0440949 Andreas van Cranenburgh, February 19th, 2010
Cognitive models of language, assignment 3.

1. The paper assumes that the syntactic annotations are representative of
children's knowledge of language, and that children produce their utterances
in an economical way: using the most likely derivation and re-using abstracted
constructions from previous utterances. A different, though perhaps unlikely,
hypothesis could be that children do not store any abstractions, but
progressively acquire more concrete elements that are recombined on-the-fly,
as in a truly memory-based model. 

Another thing is that the kind of annotations determine the sort of
constructions that will be found, because substitutions depend on matching
categories. I believe that the development of representations is the drive
behind language development, so progressive abstraction would be a result of
better and richer representations.

2.
  - Take one or more children and record every utterance they make and hear
    from birth till three years of age.
  - Measure abstractness by dividing novel utterances by total utterances
    at one month intervals. Novel utterances are utterances which have not
    been heard before by the child, not even as part of a larger utterance.
    In other words, when the child has heard "do you want a cookie", then
    the utterance "want cookie" is not novel. "Want bear", however, could be
    a novel utterance if "bear" only occurred in utterances without "want"
    before.

The advantages of this approach are that it is fully naturalistic and that it
is complete, every utterance can be retraced to past utterances. And because
the criterium for novel utterances is so strict no assumptions need to be
made about representations or anything else not directly observable. 
The disadvantage is that it is a lot of work.