Present-day English has two modal be bound to constructions, the most common one an “epistemic” one and a less frequent deontic one. It is a matter of debate whether the more frequent construction is epistemic or a case of “participant-external necessity”, but diachronic corpus data show that it appeared much later than the deontic construction. The semasiological evolution of the form be bound to consequently resembles the “path” Bybee et al. (1994) suggest for should. However, though such semasiological paths are valid explananda for historical linguistics, they should not be taken to be cognitively realistic narratives either of what happened to specific constructions or of how specific constructions came about. There may in fact not be a straight path from the deontic to the “epistemic” construction, particularly from a usage-based diachronic construction grammar perspective (cf. Barðdal & Gildea 2015).

A “radically” usage-based model of language change is one that separates individual linguistic knowledge from what is traditionally conceived of as the conventional synchronic language system, as a prerequisite to be able to account for constructional innovation in a cognitively plausible way. Comparing two explicitly usage-based models of constructional change, Fischer (2007) and Traugott and Trousdale (2013), the former is the more radical one in this sense in that it attributes innovations to analogy with constructions already present in the speaker-listener's grammar, whereas the latter inherently entangles internal and external systems through its insistence on pragmatics-induced reanalysis as the primary mechanism of change. Taking its cue from the Fischer model, this paper explores the likelihood of a number of possible analogical sources for epistemic be bound to by connecting diachronically arranged usage data for this construction with data for formally and functionally similar as well as superficially identical but semantically dissimilar constructions collected from a diachronic corpus and two text archives.