Ackerman et al. (2009) and Ackerman and Malouf (2013) argue that the organization of morphological systems allows speakers to efficiently solve the Paradigm Cell Filling Problem: how to predict a word’s inflected form given some of its other forms. For example, a given word’s possessive suffix in Hungarian can be predicted in part by its phonology and by its exponent in forms like the plural. I propose a model in which this phonological and morphological predictability is encoded in formal, constraint-based grammars. In particular, I extend the sublexicon model (Gouskova et al., 2015), in which lexically specific behavior is handled using diacritics on lexical items. Thus, I treat the Paradigm Cell Filling Problem as a problem of finding correlations between diacritics in a given lexical entry, without relying on the storage of a paradigm’s output forms.