[1-10] Constraints for Natural Language Processing

[This entry reprinted by kind permission of the author: Philippe
Blache <pb@llaor.unice.fr>, University of Nice-Sophia Antipolis]

Current linguistics theories often describe syntactic relations as
constraints on linguistic structures. It is in particular the case of
unification-based theories. But linguistic constraints are generally
far from the CLP ones and the question remains: is NLP a constraint
satisfaction problem ? [[MBJ adds an editorial note: Micha Meier
<micha@ecrc.de> feels that NLP _is_ a typical CSP.]]

It is still difficult to say what could be the actual advantages of a
CLP approach for natural language processing in general. But if we
don't have a global answer, several works propose CLP representation of
particular problems such as linear precedence (cf Patrick Saint-
Dizier), disjunctive values (cf Franz Guenthner), subcategorization,
etc ... I'm currently working on the interpretation of HPSG's
principles with boolean constraints. The problem in this case comes
from the fact that the instantiation principles of this theory can be
seen as constraints on feature structures, but using actual active
constraints need a very rigid (and heavy) representation of these
structures. A compromise between a pure CLP and a pure linguistic
approach is still inevitable.

I would be deeply interested in other approaches to this problem.

Franz Guenthner (1988) "Features and Values 1988"
	CIS-BERICHT-90-2, University of Munich

Patrick Saint-Dizier (1994) Advanced Logic Programming for Natural Language
	Processing, Academic Press, London

Philippe Blache (1992) "Using Active Constraints to Parse GPSGs"
	in proceedings of COLING'92