Discussion:
[lkb] Expanding out Defaults
Francis Bond
2006-02-01 06:46:32 UTC
Permalink
Ann,

The Korean Resource Grammar uses defaults, which means we cannot use
it with PET. Is it possible to dump the grammar with all the defaults
resolved, and then load that into PET? I think you said it was
theoreticaly possible. If there is currently a handy command to do
it, could you let us know what it is?

--
Francis Bond <www.kecl.ntt.co.jp/icl/mtg/members/bond/>
NTT Communication Science Laboratories | Machine Translation Research Group
Ann Copestake
2006-02-01 14:10:14 UTC
Permalink
Is it possible to dump the grammar with all the defaults resolved, and then
load that into PET?
in principle I would expect so (provided it's not using persistent defaults),
but there are probably complications - e.g. the way that PET uses reduced
feature structures. I think it would be OK, provided that you output the
types without any defaults and the lexical entries (and rules etc) with all
the default information included. i.e., the use of defaults should be
formally equivalent to a grammar where there is more information in the
entries.
If there is currently a handy command to do
it, could you let us know what it is?
there's code that was used to dump out the lexical entries for PET and lilfes
in io-general/outputsrc.lsp - I don't know whether it still works.

I've CC'd to Ben explicitly because he should be able to give you more
detailed info about doing this.

Ann
The Korean Resource Grammar uses defaults, which means we cannot use it with
PET. Is it possible to dump the grammar with all the defaults resolved, and
then load that into PET? I think you said it was theoreticaly possible. If
there is currently a handy command to do it, could you let us know what it
is?
Ben Waldron
2006-02-10 19:41:35 UTC
Permalink
Post by Ann Copestake
Is it possible to dump the grammar with all the defaults resolved, and then
load that into PET?
in principle I would expect so (provided it's not using persistent defaults),
but there are probably complications - e.g. the way that PET uses reduced
feature structures. I think it would be OK, provided that you output the
types without any defaults and the lexical entries (and rules etc) with all
the default information included. i.e., the use of defaults should be
formally equivalent to a grammar where there is more information in the
entries.
If there is currently a handy command to do
it, could you let us know what it is?
there's code that was used to dump out the lexical entries for PET and lilfes
in io-general/outputsrc.lsp - I don't know whether it still works.
I've CC'd to Ben explicitly because he should be able to give you more
detailed info about doing this.
Sorry for the lack of prompt response; this email slipped under my
radar. I'm afraid I probably don't know much more than you about this
issue. I wrote some code long ago which dumps the lexicon out in TDL
format (basically for use with the database lexicon -- see below). But
this simply dumps the PATH = VALUE lexicon slots in TDL, before any
expansion occurs -- so it won't help you with your defaults... I'm not
familiar with the code inside io-general/outputsrc.lsp.

LKB(17): (time (export-lexicon-to-tdl :file "~/lexicon.dump.tdl"))
(LexDB) export filename: ~/lexicon.dump.tdl
(LexDB) caching all lexical records
(LexDB) caching complete
(LexDB) emptying cache
; cpu time (non-gc) 17,340 msec user, 60 msec system
; cpu time (gc) 670 msec user, 0 msec system
; cpu time (total) 18,010 msec user, 60 msec system
; real time 19,832 msec
; space allocation:
; 15,971,259 cons cells, 128,749,552 other bytes, 0 static bytes
#<PSQL-LEX-DATABASE @ #x4483b08a>

gives

1000s_n1 := n_plur_xmod_le &
[ STEM < "1000s" >,
SYNSEM [ LKEYS.KEYREL.CARG "1000",
PHON.ONSET con ] ].

100s_n1 := n_plur_xmod_le &
[ STEM < "100s" >,
SYNSEM [ LKEYS.KEYREL.CARG "100",
PHON.ONSET con ] ].

...

- Ben

PS. Francis, any advice your group might have regarding your experiences
compiling PET with ECL would be much appreciated.
Francis Bond
2006-02-11 00:19:12 UTC
Permalink
G'day,
Post by Ben Waldron
Sorry for the lack of prompt response; this email slipped under my
radar. I'm afraid I probably don't know much more than you about this
issue. I wrote some code long ago which dumps the lexicon out in TDL
format (basically for use with the database lexicon -- see below). But
this simply dumps the PATH = VALUE lexicon slots in TDL, before any
expansion occurs -- so it won't help you with your defaults... I'm not
familiar with the code inside io-general/outputsrc.lsp.
[snip]
Post by Ben Waldron
PS. Francis, any advice your group might have regarding your experiences
compiling PET with ECL would be much appreciated.
Eric Nichols is the expert. We have everything working finr for linux
x32, but are also having problems with x64.

--
Francis Bond <www.kecl.ntt.co.jp/icl/mtg/members/bond/>
NTT Communication Science Laboratories | Machine Translation Research Group
Loading...