The human brain supports acquisition mechanisms that extract structural regularities implicitly from experience without the induction of an explicit model. It has been argued that the capacity to generalize to new input is based on the acquisition of abstract representations, which reflect underlying structural regularities in the input ensemble. In this study, we explored the outcome of this acquisition mechanism, and to this end, we investigated the neural correlates of artificial syntactic classification using event-related functional magnetic resonance imaging. The participants engaged once a day during an 8-day period in a short-term memory acquisition task in which consonant-strings generated from an artificial grammar were presented in a sequential fashion without performance feedback. They performed reliably above chance on the grammaticality classification tasks on days 1 and 8 which correlated with a corticostriatal processing network, including frontal, cingulate, inferior parietal, and middle occipital/occipitotemporal regions as well as the caudate nucleus. Part of the left inferior frontal region (BA 45) was specifically related to syntactic violations and showed no sensitivity to local substring familiarity. In addition, the head of the caudate nucleus correlated positively with syntactic correctness on day 8 but not day 1, suggesting that this region contributes to an increase in cognitive processing fluency.