Документ взят из кэша поисковой машины. Адрес оригинального документа : http://www.philol.msu.ru/~lex/melnikov/meln_en/part_3_3-5.htm
Дата изменения: Unknown
Дата индексирования: Fri Feb 28 16:51:05 2014
Кодировка:
Melnikov G.P.- Systemology and Linguistic Aspects of Cybernetics


| << back | | continuation |



3.5 NATURAL LANGUAGE AND MEANINGFUL MAN-MACHINE INTERCOURSE. 

Modelling, Structural Modelling, Sign Modelling.

After defining such concepts as the adaptive system, reflection, the sign, meaning, sense, identification and recognition, it is easy to develop our ideas on the nature of modelling, whose role is in general becoming constantly more important in scientific work, and this is especially true in cybernetics. Without engaging in a wide discussion concerning the possibility of various approaches in defining the concept of the model [32; 71; 137; 174; 188; 193], let us agree to understand by the term model, in its simplest and most obvious form, any object the study of whose properties serves as a means of arriving at conclusions about the relative properties of another object, the original. From the standpoint of our conception of the systemic approach, such an indirect study of another object through the intermediary study of another, i.e. the study of an original on the basis of data on a model, is possible to the extent that these two objects are adaptive systems.

This requirement is necessary - but it is not sufficient. A second essential requirement is the initial likening of the model to the original, i.e. the initial imposition of certain properties of the original on another object playing the role of model. In this case as we saw when analysing reflection processes, those properties of the original which in one way or another are imposed on the model represent a direct primary imprint of the active part of the original as reflected object on the body of the model as reflecting object. Also, under the influence of the primary imprint, i.e. the primary, direct deformation, the process of secondary indirect deformation begins, and the more common initial properties there are with such interacting objects, the more probability there is that the secondary imprint will come to be the pattern (more accurately the pro-pattern in the sense of continuation, prognosis, and anticipation) of a number of the original's properties, in spite of the fact that a primary imprint did not impose these properties on the model. Consequently, the subject, which has "provoked" the start of the primary deformation of the model according to the initial properties of the original which are known to it can nevertheless show, by observing the pro-pattern, i.e. the primary imprint determined by the internal parameters of the model, those properties of the model which can justifiably be interpreted as peculiar to the original, even if the subject did not know for one reason or another about the presence of these new properties in the original, and did not have an opportunity of observing them directly in the original.

It will be clear from the above that a fairly natural basis for the classification of the most obvious class of models examined is consideration of the particular features of those properties of the original which are imposed on the model with primary likening.

If the model is first of all likened substantially to the original and a result of this is the development of intentials and extentials leading to the links network structure between elements of the model being more and more likened to the structure of the original, and the common properties of the model ultimately more and more approaching those of the original, then such a model ought to belong to the subclass, "substantial". A finer division among substantial models can be based on consideration of whether we have

actual identity in the material of the model and the original, or whether this similarity is accurate only as far as the intential, potential, and extential valencies coincide on a limited number of tiers. As an example of the second variety of substantial modelling we can take the study of properties of hydrodynamic objects by means of electric chains. In this latter case people speak of analogue electromodelling.

Since mutual restrictions in adaptive systems impose on each other not only heterogeneous properties (for example, substantial ones on structural and vice-versa), but also homogeneous ones, then properties of the whole are determined in particular by which of the boundary properties of an object are set as initial ones, and which become inevitable consequences of the first. This correlation is justified when the qualitative properties of elements bearing primary boundary properties vary extremely widely; this is very convenient for the modelling of properties of various objects by means of models whose restructuring, their initial likening to the properties of the originals, is effected by combining the links structure of a limited set of elements of the model and the change of boundary properties, e.g. the form and composition of unoccupied valencies of these elements.

For example, by such a method it is possible to show with a model many properties of the original, if it is a question of variations in the positioning of apparatus within the boundaries of a particular location.

All the types of models examined can be called external since they are, in relation to the modelling subject, external objects. Among external models we have examined two types, one of which can be regarded as substantial, the other as structural.

Now it is easy to imagine that if external structural modelling is regularly carried out by the subject, and the class of originals being modelled is more or less homogeneous as are the aims of research into their properties, then sooner or later standard abstract patterns of both elements of models being employed and methods of combining these elements must be worked out by the subject.

