No need to calculate meaning and knowledge when they can be extracted from computer memory

    As the great classic Aristotle said, “the known, it turns out, is known to few .

    Linguists all over the world are subject to one pernicious illusion, believing that if they manage to build a “correct syntactic graph” (that is, a “phrase tree”), then in this case they will finally solve this damned problem of machine processing of natural-language texts (eat). So linguists are looking for day and night for some mythical connections and relations between words (semantic units) in sentences and paragraphs of texts. Yes, and cybernetics have connected to these unsuccessful searches. Half a century has passed of such research, and the cart, as they say, is still there. This graph is not built in any way, giving many years of stable 50% errors. Hundreds of millions of dollars have already been spent. The Watson project alone is worth it. But, in principle, these “ties and relations” are not there in reality. All this, if you think carefully, artificial science-fictions,

    Linguists are like those aliens in whose hands the chicken egg, ordinary for earthlings, fell. Here they can study it this way and that. And measure length and breadth, and weigh, and illuminate with an X-ray, and examine the shell for composition, and measure its Mohs hardness, measure its fragility, and calibrate the color. In general, to carry out everything that can be measured, but to understand how it is formed, the poor people will not be able to, because the hens themselves do not know and did not know. In the same way, the natives of the oceanic islands of our planet could not understand how the transistor receiver brought to them by Western civilization could make articulate sounds or sound melodies. And as if they did not study this receiver, did not try it for a tooth or to taste, they would never have understood for themselves that a radio station is also needed for such a device.

    And then what are the scholars of psychology? Why should linguists and cybernetics not unite with them and at the junction of the three disciplines not try to find the desired result. But no! It turns out that brain specialists are also busy studying their “egg product,” that is, gray matter. As soon as they did not measure it and studied it, right down to the very atoms, that is, by the very dendrites and axons. That's even the “neuron-shaped semantic network” in this field invented. Yes, only use in it, as with a goat of milk, since it is also impotent in terms of the possibility of a meaningful perception of it. Or maybe philologists here would say their weighty word? After all, they have something to do with thinking. Unfortunately, they cannot help in any way, since they are passionate about external manifestations of human mental activity (this is their “egg”),

    So what to do? And exactly one thing is to study the procedures of our thinking and try to implement them in the computer plane. Indeed, a person starting to comprehend colloquial speech simply does not know about the existence of morphology and syntax, which he has yet to study in school, and does not at all think about the connections and relations between words, and does not calculate predicates, but simply starts talking himself and understand what others are telling him. This happens with the help of the so-called “Models of Behavior of Images” (MPO), which people remember and accumulate in their memory their whole lives, perceiving the surrounding reality and building in their minds a virtual “Model of the Universe” (MM). It is these MPOs and MMs that help us to decipher the verbal code that encrypts the individual’s thought forms, creating verbal or epistolary language messages. Simply put, people do not understand each other directly with the help of mathematical and statistical processing of “verbal relationships,” but using these same MPOs to extract meaning from a message, as some idea of ​​the author of this message. So the computer should be taught how to extract (recognize) these MPOs in the text and memorize them by entering them into its computer memory. And in order for the computer to learn this craft, in addition, everything still needs to be brought to a formalized form. That is, formalize the procedures for extracting meaning, formalize the knowledge gained, while creating, not a database, as is now done, but a full-fledged “Knowledge Base” (KB), so that there is no primitive Data Mining, but a highly effective “Knowledge Mining” ". Finally, one must learn not to compute meaning and knowledge, but simply getting them out of the knowledge base in the same heuristic-associative manner similar to how all people do it. But computer linguistics just cannot do this.

    It is a pity, of course, that a lot of resources (intellectual, material, temporary) were spent to achieve such an understanding, but it is gratifying that there are already the first enthusiastic researchers who are taking this new unpaved path, getting the first and very promising results.

    Also popular now: