This paper examines implications for morpho-phonology of a model that minimizes the role of an innate linguistic endowment in grammar formation. ‘Bottom-up’ learning results in mental representations that form sets from perceived morphs but do not involve abstract ‘underlying’ representations. For production, syntactic/semantic features (S-features) identify morphs to be compiled into words. When multiple morphs bear the same S-feature, the grammar must select among the possible contenders. Selection involves phonological regularities or sub-regularities and morpho-phonological as well as idiosyncratic choice; when all else fails the default morph is selected. The model unifies the formal characterization of suppletion, sub-regularities, allophonic patterns, as well as unifying suppletion and zero morphs. Examples come from English, Southern Min, Yoruba, and Kinande, and other languages.