Re: veridicality in grammar

[Chris Handley <chandley@OTAGO.AC.NZ>]

Bob writes:

>It is grammatically incorrect to put a value into the wrong
>grammatical category.  In English it is grammatically incorrect to put
>a plural value into a singular grammatical category.  Among the
>Dyirbal, it is correct to put cigarettes and edible fruit into the
>same grammatical category and incorrect to put them into the same
>category as meat and fire, which are in two other categories.  In
>Lojban, it is grammatically incorrect to put a `for real' value into a
>`I designate as' category.
>
>You are right, this Lojban parse goes against the way a grammar works
>that is based on parse trees.  I run the parser program on both
>utterances, and both are OK.  Nonetheless, a fluent speaker should
>*feel* that the grammar are wrong, just as you *feel* the grammar
>error I just made.
>
>As far as I can see, this means either that Lojban has a grammar that
>cannot succeed in practice (meaning that fluent Lojban speakers will
>not be able to distinguish {le} and {lo} by `feel' as I am
>hypothesizing), or it has a grammar that has more characteristics than
>a contemporary computer programming language.
>
>I suspect the latter, but rather surprised and saddened by it, since I
>did not expect it, and was hoping for language with a `complete'
>definition.
>
The problem is that all the parsers we write are based on context free
grammars. If I write the local equivalent of
   a <- b + c
in most programming languages, this would be acceptable as an assignment of
the value of an expression and therefore would be syntactically correct (it
would parse). If, however, I was using a strongly typed language and a, b
and c were of incompatible types then it would be unacceptable, but _at a
different level_ (and quite possibly checked by a different part of the
compiler). It would in fact be tha same as the category errors Bob was
talking about.

Once we admit this possibility our language becomes a context sensitive
language, and the grammar is no longer LL(1). If one one wishes to catch
errors of this type in your parser, it will become several orders of
magnitude larger, and still won't always work. These questions, in the same
way as the usages of tense and number, will have to be left to fluent
speakers of the language.

Chris Handley.



======================================================================
Chris Handley                                     chandley@otago.ac.nz
Dept of Computer Science                       Ph     (+64) 3-479-8499
University of Otago                           Fax     (+64) 3-479-8529
Dunedin, NZ
______________________________________________________________________
 "Dyirbal speakers mark entities as belonging to the `balan' category,
  which includes `women, fire, and dangerous things'."

                                                    Robert J. Chassell