[topicmapmail] TM and NLP

[Dan Corwin]

Bob Parks wrote:

   > Has anyone seen an application of topic maps to linguistic
   > data? I'd appreciate comments, suggestions, pointers.

Topic Maps make an excellent conceptual memory model
for a natural language parser.  NPs it detects can map
readily into topics.  Sentences and prepositional phrases
map readily into their associations.  Roles used for the
NP topics can be formally constrained by linguistic theory
for case frames - and enforced by OWL, TMCL, etc..

Scope seems helpful for tagging any competing parses that
emerge during syntax analysis.  And after ambiguities get
resolved, the parser can dump out surviving topic models
into an XTM file, for independent high level processing
by remote client applications.  More thoughts are at:

      http://www.lexikos.com/nlptools.jsp

These web pages are merely drafts, still disjoint from my
main web site.  They propose a project to build MODELER, a
toolkit for encoding NLP lexicons under XTM 1.0.  They also
show other uses TMs might have in a text-analysis suite.

<soapbox: building it will take non-trivial teamwork.>

To develop this, I must advance from a part-time lurker
on this list into a full-time XTM hacker.  (You have
fully convinced me XTM is the right technology for it.)
But I also must attract help on several big sub-tasks,
many of which demand more than just discussions here.

So on New Year's Day, I resolved to start an open source
effort on MODELER, and seek out support within NLP, XTM,
and J2EE interest groups for developing it.

Bob, you showed NLP activity via your September post on
dictionary APIs.  Perhaps you - and/or other list lurkers or
readers with NLP/XTM thoughts - would be willing to put
real resources into making this product concrete and useful.
By design, it is readily customizable to multiple goals.

If so, RSVP - here or privately.  NLP and XTM complement
each other nicely, but they need some willing firms and
individuals to follow suit.  I'd welcome your support.

Cheers,
Dan Corwin