Languages exhibit variation which may be conditioned or partially conditioned on sociolinguistic cues such as social context, gender, and ethnicity. While the existence of this kind of socially conditioned variation is well-established, there is debate as to the role of such cues in child language learning (e.g., Labov, 2013; Smith, Durham, & Richards, 2013). Namely, several studies of naturalistic language use provide examples where children’s production of sociolinguistic variants appears to be adult-like. It is, however, hard to establish whether the data demonstrates children’s knowledge of conditioning cues or systematic differences in the input children from different social groups receive (Smith et al., 2013). In this talk, I present data from four semi-artificial language learning experiments with 6-year-olds (and adults) that address this issue by investigating the ability to track artificial (novel) associations between different speakers and their usage of different forms. Results demonstrate that children (and adults) are sensitive to both deterministic and probabilistic statistical cues involving speaker identity, an ability which is fundamental to tracking sociolinguistic variation. Implications of these findings for the literatures on developmental sociolinguistics and regularization of unpredictable variation are discussed.