David Barbour <[email protected]> writes:

> Regarding the language under-the-hood: If we want to automate software
> development, we would gain a great deal of efficiency and robustness by
> focusing on languages whose programs are easy to evaluate, and that will
> (a) be meaningful/executable by construction, and (b) avoid redundant
> meanings (aka full abstraction, or near enough). Even better if the
> languages are good for exploration by genetic programming - i.e. easily
> sliced, spliced, rearranged, mutated. I imagine a developer who favors such
> languages would have an advantage over one who sticks with C.

My interest in Programming Language Theory sprang from my interest in AI
and code generation (things like the Goedel Machine which Simon
mentioned), specifically what makes a language easy/hard to use by a
machine.

Regarding 'syntaxless' languages, I agree that this is a desirable
property for the IR of a program-generator. Due to the combinatorial
nature of the problem, we need all the efficiency we can get. Removing
syntax can improve this by dramatically reducing the search space
without sacrificing potential solutions.

However, there can often be a semantic cost in trying to assign meaning
to arbitrary combinations of tokens. This can complicate the runtime
(eg. using different stacks for different datatypes) and require
arbitrary/ad-hoc rules and special-cases (eg. empty stacks).

These can make the landscape smoother, but lead to larger solutions (in
bits), which is the dominant factor on scaling. It can also make
learning/reasoning about the language and problem domain harder for
meta-level algorithms, eg. Estimation of Distribution.

I think this semantic cost is often not appreciated, since it's hidden
in the running time rather than being immediately apparent like
malformed programs are. At the end of the day, it doesn't matter if we
hit millions of syntax errors if we get good quality solutions in an
acceptable amount of time. It may be the case that telling our
algorithms when they're making no sense will give them the information
they need to find better solutions faster.

Cheers,
Chris
_______________________________________________
fonc mailing list
[email protected]
http://vpri.org/mailman/listinfo/fonc

Reply via email to