On Sun, Feb 10, 2013 at 10:38:03AM -0300, Felipe Gonçalves Assis wrote:
> If you input to a computer
> (1) {la .djan. cortu lo birka}
> (2) {.i ri birka ma},
> it could infer the answer to the question from the following general principles,
> which you may call semantic axioms:
> (3) {roda rode ganai cortu gi se pagbu}
> (4) {ro da poi birka ku'o ro de poi danlu ganai pagbu gi birka}
> and some data providing
> (5) {la .djan. danlu}
> Question (2) could trigger a search for data expressing a {birka} relation, or
> principles having it as its consequent. Finding (4) would trigger a search for
> things that imply a {pagbu}. This would lead to (3) which by its turn would
> lead to (1). Then it would be just a case of checking that {la .djan. danlu} and
> the trivial {lo birka cu birka}.
>
> Of course there would be typically many side attempts leading to dead ends,
> but nobody said that artificial intelligence is easy and, not really having any
> experience in the area, my point is that the inference is possible, from very
> general semantic information, and with a pretty simple reasoning.
People like Patrick Hayes and Drew McDermott tried to do that for a loong time
even with logical formulas instead of language(though lojban might be somewhere
between). Forget about it. Human everyday reasoning is not based on symbolic
inference. It just doesn't work. Abduction, analogy reasoning, etc. might work
(at least better than inference), but you're explanation is a very nice example
of the reasoning back in the 1980th.
v4hn
Attachment:
pgpm4Z3d2mppn.pgp
Description: PGP signature