How do learners acquire and generalize knowledge about structural alternations based on limited input? One possibility is that learners are verb-wise conservative, i.e., they largely restrict their use of structural alternations to verbs that they have experienced them in. The opposite is that they generalize all alternations across all verbs. Yet another possibility is that they are sensitive to semantic properties of verbs (i.e., narrow conflation class; “give” and “hand” belong to the same class as they imply transfer, while “show” does not) and generalize to within-class verbs only. English monolinguals described “give”, “hand”, and “show” dative events in a Korean-English hybrid language. The language was constructed with Korean word order, case markers, and verbs, but English nouns. In the exposure phase, all subjects saw structural alternations only in “give” sentences. The non-scrambling group was exposed to the canonical order (NOM-DAT-ACC) and an alternation inside verb phrases (NOM-ACC-DAT). The scrambling group was exposed to scrambling (DAT-NOM-ACC) in addition to the orders shown to the non-scrambling group. Both groups saw only the canonical order for “hand” and “show”. In a subsequent picture description task with novel pictures, scrambling group learners produced the alternated structure (NOM-ACC-DAT) equally often with all verbs, but non-scrambling group learners produced the alternated structure more with “give” than with other verbs (interaction *p* < .05). The analogous pattern was found in a subsequent acceptability judgement task (interaction *p* < .05). Thus, learners were verb-wise conservative only when they saw no scrambling (i.e., when they saw no evidence that the structural alternations occur outside verb phrases). We found no evidence suggesting an effect of narrow conflation class. In conclusion, learners do not merely track statistical patterns in the input, but use internal linguistically sophisticated biases to generalize structural alternations.