ABSTRACT
Marcus, Vijayan, Bandi Rao, Vishton’s experiment (1999) concerning infant ability to discriminate between simple syntactic structures has prompted many connectionists to strive to demonstrate that certain types of neural networks can replicate those results. In this paper we take a closer look at two such attempts: Shultz & Bale (2001) and Altmann & Dienes (1999). We were not only interested in how well these two models matched the infants’ reported results, but also whether or not they were able to learn the grammars involved in this process. After performing an extensive set of experiments, we found that, at first blush, Shultz & Bale’s model replicated the infant’s known data, but the model largely failed to learn the grammars. We also discovered serious problems with Altmann & Dienes’ model, which failed to match most of the infant’s results and to learn the syntactic structure of the input patterns.
