[developers] morphology - testing and a couple more things

Francis Bond fcbond at gmail.com
Tue Sep 6 14:54:59 CEST 2005


> Official testers - please take this as a request to test if you haven't already
> done so!  I have had no feedback whatsoever about the revised code.  I am not
> sure whether that simply means nobody has tried it on their grammars.  If you
> have tried it, and it works, please let me know that too.  It is important
> it gets tested reasonably soon because it's very difficult for me to find time
> to fix bugs during term time, plus I'll forget the details of the code.

Sorry I haven't been able to do this yet.   I probably won't get to it
until next week, at the MT Summit.

Is there a compiled version of this version of the lkb?  I am not sure
if all the official testers have ACL (Jong-Bok do you?).

> I thought I'd allowed for every case of escaping in the code, but I didn't
> think of people wanting to name letter sets !, ?, *, \ or ).  Rather than
> changing things again, and complicating the code some more, I would just ask
> people to avoid using those characters as the names of letter sets and not to
> put escapes inside letter set specifications in any context.  I'm afraid I
> think this means the checked in version of the ERG may not work because it has
> !\?. The fixes I have made to it in my local version are to rename !?, to get
> rid of !b (which was giving an unmatched letter set warning, although that's
> not fatal) and to not load the file with the double escape characters.
> The other thing I thought I should mention is that because the rule table no
> longer gets built using all rules, you will not get warnings of possibly cyclic
> applications of rules which have a morphological effect.  This is nevertheless
> something that should be avoided when it is impossible, since it will lead to
> unnecessary analyses being tried and may give some spurious parses.  If it's
> thought useful, I can easily make it possible to check all feeding
> relationships - e.g., make it a menu option.  Let me know.

Has anyone else had a look yet?

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

More information about the developers mailing list