Sociolinguistic studies generally focus on specific sociolinguistic variables. Consequently they rarely examine whether different sociolinguistic variables have coherent orientation in a specific language variety (a social or a regional dialect) or whether the speakers freely mix sociolinguistic variants. While different attempts have been made to identify coherence and mixing in the production or perception of dialects, our aim is to answer this question at the level of the cognitive representation of varieties. For this purpose, we draw on the phenomenon of sociolinguistic restoration: when they repeat sociolinguistically mixed utterances, people tend to make them homogeneous. The first experiment—a repetition task—reproduced sociolinguistic restoration in an experimental setting. The second experiment—a judgment task—ensured that participants perceived the difference between homogeneous and mixed utterances. We conclude that high-order coherent representation influences the reconstruction of utterances during the repetition task.