[logon] [peter.adolphs at dfki.de: stable leaf type order]

[Stephan Oepen]

dear all,

originally, i believe, i had proposed this past week as the code freeze
deadline for the LOGON Barcelona release.  for various reasons, we are
running a little behind schedule.

i checked in a number of revisions yesterday and today: new binaries of
PET (see the message below for details), some changes submitted by eric
and berthold, the latest revisions from the LKB `trunk', minor updates
to the ERG (tokenization and chart mapping), and fresh LOGON binaries.

i would encourage everyone to `make update' their LOGON tree and start
testing using the emerging release candidate.  francis and eric are now
finalizing JaCY and JaEn, and dan and i still plan a few small changes
to the ERG.  other than that, i am not aware of pending revisions that
should be included in this release?

my proposal is to finally freeze the repository for burn-in testing on
tuesday this coming week (14-jul).  please let me know if you had plans
to commit further updates for this release.

                                                      all best  -  oe

+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
+++ Universitetet i Oslo (IFI); Boks 1080 Blindern; 0316 Oslo; (+47) 2284 0125
+++     CSLI Stanford; Ventura Hall; Stanford, CA 94305; (+1 650) 723 0515
+++       --- oe at ifi.uio.no; oe at csli.stanford.edu; stephan at oepen.net ---
+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++


[--- start of forwarded message ---]

Date: Thu, 09 Jul 2009 17:19:02 +0200
From: Peter Adolphs <Peter.Adolphs at dfki.de>
To: Bernd Kiefer <kiefer at dfki.de>, 
    Dan Flickinger <danf at csli.stanford.edu>,
    Stephan Oepen <oe at ifi.uio.no>, Yi
    Zhang <yzhang at coli.uni-sb.de>
Subject: stable leaf type order

Dear colleages!

I've changed the flop code to create a stable partition of all types
into leaf types and proper types, i.e. they are now arranged in the
same order as they are defined in the TDL files. I also changed cheap
to store and apply the rules in the order as they are defined in the
grammar dumps instead of the reverse order, as it was up until now.

These changes are crucial for chart mapping, because the rules have to
be applied in the order as they are defined in the source files. That
everything worked fine with the current cheap + erg was just a
coincidence.

The changes should have no effect on normal parsing (if no limits are
hit) but of course on chart mapping. If you want to use chart mapping,
you should make sure to use the latest version of flop and cheap,
otherwise chart-mapping rules will be applied in the wrong order. The
performance of both flop and cheap wasn't affected according to my
experiments. I didn't get any different results with -nsolutions=0
(./parse --best 0) on the Cathedral and the Bazaar.

Bernd, Stephan and I decided to increase the version number of the
grammar dump because of the renumbering of types. If the version of the
grammar dump and of cheap don't match, there will be a warning but the
grammar is still loaded. This way, we can ensure that users with old
cheap + new dump or new cheap + old dump will get a warning.

I've commited the changes to the cm branch. I suggest to apply them to
main before Barcelona if you don't encounter any problems with it.

Cheers,

Peter

[--- end of forwarded message ---]