Proceedings of the Annual Meeting of the Cognitive Science Society

This paper probes the influence of a particular kind of domain-general cognitive bias in first language acquisition with the aid of computational models. We introduce a novel task: inducing functional categories from morphemically tokenised sentences, and supply a manually annotated dataset of English child-directed speech (CDS). We operationalise a widely assumed type of cognitive bias, "less-is-more", as three computational principles—ordering input, gradually increasing model complexity, and priming the learner—and develop a theory-neutral experimental setup to evaluate their impact on functional category induction. Our experiments with CDS demonstrate that models incorporating reflexes of "less-is-more" outperform the purely statistical baseline. As part of our exploration of ordering effects, we employ the morpheme acquisition order proposed by Brown (1973) and, for the first time in literature, present statistical evidence that Brown-compliant orders outperform non-Brown-compliant ones.