[developers] PET with inflectional and lexical rules
Stephan Oepen
oe at csli.Stanford.EDU
Wed Apr 5 14:20:03 CEST 2006
hi yi,
> (4) `abraham likes the cat.' [0] --- 0 (-0.00|0.01s) <25:21> (687.2K) [0.0s]
[...]
> Or is it likely to be a bug in PET?
i am afraid, the latter: until recently, the ERG was developed against
a different branch of PET (the older `oe' branch; see the wiki pages),
so for use with the ERG you had to install a different PET version.
however, your message gives us a reason to summarize some good news for
the `developers' list. bernd and i recently sat down jointly to merge
the two PET branches. we identified three changes that had to be made
in the `main' branch for use with the ERG:
- free interspersing of orthographemic and non-orthographemic lexical
rules (this is the problem you encountered above: the period needs
to be analyzed by an orthographemic rule prior to a null-inflection
of the singular noun);
- escaping of quotes and backslashes in [incr tsdb()] output; and
- correct measurement of `tcpu' and `total' values for [incr tsdb()].
bernd is in the process of preparing a new PET release that should then
fully support the ERG. i believe dan will develop against that unified
PET version in the future.
to get something going in the meantime, i had sent instructions on how
to obtain a LOGON tree to tim (for him and you to experiment with parse
selection performance). once you install the LOGON tree, it includes a
binary version of PET that already includes the changes above (i.e. in
a sense a pre-release of what bernd is preparing). you should be able
to just use the `flop' and `cheap' binaries from $LOGONROOT/bin/.
let us know how things go! all best - oe
+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
+++ Universitetet i Oslo (IFI); Boks 1080 Blindern; 0316 Oslo; (+47) 2285 7989
+++ CSLI Stanford; Ventura Hall; Stanford, CA 94305; (+1 650) 723 0515
+++ --- oe at csli.stanford.edu; oe at ifi.uio.no; stephan at oepen.net ---
+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
More information about the developers
mailing list