Re: SUO: WordNet Mappings
Also, of course, WordNet is being used more generally for
taxonomy/classification induction over document collections by emerging
software utilizing (generally non-symbolic) NLP technology. Some of
these systems are quite useful, but their value correlates with the
value of the underlying knowledge resources employed (such as WordNet or
ontologies).
Leo
John Bateman wrote:
>
> Hi Ian,
>
> thanks for the mapping paper: will read with interest.
>
> > First, nouns are clearly
> > favored in WordNet - there is much more information about nouns that there
> > is about other parts of speech. The second reason is that most verbs have a
> > nominal form,
>
> well, linguistically it is well known that one of the functions
> of nominalisations is to drop a lot of information--e.g., information
> about their temporal profile, their configuration of participants/semantic
> roles, aspect, and so on--so, at least from my own background
> in Natural Language Generation, we stick to events (and the verbal configurations
> that express them) as the basic starting point and if justified (from
> their use in a text) we can throw away a lot of information and use
> a noun, since most nouns behave pretty much the same. This is just to
> adopt an event semantics as "basic" whenever we can.
>
> > so that mapping all of the nouns gives you most of the verbal
> > > meanings.
>
> well, again, it might have done if, as you say,
>
> > it would be nice to have a pointer from the noun
> > synset to the corresponding verb synset, and I hope the WordNet people will
> > introduce this at some point.
>
> but they don't. This is probably another reason for moving on to the
> EuroWordNet upper structure rather than WordNet. I think these things
> have been thought of there more.
>
> There is also the question about the hierarchies of the synsets concerned
> with verbs and those concerned with nouns: I think (haven't looked at
> these for a while now) these look very different in WordNet, so there
> is some different information being taken into consideration.
>
> > However, we are not using this
> > structure in our mapping work. We are associating individual synsets with
> > SUMO concepts.
>
> This is a relief! :-)
>
> Clearly a lot of interesting stuff here: and a link to WordNet, for all
> its (i.e., WordNet's) drawbacks, is obviously a Good Thing. I look forward to
> more. There have been some moves to using WordNet information as a lexical
> resource for doing Natural Language Generation--I think having a link to
> something that is better organised, like the SUMO-to-be, will be a great
> help.
>
> Best,
> John B.
-