Consequently structural modelling and the revelation of unknown properties of the original will be able to be carried out in time by the subject (in reasonably standard cases) without the aid of a corresponding structural model, on the basis of just thought patterns of boundary properties of these models' elements, and on the basis of schemes for combining the elements.

Such abstract thought units, which are specifically for the structural modelling of objects' properties of a particular type, represent constructs, as units of a certain calculation, or some formal theory.

Constructs, if used by a group of people to carry out structural modelling of objects in a certain area of research, naturally are usual and social thinking units. If they are examined from the point of view of communication needs, the sign expression of such abstractions in the role of communications' senses is most easily accomplished by means of usual signs, where the motivation of the signs' properties by properties of these usual senses is not essential.

Consequently among the members of a group of people using the same constructs for structural modelling intercourse is easily achieved in order to transfer content obtained as a result of structural modelling by means of constructs of a given calculation. The process of structural construct modelling itself, the initial conditions etc, are just as easily expressed through these signs. Those speech ( or textual) works which express named contents by means of conditional usual signs represent sign models.

The most widespread type of structural modelling is indeed sign modelling, and the most universal sign models are worked out by a special science, mathematics, although sign structural modelling is possible which is not universal but narrowly specialised, with its set of special constructs and signs for expressing them. [117-125]. 

Structuralism, Systemic Approach, and Generative Grammars.

Since in order to model a preliminary transfer of initial known properties of the original is essential, in the first formation stages of any science, when only little data has been accumulated about the properties of objects studied by it, the modelling method is not applicable in practice. The researcher's main concern is the properties of the whole, then the properties of the parts etc, and his purpose is to explain the properties of the whole through the properties of the parts, i.e. the substance. This can be done successfully until a certain stage but sooner or later it is revealed that new knowledge about substance properties has less and less explanatory force and the reason for this is the lack of attention paid to the combination and interaction methods of their parts, i.e. to the structure of the object.

After this researchers switch over to a search for the laws of linking between the structural features and properties of the whole: this opens up the possibility of using structural modelling in general and sign models in particular in order to describe object properties and to explain these properties. We call this phase of development structural. Among specific sciences physics become structural first of all, it seems, after Galileo's work when he found methods of describing the most important physical phenomena and objects by their structural characteristics, expressed through special sign systems. [103; 119; 120].

With the successes shown by methods of structural sign modelling in the first stages of their use the impression is gaining ground with the representatives of the corresponding scientific discipline that taking account of substance characteristics of an object under study in order to explain its properties is in general non-essential and that the object's elements must be regarded not as bearers of certain immanent properties, but exclusively as pure significances, like intersection bundles (or bunches) of links and relations in the structure of the whole. Consequently if at first structural methodology replaces "elementarist" nations by seeing beyond the divisibility of an object its wholeness within a links network between elements (and in this sense the structural approach to an object is more systemic than substantial), then, when an awareness of the importance of taking into account an object's structural characteristics develops so that the importance of the substantial characteristics is ignored, research methodology has moved to the stage of structuralism, which is just as non-systemic as pure substantialism, although at the opposite pole of "nonsystemicness" [48; 73; 83; 84; 89; 91; 96; 99; 100; 102; 103; 121].

It should be noticed that the concept "structuralism" has only been developed in science in the twentieth century (in spite of the fact that it has existed as a phenomenon since the beginning of the seventeenth century) first of all among linguists when humanitarian disciplines developed from the stage of substantialism to the stage of awareness of the importance of studying the structure of their objects by using structural models as a research instrument. Certainly in modern methodological literature structuralism is mostly identified with the systemic approach [23] and as a result of an object's essence is identified with its structure, this is supported in fact by the proposition concerning the drawing together of an object's substance, i.e. its elements and parts, into bundles of "pure" relations, or "pure functions". Structuralists deprive themselves through this of the possibility of taking account of interactions between the properties of material and the structure of a whole, and are obliged to regard material as absolutely amorphous and pliable. Therefore structuralists do not raise the "systemicness" of their general conception any higher than the tomists' philosophy.

