[Corpora-List] TextGraphs-6 - Call for Papers - ACL-HLT 2011 Workshop

John F. Sowa sowa at bestweb.net
Mon Feb 7 19:58:05 UTC 2011


On 2/7/2011 1:01 PM, Rada Mihalcea wrote:
> In addition to semantic networks, which are indeed an instance of "graphs
> for NLP," there is also a lot of work involving syntactic graphs,
> discourse graphs, graphs built using sentence or document relations, etc -
> so I think this area of work is broader than just semantic networks.

All those aspects have been included under the term 'semantic network'
since the beginning (early 1960s).

If you like syntactic graphs, note that the 1961 conference included
David Hays, who formalized a version of Tesnière's dependency graphs
and their relationship to context-free grammars.  At that same
conference, Margaret Masterman used the term 'semantic network' for
her lattice structures.  And Silvio Ceccato presented correlational
nets, which included a set of 56 different relations for case roles,
part-whole, type-subtype, and inter-sentence relations.

In the later 1960s, Robert Simmons helped to popularize the term
'semantic network' as a generic for all of them.  Roger Schank
adopted dependency graphs, but put the word 'conceptual' in
front to emphasize a shift away from syntax.  Stuart Shapiro
adopted the term SNePS (Semantic Network Processing System)
for a version that was supported full first-order logic.

The SNePS system has been in continuous development and use for
research and applications from 1969 to the present.  In addition
to NLP, Shapiro's students and colleagues have used SNePS to support
language interfaces to computer vision, robotics systems, learning,
etc.  Following is a list of their publications from 1969 to 2010:

    http://www.cse.buffalo.edu/sneps/Bibliography/

For discourse, I showed that Kamp's discourse representation
structures are isomorphic to Peirce's existential graphs,
which I used as the logical foundation for the version of
conceptual graphs that I developed.  Kamp acknowledged that
point in the preface to Kamp & Reyle (1993).

For the broader use of graph algorithms, the chemists have been
in the forefront of developing the most sophisticated algorithms
for processing, indexing, and classifying large graphs and
large collections of graphs (multimillions of them).  The first
high-speed system for indexing and searching large systems of
semantic networks was developed by applying chemical algorithms
to conceptual graphs:

Levinson, Robert A. (1992) Pattern associativity and the retrieval
of semantic networks, Computers and Mathematics with Applications,
vol. 23, no. 6-9, pp. 573-600.

> Indeed, there is work on eigenvalues on semantic networks...

Yes.  The chemists have been using that work for years, and we have
been applying it to conceptual graphs at our VivoMind company.  We
also use the relations of Rhetorical Structure Theory to relate
graphs of sentences, paragraphs, and documents.

On 2/7/2011 1:20 PM, Vlado Keselj wrote:
> Eulerian graphs derived from n-gram tables should be added...

Yes.  The field has been continuously adopting and adapting
algorithms from everywhere since the beginning.  See the
following two collections of papers from the early 1990s:

Sowa, John F., ed. (1991) Principles of Semantic Networks:
Explorations in the Representation of Knowledge, Morgan Kaufmann
Publishers, San Mateo, CA, 1991.

Lehmann, Fritz, ed. (1992) Semantic Networks in Artificial
Intelligence, Pergamon Press, Oxford. Also published as a
special issue of Computers and Mathematics with Applications 23:6-9.

In all these developments, there is a continuum with multiple
overlaps and researchers borrowing and adapting ideas from
other researchers.  There has never been a sharp break.
Even the Semantic Web is just a giant semantic network.

John

_______________________________________________
Corpora mailing list
Corpora at uib.no
http://mailman.uib.no/listinfo/corpora



More information about the Corpora mailing list