This article examines lexical conservatism (Steriade 1997), a phenomenon whereby the distribution of stem allomorphs in a morphological paradigm influences the way that paradigm accommodates derived members. Specifically, a phonological alternation applies in a derived member only if an existing form is present elsewhere in the paradigm that offers the needed phonological material. Thus illústrable undergoes stress shift because the existing word illústrative contains the illústr- stem allomorph. In contrast, *irrígable is judged worse than írrigable, since there is no existing form in *irríg-. In four experiments with speakers of English and Mexican Spanish, I demonstrate that this dependency between paradigm structure and application of phonological processes generalizes to entirely novel words in a probabilistic manner. Further, I find that a broad range of stem allomorphs in a paradigm play a role in determining the form of the novel word, rather than only those that could reduce the markedness of the novel form, contra previous studies. I propose a novel grammatical model where bases get to ‘vote’ on the shape of the novel form: all stem allomorphs in a lexical entry stand in a correspondence relation to the novel form and exert their influence via multiple faithfulness constraints, which compete with standard markedness constraints in a probabilistic phonological grammar.