Aslin, R.
University of Rochester, USA
The acquisition of grammatical categories involves solving two problems: (a) determining how many categories there are and (b) assigning auditory word-forms to these categories. Although in principle there is sufficient distributional information in the input to solve both of these problems, even degenerate cases of learning artificial grammatical categories in the lab have required additional sources of information, such as prosodic cues or absolute position in a string, to be successful. This has spawned a view that distributional (or statistical) learning and more abstract (or rule) learning are qualitatively different mechanisms. I will present the results of a series of experiments from adults, and on-going experiments from children, that provide evidence (a) that distributional information alone is sufficient for the formation of grammatical categories, and (b) that statistical learning and rule learning comprise a single mechanism whose breadth of generalization is determined by the specific patterns present in the input.