Universal dependency features in CG

[Edward Garrett]

Hi,

I was wondering whether anyone in this group is using "universal features" (http://universaldependencies.org/u/feat/index.html) as part of their Universal Dependency part-of-speech tagging in Constraint Grammar. If so, are there any conventional or recommended ways of tagging universal features within CG. For example, a word could have the following set of features in addition to its POS tag: VerbForm=Fin|Mood=Ind|Tense=Pqp

The resource below on modelling dependency grammar in CG is helpful, but it does not address this point.

visl.sdu.dk/~eckhard/powerpoint/CG_ESSLLI_Thursday.pdf

Would there be any interest or value in adding and being able to operate on Key=Value type tags?

Thanks,

Edward

[Francis Tyers]

We're using CG for Universal dependencies for the Turkic languages, but we're only using them CG-natively with the deprels for now (we often do hand-annotation and using Key=Value would be a lot more cumbersome). We have a lot of scripts for converting between CG3 format and CoNLL-U format (including conversion of morphological features).

If you like I can write up how our "pipeline" works. Which language(s) are you interested in using it for?

Fran

--
You received this message because you are subscribed to the Google Groups "Constraint Grammar" group.
To unsubscribe from this group and stop receiving emails from it, send an email to constraint-grammar+unsub...@googlegroups.com.
To post to this group, send email to constraint-grammar@googlegroups.com.
Visit this group at https://groups.google.com/group/constraint-grammar.
For more options, visit https://groups.google.com/d/optout.

[Eckhard Bick]

Hello!

I think feature attribute pairs should stay implicit in CG itself, as
they are now, but they are easy to create with a post processor, which
you would need anyway to create output formats like xml or UD's CoNLL
format.

Even if one mapped VerbForm=Fin on a word with CG (which is entirely
possible), this would then be a tag as a whole, and make everything
manual more work, like rule writing or post-editing, because the new tag
is much longer than "Fin" on its own.

Best,
Eckhard

> --
> You received this message because you are subscribed to the Google
> Groups "Constraint Grammar" group.
> To unsubscribe from this group and stop receiving emails from it, send

> an email to constraint-gram...@googlegroups.com
> <mailto:constraint-gram...@googlegroups.com>.

> To post to this group, send email to

> constrain...@googlegroups.com
> <mailto:constrain...@googlegroups.com>.

> Visit this group at https://groups.google.com/group/constraint-grammar.
> For more options, visit https://groups.google.com/d/optout.

--
Eckhard Bick,
cand.med., dr.phil.
University of Southern Denmark
e-mail: eckhar...@gmail.com
web: http://beta.visl.sdu.dk