaticalization path, but this is not enough to drive change beyond existing conventions.