Re: Runtime "learning"

Thant Tessman <thant@acm.org> wrote in message news:01-11-023...
> Doru-Catalin Togea wrote:
>
> [...]
>
> > What I am asking for is: Is there any tool which can take me a step
> > further? Can it build an EXECUTABLE file, which, AT RUNTIME, learns
> > a language by reading its grammar, and thus becames able to
> > recognize (and take actions based on) sentances in that language?
>
> Out of curiosity, do you plan on somehow dynamically building new
> actions to associate with new productions? This is something I can't
> imagine being easy in any non-functional programming language. And


Even if the set of possible actions is Turing-complete, you define it
in an interpreted programming language and pay the performance
penalty.


> without this ability, I have a hard time imagining how dynamically
> modifying a grammar would be useful. Changing whether or not something


In programming we could use a recognizer for an arbitrary text file
that finds out its language where possible. In natural language
processing world-wide, writers of emails often lapse into their first
language to express some thoughts, sans peur et sans reproche.


> is considered a keyword by the lexer at runtime doesn't require
> modifying the grammar itself, does it?


In the general sense it does. Language designers have finally learned
NOT to redefine the language by means of "optional parameters" or
controls on the graphical user interface since this makes code
uniquely not reusable.


> Also, what do you want to happen if a new production introduces an
> ambiguity into your grammar?


I am not a formal language theorist, although I play one in the
movies, and my understanding is that you can test for this sort of
thing.


> [...]
>
> -thant
> [These problems were also solved 25 years ago. -John]
[These problems were still solved 25 years ago. -John]

Return to the comp.compilers page.
Search the comp.compilers archives again.