consider incorporating CGELBank's xpos

[nschneid]

CGELBank includes a bit of :xpos information—PTB tags for verbs (incl. auxiliaries) and numbers.

Is it worth putting this into the annotation task, or should it be added post hoc?

Asking the parser to distinguish VB vs. VBP, VBN vs. VBD, MD vs. other V_aux, etc. may lead to more accurate structures. OTOH, its tag predictions would have to be checked more carefully.

Note that the aux/non-aux distinction is partially orthogonal to the xpos distinctions from PTB. A tagset combining the info could be: MD VB VBaux VBD VBDaux VBG VBGaux VBN VBNaux VBP VBPaux VBZ VBZaux, replacing current V Vaux. (As well as adding CD to distinguish numerals from other Ds.)

activedop already does some transformations between the ptree/gtree and the ctree, so in principle we could add a bit more notational divergence (this would not affect structure).