3 thoughts on “Learning verb argument-structure: Syntax and statistics

  1. Thank you for such a wonderful talk, Cindy! I am wondering whether there is also a temporal hierarchy in processing/learning verb-specific vs. verb-general statistics both for an individual learner and across development. Is verb-specific statistical information more malleable or verb-general statistics more malleable?

    1. Thanks, Zhenghan!
      Both levels seem to be quite malleable. I stripped this down to fit in 35 minutes, but Thothathiri & Braiuca, 2020, have data patterns similar to our multi-level learning result with adults. That paper did not include the same within-experiment comparison, so it is difficult to compare quantitatively, but to a first approximation it looks as if both the verb general (does verb identity predict structure in my local environment) and verb specific (does verb X predict structure Y) levels can be readily pushed around, and that there is considerable continuity across development in the nature and the size of these effects. This is one of the kinds of effects that show, as Chang, Dell & Bock 2006 put it, that ‘language learning never stops’. But it’s also true that, if the mature system has firmly learned that, say, verb identity is not a good predictor of active vs. passive structure, then it might be difficult (for adults) to learn new verb biases with respect to active/passive choice within a short task. A graduate student here, Amanda Kelley, has found just this: She also finds that adults can learn new dative verb biases in language production, but that they seem not to learn active/passive verb biases. It might be that children would have more flexibility there, simply because they might have less firmly learned that the active/passive choice does not depend on the verb.

      1. The interaction between verb-general and verb-specific learning is so cool!

        I imagine an efficient/ideal learner would process each occurrence of language inputs at multiple levels simultaneously, i.e. does verb X prefer this argument structure & are verbs informative for parsing in general. The latter seems to rely on memory trace that is more widely distributed across items, and therefore more difficult to keep track of. But if young children are readily able to do both, then attention and working memory capacity should not be an issue here.

Leave a Comment or Question Below