[developers] HCONS in MRSs produced by PET
fcosta at di.fc.ul.pt
Fri Jun 5 12:21:34 CEST 2009
I'm having problems with the MRSs produced by PET with a grammar for
Portuguese. I can't reproduce them with the ERG or GG, so I'm asking if
someone has had a similar problem in the past and has any clue of what
may be wrong.
In short, I get no more than 1 or 2 elements in HCONS when parsing a
sentence with PET, even though the MRSs are fine under the LKB.
If I also use the -tsdbdump option and then export the sentences in
[incr tsdb()], I get feature structures for the top node that seem
alright, but the MRS are still not showing the entire HCONS.
For example, for a sentence with a fs for HCONS like:
HCONS diff-list & [
LIST #35 & cons & [
FIRST #36 & qeq & [
HARG #9 ],
REST #37 & cons & [
FIRST qeq & [
HARG #19 ],
REST #38 & cons & [
FIRST qeq & [
HARG #28 ],
REST #39 & list ] ] ],
LAST #39 ] ],
which seems ok and has three elements in it, I get MRS representations
with the following HCONS:
HCONS: < h5 QEQ h7 h1 QEQ h8 >
with only 2 elements. All formats of MRS and RMRS exportable by [incr
tsdb()] show only these 2 elements in HCONS.
The composition of HCONS and RELS in the Portuguese grammar is done in
the exact same way, and there is no similar problem with RELS.
Thanks in advance,
More information about the developers