MEASURING THE UTILITY OF INFANT GRAMMATICAL CATEGORY KNOWLEDGE
Katie Khuu, Lisa Pearl.
University of California, Irvine, Irvine, CA.
Children learn much about their native language without explicit instruction. Yet, what exactly are they learning in the early stages of linguistic development? One goal might be to learn knowledge that is useful for understanding the language around them, even if this knowledge is not adult-level knowledge. As a case study, we investigated how children first learn grammatical categories, like nouns and verbs. This process begins around 12 months when children are capable of using 1) distributional cues such as which words appear together, and 2) communicative cues like what utterance types (e.g., questions, statements, or commands) in which words appear. We applied a promising learning strategy using both of these cues to an age-appropriate dataset. To evaluate whether the strategy was successful, we compared the inferred categories to adult categories, like nouns and verbs, and evaluated how useful the inferred categories were. Our main findings are 1) the quantitative comparison against adult categories shows that the learning strategy does not fare as well on an age-appropriate dataset as it did in the original study demonstrating its performance, but 2) the inferred categories still seem to be useful given qualitative and information-theoretic analyses. So, in short, what children would learn with this strategy is not crazy; the knowledge is very useful even if it is not yet the adult-level knowledge. These results can inform expectations about what typically developing children should be able to learn about language early on, and what appropriate evaluations are for assessing when they are (a)typical.