[Corpora-List] context-DEPENDENT grammar simulator.

[Michael Maxwell]

Rob Malouf wrote:
> NLTK also supports context sensitive grammars,
> in the form of feature-based grammars...

AFAIK, the use of features, even when it enforces agreement, does not make
a grammar context-sensitive; and I don't believe the feature-based
grammars in NLTK have context-sensitive power.  The theory of HPSG (and
before that, GPSG) is based on the assumption that feature grammars are
still context-free.

There are very few examples in natural language that seem to require
anything more than context-free grammars, and there's a substantial
literature on those few examples of "mild context sensitivity." 
Tree-Adjoining Grammars (TAGs) are claimed to have the appropriately mild
context sensitive power.

Phonological rules are often written (at least by those of us who haven't
succumbed to Optimality Theory :-)) using a notation that looks like
context-sensitive phrase structure rules.  But in fact phonological rules
have been shown to have only finite state power, under the assumption that
they cannot (without bound) re-apply to their own output.

Note that I am using the term "context-sensitive" in the mathematical/
Chomskian sense, not in the intuitive sense that what we say is sensitive
to the knowledge we have of the listener, environmental noise level, etc.
(as this term was used in one of the previous postings on this thread).

   Mike Maxwell
   CASL/ U MD


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