|Abstract:||In this paper, we focus on the learnability of marked interactions of phonological processes. In particular, we explore the learnability of opaque interactions relative to transparent ones using an artificial-grammar learning paradigm. Participants behaved in one of three ways, either correctly learning the opaque grammar, not learning anything, or leveling the grammar into a transparent grammar. This grouping of participants accounts for the typological pattern found with respect to opaque interactions: They persist in languages despite being marked without any apparent benefit in perceptibility (first group), but they are less common than transparent relationships (last group). Furthermore, to explore the role memory has in supporting the acquisition of generalizations and opacity, these same participants took a battery of standardized cognitive and memory tests to assess whether performance on these general cognitive tests predicted language-learning success.