However it must be noted that if an object of structuralist research is a deeply adapted system and we are not interested either in the prehistory of the formation of this system or in a prognosis of its future changes, i.e. we are only looking at a synchronic section of a system, then, on the strength of deep agreement between structural and substantial characteristics of the system we can also interpret the substance properties of its elements through the particular features of the significances of these elements, and through the features of that links junction in which this element adapted in order to carry out certain functions in the system, i.e. through the particular features of the structural parameters only. It is only the causes and tendencies of development which we cannot study by purely structural methods; for this proper systemic notions and concepts are necessary. Both structural and substantial methods prove to be absolutely essential, by supplementing each other, if of course the researcher understands systemic methodology and knows the laws of interaction and interinfluence of the structural and substantial parameters of objects during their adaptation in order to carry out a particular function in a certain meta-system.

And so, if the object under investigation is a deeply adapted system and is examined only in the synchronic plan, then on the strength of the accord of its substantial characteristics with structural ones, the most effective approach is the study of all its properties by revealing the features of the structure. Then the possibility arises of using structural and sign modelling as the main means of studying the object, that is, if we are not interested in the cause of the rise of the properties discovered and forecasts of their change.

In particular, structural methods of studying natural language are extremely fruitful, but only that side of it which satisfies the requirements given: it can be regarded as deeply adapted owing to a lengthy preceding evolution and adaptation and relatively invariable in the synchronic section examined. From everything examined in the speech communication scheme it follows that such a part of natural language and linguistic thinking in general embraces all those units which are usual, i.e. meanings, gestalts for identifying and reproducing speech units, settled combinations of signifying units, e.g. words, obligatory schemes of their external agreement in the speech current, and also reproduced, i.e. usual, senses, taking the role of intermediaries in acts' of naming occasional senses.

At the same time it is clear that everything that is a single event, irrepeatable, unique, and occasional is not supported by strict, reproduced contiguity links, but by suppositional resemblance associations. It cannot turn out to be usual, repetitive, or deeply adapted, it does not lend itself to study by structural methods only, and it is inexpressible through constructs of any calculation and therefore only to be understood meaningfully, on the basis of systemic notions. Among such "non-calculable" processes even simply within the framework of linguistic thinking are included acts of selecting meanings and usual senses for hinting at occasional senses and, correspondingly vice-versa, acts of guessing from received signs and meanings (linked with them usually) about the occasional sense which was in mind when the speech section formed. Processes of communication formation are not "subject to" structural sign modelling insofar as they are based on hypotheses of the speaker about the listener's knowledge and on his taking into account all the contextual and situative conditions when choosing those senses which require linguistic hinting and those which are obvious, without that, to the receiver.

It is true that the attentive reader will have already noted that between linguistic and speech units on the one hand and units of sign modelling on the other a very deep isomorphism is traced, in the light of which the contrast of natural language and formal calculations of structural modelling can appear unnecessarily sharp. Indeed in both cases we have external (speech or textual) signs or symbols, internal signs - gestalts - for reproducing and identifying external signs, and also elements of abstract usual content - the meaning of language or constructs of a sign model. With the specific use of linguistic signs the occasional establishment of similarity between sense and meaning is required, and when the sign model is being used interpretation is required of notions about the object under study through constructs of the selected sign model, i.e. the once again occasional establishment of similarity between sense in the act of modelling and meaning in the appearance of the model construct.

But after the explication of this isomorphism the distinction is clearly revealed between sign modelling and sign communication. With modelling the process of choosing the model and the process of interpreting the sense in the model constructs is recognised as the creative part of the researcher's work, and nobody claims that a process of formalisation will ever be formalised [5; 21; 34; 42; 62; 70; 65; 66; 140; 153; 154; 161; 162; 169].

Speech intercourse is an unbroken formalisation and interpretation of senses through meanings, and of meanings through senses. Nevertheless the basic majority of mathematical linguists and cyberneticists are convinced that it is possible to create a formal description of the intercourse process in natural language and consequently the creative act of transfer from content to the formal can be reduced to a formal routine operation [57; 133; 173; 184-187; 197; 211].

It is clear that one can describe formally in speech communication the simplest part, with the minimum dependence on content, of the formation of already creatively selected linguistic units into a syntactically co-ordinated whole : the sentence. But the choice of units and even their syntax during intercourse depends above all on the particular features of occasional senses, while syntax in sign models is set fully by the properties of meanings, i.e. constructs.

