Recent work in cognitive linguistics has hired the term coercion which was formerly associated with computational linguistics to apply to a number of instances in the study of natural language in which there is an incongruity between the semantics of a syntactic frame and the semantics of lexical items found in it. Some of these instances illustrate areas which could well be described as extensions of aspectual boundaries in which the harmony between lexical and grammatical aspect has been penetrated; others illustrate mismatch between count and mass noun types, in which the usual parameters that apply to either type have been violated. Although typical cases of coercion have been associated with the processes of metonymy, the purpose of the present paper is to analyse each individual case as it has been presented in recent accounts, and to investigate whether the processes involved in metonymic inferencing can be summarised in pragmatic terms, or otherwise in the light of their diachronic development and grammaticalisation. If so, it will thus be questioned how coercion can be justified in the context of natural language at all. [Copyright 2006 Elsevier B.V.] (LLBA, (c) CSA [2007]. All rights reserved.)