Dear Rui,
The manual needs some updating in these regards. Currently, we have two training sets. The Wright-train is from adults doing picture descriptions and narratives and the Eve-train includes all 20 of the Eve files from Brown 1973. The Eve files were very much out of date, but I brought them back up to date just now. In general, tagging of child utterances, particularly the shortest ones, is never going to be determinate. However, using TRNFIX, the total number of words differing between the %trn and the %mor lines is not too large and mostly is for relatively uninteresting ambiguities. We are developing some POSTMORTEM rules to fix some of these remaining things, such as the treatment of “right” when it appears alone.
There is now no difference between the Eve corpus in the ENG MOR grammar and the one in the database, but make sure you get new versions of ENG MOR for your work.
—Brian MacWhinney