[Corpora-List] context-DEPENDENT grammar simulator.

[Mike Maxwell]

(Rob and Torbjörn, perhaps we should take this off-line, before someone 
blows the whistle on us...  If anyone wants to join this discussion, let 
us know)

Torbjörn Lager wrote:
> On Thu, Apr 10, 2008 at 6:27 PM, Rob Malouf <rmalouf at mail.sdsu.edu> wrote:
>>> Rob Malouf wrote:
> ...
>>  I think it should be fairly easy to write an HPSG grammar for the
>>  a^nb^nc^n language using subcat lists [notice the hedges... meaning I
>>  haven't actually done it].
> 
> I just did it, but in Definite Clause Grammar (DCG):
> 
> s --> a(N), b(N), c(N).
> 
> a(0) --> "".
> a(s(N)) --> [a], a(N).
> 
> b(0) --> "".
> b(s(N)) --> [b], b(N).
> 
> c(0) --> "".
> c(s(N)) --> [c], c(N).

Prolog used to be my favorite programming language, but I forget... I 
presume that s(N) means "successor of N", i.e. the number that comes 
after N (N >= 0; otherwise a 'cut' would be needed to prevent this from 
going on infinitely, with N becoming negative, and no guarantee that the 
a's, b's and c's would then be equal).  If that's the right 
interpretation, then this is taking advantage of the fact that Prolog 
can do infinite arithmetic.  But arithmetic cannot be defined in CFPSG, 
AFAIK (beyond the trivial listing of a finite number of arithmetic 
facts), so I don't think this counts as a counterexample to the claim 
that HPSG cannot do strictly CSPSG (or possibly "mildly context 
sensitive PSG").

> So that's a context sensitive _language_. I would hesitate to call the
> DCG a context sensitive _grammar_ though. As far as I remember,  a
> context sensitive grammar has to have a particular form, with rules
> allowing more than one symbol on the left hand side.

That's how it's usually defined, but the unchanged symbols can be moved 
over to an environment, giving you CSPSG rules that look much like 
phonological rules.  It's just a different notation.  I suspect DCGs 
have general cspsg capacity, since they can contain arbitrary Prolog 
clauses.
-- 
    Mike Maxwell
    "We signify something too narrow when we say:
    Man is a grammatical animal. For although there
    is no animal except man with a knowledge of grammar,
    yet not every man has a knowledge of grammar."
    --Martianus Capella, "The Seven Liberal Arts"

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