Led by DPF and FCB, scribed by FCB.
We had a very informative discussion. I have tried to put most of the information in the new GrammarPerformance page (and not recorded it here), but there was a new idea that did not fit well there.
An idea for a useful tool (maybe part of TypediffTop):
Could we parse something, look for edges that appear and are never in a real sentence, and offer them to the grammarian as places to try to get rid of. Look at something like the small trees used in the maxent model? (Glenn and Dan)