[developers] Semantic treebanking without grammar?
berthold.crysmann at gmail.com
Sun Jun 30 20:10:08 CEST 2013
On 30/06/13 00:41, Stephan Oepen wrote:
> hi berthold,
> i think there are two variables that need to be set:
> (setf lkb::*tree-discriminants-mode* :modern)
> (setf lkb::*tree-display-view* :modern)
> plus there was an assumption that the ‘mrs’ field be populated in the
> ‘result’ relation, but the ‘derivation’ field be empty. i have purged the
> latter assumption, and added to the [incr tsdb()] podium two toggles:
> Trees|Switches|Semantic Discriminants
> Trees|Switches|Show Elementary Dependencies
> there is a new 64-bit run-time binary in the LOGON tree, so i would
> suggest you ‘make update’ and then try with the two switches above
> set from the podium.
> please let me know in case you run into more problems! oe
thanks a lot for this. I shall try it out over the next few days,
hopefully tomorrow already.
BTW: WIll it be possible to use the LOGON infrastructure to train and
test models derived from these treebanks? I guess we can produced TAG
derived trees in Lisp format, and since most treebanking features (types
vs. instance being the exception) operate on the CF backbone I do not
see why in principle that shouldn't work.
If that should not be easily feasible, do you reckon the export will
work oput of the box?
>> we would like to do some MRS treebanking on the output of a French TAG.
>> Obviously, we cannot replay derivations with the LKB, so
>> we thought that we could simply treebank using the discriminators extracted
>> from the MRS. I quickly tried this out with a Hausa treebank, i.e.
>> activating Trees|Annotate without any grammar loaded, and the fine system
>> just outputs error messages complaining it cannot reconstruct the trees.
>> *tree-discriminants-mode* is set to :modern.
>> Is it possible at all? I remember an LREC 2006 paper on this. What do I have
>> to do configuration-wise to make it possible?
Berthold Crysmann <crysmann at linguist.jussieu.fr>
CNRS, Laboratoire de linguistique formelle (UMR 7110), U Paris Diderot
Case 7031, 5 rue Thomas Mann, 75205 Paris cedex 13
Bureau 545, bâtiment Olympe de Gouges, rue Albert Einstein, 75013 Paris
More information about the developers