Intervocalic consonant epenthesis is used as a case study for investigating grammar change. An emergentist framework is adopted, whereby a simple learning mechanism transforms a phonetically-based sound change into a synchronic phonological process. A two-part model of such 'grammaticalizing' change is developed, along with a formal analysis of the necessary model properties. This work demonstrates that perception-based consonant loss could, in principle, lead to synchronic epenthesis. However, the larger number of historic conditions required for its emergence are predicted to make it less likely than other outcomes such as deletion or suppletion (unpredictable alternation). An important corollary of the latter result is that theoretically dispreferred grammars do not necessarily have to be explicitly marked or removed from the learning space. Input to the hypothetical learner is automatically filtered by asymmetries in the way sound changes occur, the way sounds are organized in words, and the way words are organized in paradigms. The conclusion is that mechanisms other than Universal Grammar are sufficient to produce the observed epenthesis typology without overgenerating. Furthermore, it is argued that the research methodology is a promising one in general for explaining the universal tendencies of human languages.