The present experiments investigate how young language learners begin to acquire form-based categories and the relationships between them. We investigated this question by exposing 12-month-olds to auditory structure of the form aX and bY (infants had to learn that a-elements grouped with Xs and not Ys). Infants were then tested on strings from their training language versus strings from the other language using a preferential-listening procedure. Importantly, the X and Y elements were new at test, requiring infants to generalize to novel pairings. We also manipulated the probability of encountering grammatical structures of the training language by mixing strings from two artificial languages according to 83/17 and 67/33 percentage ratios in Experiment 2. Experiment 1 shows that 12-month-olds are capable of forming categories of X- and Y-elements based on a shared feature and, furthermore, form associations between particular a- and b-elements and these categories. Experiment 2 shows that learning was sustained even when 17% of instances from another language were present during training. However, infants failed to generalize when exposed to a larger percentage of strings from another language. The findings demonstrate that the first step of form-based category abstraction (the ability to generalize based on marker-feature pairings) is in place by 12 months of age.