universality of gfs

[Avery Andrews]

Rereading this, I realized I didn't say quite the right thing:

> How well does it really work to do it in terms of strictly semantic
> Agent and Patient concepts?  ...

What I should have said is `how well do Agent and Patient really work
for cross referencing' (and case-marking if there is any), since these
are easy to observe for large numbers of verbs, while the standard
grammatical tests aren't (easy to do for 5 verbs, pretty hard for 500).

The idea being that if Agent and Patient can't really account for the
basic morphosyntax, then there is are some semantically abstract
structural notions in the system.

  Avery