On algorithmic natural language analysis and understanding

摘要

In this article an attempt is made to substantiate that understanding and analyzing natural language texts is not algorithmic, that is to show that there is a natural boundary separating the human from the algorithm supposed to model understanding and syntax analysis. Our main concern will be to discriminate the human from the algorithm with respect to the notion of truth; the analogous discrimination with respect to understanding and analyzing natural language will then be shown reduceable to the first mentioned one.To this end we cannot apply directly the classic results, such for example as Goedel's incompleteness theorem. In fact all these results attempt to show inevitable shortcomings in the formalism's way of handling some important tasks. Now, however, we are confronted with the problem of differences between the human and the formalism. This problem differs from the one set in Hilbert's programme in that now a formalism is wanted which would act, (e.g. in handling the notion of truth) as imperfectly as the human does. We attempt to show such a formalism to be impossible.Our analysis cannot be a mathematical one (and not even a formal deduction in general) as for example it is impossible to obtain formally the conclusion “intelligence is not formalizable” after having started the investigation with formalizing intelligence. This certainly calls down methodological difficulties, which better are avoided, and apparently weakens the discourse. But, those not having what they like have to like what they have. Besides we are not within a deductive science and there is no reason to limit our methods to deductive ones. And, after all, we are in good company: the present investigation seems methodologically related to metaphysical discourses of physics (such, e.g. as Gendanken-experiments, first named but not first used by Einstein) or speculations of astro-physics. Finally, the problem of limitations for Artificial Intelligence, as well as for Computer Science in general, just exists independently from whether you are able to handle it mathematically or not.