In the light of what has been said it is reasonably clear that the various generative grammars which are popular among cyberneticists and whose methods of creation was suggested about twenty years ago by N. Chomsky, are extremely effective when describing just these external, non-sense communicative processes with a fixed and, consequently, only a usual, set of meaningful or content units. But at the same time these grammars do not have any relation to the actual process of communication in natural language. Insofar as the chain-link with the hinted - hinting relation, without which circuit intercourse is impossible, cannot be conceived of in principle in generative grammars; also the influence of occasional senses even on syntax is not reflected, as there simply are no occasional senses in the scheme. Let us give a simple example to illustrate this proposition.

In ancient Rus (as Russia was named in the Lievan epoch) educated people would remind certain erudite figures of the proverb taken from Greek, "Omniscience is not wisdom". From the point of view of the laws of generative grammar this is a perfectly correct sentence.

Simple people who did not know Greek proverbs themselves arrived at the same truth, but clothed it in different form: "he knows much, but understands little". It is extremely doubtful whether among all the variants of the first phrase generative grammar would yield the second. But we can certainly exclude the possibility that on the basis of generative rules a phrase will be formed which the companion of the writer and traveller Arsenyev, Dersu Usala, uttered: "know - much, understand - no". He did not know Russian well, but characterised an erudite person no less to the point than the conoisseurs of Greek literature.

Obviously it is far more valuable for cybernetics to create machines which could arrive at the truth, and if initially they express it like Dersu Usala or even worse, intercourse between man and machine in natural, though broken, language will prove to be possible. The communication mechanisms we have examined bring us nearer to working out the methods of such intercourse, while the grammatically correct phrases generated by Chomsky's grammar do not even give us a hook with which to attach speech signs to occasional content. Even colleagues of Chomsky are beginning to understand this clearly although ways of overcoming the semantic limitations of generative grammars continue to be sought in the construction of new formalisms [70]; and in the "speaking machine" scheme there have not been up to the present time any channels of interaction.

Let us now examine a number of real problems of cybernetics, for which the lack of clear concepts concerning the nature of linguistic categories and their relation to content have up to this point been a serious hindrance to their being successfully resolved.

Ways of effecting machine translation in practice. If we start with the quite detailed ontological scheme we examined of language as a psychic, usual mechanism of occasional communication, it is clear that with machine translation of texts from one language to another, if the machine is not capable of thinking and having its own life experience there is in general no semantic plan. The most that a machine can do is find from identified symbols of a text in one language patterns for recreating symbols of a text in another language, while the move from symbols to content can only be carried out in a person's head. The rules for choosing output symbols from input symbols can be extremely primitive; a symbol is set to correspond with a symbol ("word by word" translatio), but they can include both complex instructions from the presence and links of many symbols from the input text and the already found symbols of the output text before a regular symbol of the output text will be chosen.

What have the efforts of specialists in machine translation been directed towards since the start of work on this problem right up to the present time? - Towards the point where the sequence of symbols of the output text are as if the input text had been interpreted by a human interpretor, and not a machine.

But an interpretor, having received a text, first of all understands it, i.e. he identifies meanings sequentially from its symbols, and from the usual senses he identifies occasional senses.

After this the interpretor is justified in forgetting how he "obtained" the occasional senses. For him they represent that sense intent (or plan), which must be expressed through the means of the output language [55; 56].

For this the interpretor must place understood occasional senses, where possible, under the usual senses of another language and through so doing determine the group or set of meanings and linguistic signs of the other language for these usual senses.

For those occasional senses which are not under the usual senses he must select the necessary meanings and, through them, linguistic signs, guided directly by the resemblance between meanings and fragments of occasional sense and also an idea of what associations can be aroused by these meanings in the consciousness of a reader of the translated text. Then the interpretor can already progress to speech signs both for usual and for a second group of occasional senses [86].

But since modern computers do not have an occasional and usual senses plan, nor a meaning plan, all the procedures we have examined cannot be put into effect in modern electronic computers, and carrying out a translation orientated towards the result described above is not possible in principle. Those external similarities of a translation which are achieved these days no more resemble a real translation than a toy dog resembles a live one, and the "perspective" of content transfer has to change only slightly for the insurmountable nature of many difficulties to become evident, difficulties which a living person does not even regard as such during the act of translation.

