W3C home > Mailing lists > Public > public-swbp-wg@w3.org > September 2005

Re: [WNET] RE: WordNet Conversion

From: Mark van Assem <mark@cs.vu.nl>
Date: Tue, 13 Sep 2005 18:31:22 +0200
Message-ID: <4326FEDA.9030302@cs.vu.nl>
To: public-swbp-wg@w3.org


Hi Brian,

> I think there are use cases where what is important is:
> 
>   1. the ability to name words and wordsenses
>   2. the ability to navigate the Wordnet database - e.g. find synonyms
> etc
> 
> This can be accomplished without using Owl at all.  Have we other

For these use cases, you are right, RDF(S) is enough. OWL can help in 
e.g. (1) checking the data and (2) automated classification of 
instances and class subsumption.

(1) For example, I did not detect an error in an earlier version of my 
conversion in which some synsets were both a noun and a verb. Stating 
wn:Noun owl:disjointWith wn:Verb and running the data through an OWL 
reasoner would prevent this. Using OWL for these kinds of check is 
probably simpler than writing a checking program each time for each 
separate dataset (of course this issue plays a role not only in WN).

(2) For example, the class SynsetUsedAsClassifier is equivalent to the 
Synsets that have a value for the property wn:classifies (see [1]). 
Other definitions are possible, e.g. MonosemousWord which is the Words
that have exactly one value for the property wn:sense (which has range 
WordSense).
Membership of these classes does not depend on explicitly asserted 
rdf:type links, but can be inferred instead.

In my opinion, it is possible to let RDFS and OWL co-exist. There are 
two options:

(a) two seperate versions: RDF data + RDF schema; RDF data + OWL schema)
(b) one version of the RDF data, two separate schemas (one RDFS, one OWL)

In option (a), the RDF data that belongs to the RDF schema requires 
all explicit rdf:type and rdfs:subclassof instances that can be 
inferred in the OWL version (the RDF version belonging to the .

In option (b) the RDF data again contains all the explicit instances, 
on which both schemas can operate.

 > 2.  Naming of things

<snip>

 > from the concept id part.  This would avoid the redirect, but I've
 > sneaking suspicion I've forgotten why that is a bad idea.  I also

I didn't follow this discussion, so any suggestion on how the URIs 
should be constructed and why is welcome.

 > wondered about using synset id's which change from version to 
version of
 > wordnet.

This problem is a tough cookie, and actually exists for ANY resource 
that is periodically converted to RDF, which is why I am uncertain on 
how it is best handled. See also the discussion we had in the PORT TF 
about this [2].

 > 3.  Do we know what other Wordnet's in RDF are out there and how they
 > differ?
 >

See [3] for some WN's. Important differences: some are incomplete (not 
all relations converted), some do not have a URI for WordSenses, see 
also [4].

Regards,
Mark.

-----
[1] http://www.w3.org/2001/sw/BestPractices/WNET/wordnet_datamodel.owl
[2] http://lists.w3.org/Archives/Public/public-esw-thes/2005Jul/0051
[3] http://www.w3.org/2001/sw/BestPractices/WNET/tf.html
[4] http://www.cs.vu.nl/~mark/wn/wordnet_rationale_mark.html

-- 
  Mark F.J. van Assem - Vrije Universiteit Amsterdam
        mark@cs.vu.nl - http://www.cs.vu.nl/~mark
Received on Tuesday, 13 September 2005 16:31:35 UTC

This archive was generated by hypermail 2.4.0 : Friday, 17 January 2020 17:31:12 UTC