What constrains children's learning of novel shape terms?

In this study, 3-year-olds matched on vocabulary score were taught three new shape terms by one of three types of linguistic contrast: corrective, semantic, or referential. A 5-week training paradigm implemented four training sessions and four assessment sessions. Corrective contrast ("This is concave, it is not square," where square is the child's label for the target) produced more learning than did either semantic or referential contrast. In addition, regardless of group, more was learned about those targets that were classified more variably at pretest. Avoidance of lexical overlap (i.e., using more than one term for the same dimension) might make it more difficult for children to learn new dimensional adjectives, and a "shape bias" might make learning shape terms easier. However, children's expectations about the speaker's communicative intent interacted with the potential benefits of contrast in the semantic condition, and children in that group learned no more than did controls.