It is generally accepted in usage-based theories of language that language change is facilitated by contexts of use that allow for semantically and/or structurally ambiguous readings of a construction. While this facilitative effect is well attested in the literature, much less attention has been paid to factors that constrain language change. Furthermore, the idea that ambiguities may also promote stability and discourage change has not been explored in detail. In this article I discuss the development of three participle constructions that have gained increasingly adjective-like uses in recent history: ADJ- looking (e.g., modest-looking), N-V ing denoting a change in psychological state (e.g., awe-inspiring), and the adjectival - ed participle headed by a psych-verb (e.g., surprised). I argue that the development of these constructions has been significantly constrained by unresolved ambiguities as well as source structures that continue to support the earlier, verbal categorization instead of an adjectival one. As a consequence, the participles have acquired adjectival uses gradually and often at a slow rate. The data are analyzed from a constructionist perspective, where constructions are connected in a network, and categories like “verb” or “adjective” are regarded as emergent schemas that arise from actual patterns of use.