Research on child language acquisition would benefit from the availability of a large body of syntactically parsed utterances between parents and children. We consider the problem of generating such a ``treebank'' from the CHILDES corpus, which currently contains primarily orthographically transcribed speech tagged for lexical category.