My colleagues and I just got a paper accepted at Psych Science on this debate: we looked at the degree of generalization kids show in using determiners/articles

My colleagues and I just got a paper accepted at Psych Science on this debate: we looked at the degree of generalization kids show in using determiners/articles ('a','an' or 'the') with nouns, also accounting for imitation of caregiver speech. If a child has heard "the cat" and "a cat", but only ever "the fish", when do they know they can generalize and say "a fish"? TLDR is that it's hard to say conclusively without larger samples of caregiver speech, but it seems like early speech (<2 yrs) has little evidence of category-based generalization, but it rapidly increases after that. Of course it's possible to contest that kids have more grammatical knowledge than what's reflected in what they say (the performance vs. competence distinction), but these methods can't address that. http://langcog.stanford.edu/papers_new/meylan-inpress-psychsci.pdf http://langcog.stanford.edu/papers_new/meylan-inpress-psychs... HN readers may appreciate the methodology: we construct a beta-binomial model as a simple approximation of the generative process, and measure how the parameters change over time when we fit to read data. The child's definite vs. indefinite uses for each noun are treated as flips of a weighted coin; the weighting is determined by the definite and indefinite uses by the parent combined with a beta prior. The peakedness of the beta prior indicates how similar a kid thinks nouns will be to one another (in the absence of evidence). We fit this model to all the successive samples of a bunch of longitudinal developmental datasets, including the Speechome project out of MIT. Nonzero, constant peakedness would be consistent with access to syntactic categories from birth; increasing peakedness of the scale parameter corresponds to growing syntactic generalization.