- From: Danny Ayers <danny666@virgilio.it>
- Date: Sat, 12 Jun 2004 12:40:07 +0200
- To: gangemi@loa-cnr.it
- Cc: 'SWBPD list' <public-swbp-wg@w3.org>
I just read the paper [1] by Aldo et al about the application of the OntoClean methodology to WordNet's top level. Interesting stuff. The paper describes research in progress - I'd be very grateful to hear the current status. I also think it highlights a potential difficulty that is likely to be encountered by other groups wishing to express other wordnets in RDF/OWL. The WordNet vocabulary is derived from the scruffy world of human language, and the description of the relationships between the terms presupposes that there are systems to be found in the data. The first pass at going meta (WordNet) yields something that was still fairly random when viewed from above (the meta-properties of OntoClean). The amount of heterogeneity (big variations in granularity, uncomfortable mixes of types) in the pre-OntoClean top level, and the amount of 'cleaning' that was needed might suggest that the DL style of modelling might not in fact be such a good fit. Is it safe to assume the heterogeneity was simply the result of human imperfection at classifying terms? Might that be helped by provision of a framework extending up to OntoClean's level earlier in the process? Probably stretching the point, I read somewhere recently how the categorisation used by Roget in his thesaurus said a lot about his own world view (some religious connection?). Might there not be a danger of WN in OWL saying as much about the OWL world view as the information it is intended to capture? I don't think it would actually matter too much in practice if the primary application of OWL wordnets is to provide common reference points between disparate ontologies or data sets, but it might suggest to what extent such mappings could be relied on. One of the proposed TF objectives is to provide guidelines for wordnets-to-OWL translation, so perhaps this might call for a little further analysis. I'm not familiar with the literature, presumably there will have been plenty of previous attempts at mapping from human lexicons to formal structures - is there any way of judging which approaches were successful? Cheers, Danny. [1] http://citeseer.ist.psu.edu/gangemi01conceptual.html -- Raw http://dannyayers.com
Received on Saturday, 12 June 2004 06:41:47 UTC