So is there a way out of this situation? In the light of all that has been discussed it is clear that there is, indeed more than one. The first is the most reliable, but the most difficult, capable of giving the required result only very slowly (many consider it to be in general non-realisable), - this is the construction of a thinking computer, which, being able to think, will be able to both speak and translate.

Since real thinking in modern digital and analogue computers has so far not been successfully effected (in particular because thought is identified with communication), we will leave this first way to the future.

Let us take a second way: extract all the information which is contained in the text of the input language, and present that and only that (because all other information is drawn from the life-experience of the interpretor, and a computer does not have this) in output language symbols, i.e. describe it in the output language.

It is clear that then the text will not resemble the one which an interpreter could give, but then on the other hand the presence of all the textual information simplied by the author of the initial text allows the reader of the translated text to recreate very fully from it the sense of the transmitted initial text.

But how could this be done in practice? What sort of information is presented by the text, and why is it not that sense which the creator of the text had envisaged?

Let us look at this in a little more detail.

A speaker selects the minimum number of meanings from which the implied, hinted, sense can be obtained.

Sense in our scheme consists of patterns of actual or imagined phenomena; meanings are also patterns, but generalised, combined ones; they are patterns of classes of senses which could be suitable divided only when there is a certain method of communication, and grammatical structure. In the other language the breaking-down into communicative classes occurs in a different way and therefore the group of meanings is different in the other language, as communicatively significant generalised patterns..

The transmitter has selected meanings so that a receiver guesses the sense from them. But the transmitter has here presumed that the receiver's set of meanings is the same as that of the author of the text himself.

But this hypothesis is not true if the receiver does not know the transmitter's language, and we must correct this mistake where possible with a translating-machine. Since it is a question of the meanings involved, i.e. usual and social units, to a high degree similar for all members of the linguistic collective (it is not a question of occasional senses which are always unique), then a basis exists for discussion of automation. Since a transmitter must above all stimulate by means of speech signs only a group of fully determined meanings in the receiver's conscious -ness, and, as we saw earlier, further transfer processes to the sense occur in the consciousness of the receiver himself on the basis of the laws of similarity and association of patterns, then the computer must be capable of helping to stimulate in the receiver's consciousness, if he knows the speaker's language, those very patterns which would be stimulated as meanings in the consciousness of the receiver who knows the speaker's language.

In other words patterns identical to those which are meanings for a receiver speaking the same language must first of all become a foreign language speaker's senses. Only after this will the different language reader be able to begin the transfer process from the hint to the hinted, i.e. to occasional senses of the text being translated, according to intelligent meanings of the foreign language, as hints of the occasional sense in the mind of the author of the text.

It is easy to see that such a transfer from the foreign text to its content by means of modern computers can indeed be effected.

Let us examine the principles involved in carrying out this process in a little more detail.

The principles of translation of natural language texts by means of an information computer.

Any natural language is a universal means of expressing content. Consequently, if on the basis of linguistic analysis we reveal the non-contextual meanings of elements of a certain language which is alien to us (for example English) we will be able to describe each such meaning by means of Russian words. In the most favourable circumstances the description will also consist of one word (although there occur few such happy cases). For example it can be shown that the non-contextual Russian, "Da", and English "yes" do not differ in meaning.

If there is no coincidence not just one word but several, possibly many, will be required in order to describe the meanings of each morpheme and each signified unit.

This description is made by people on the basis of a deep study of both languages. But it is important here that as soon as a description has been compiled, it can already be introduced into the computer without any content: the computer will simply remember that it is necessary to produce a certain symbol (or chain of symbols) of the output language for certain symbol of the input language.

Consequently if a text being translated is now entered in, a text of symbols and chains will appear at output which are received and interpreted by the representative of other language. He will first of all identify from them, as we have already noted, through the meanings of symbols of his native language, those senses which coincide with meanings, which would arise in the mind of the receiving person, if he knew the language of the transmitting person and read the input text directly.

Consequently the computer has helped to produce the same psychic effect on the receiving foreign language speaker (to be sure after many stages in the receiver's consciousness), as on the receiver who knows the input language and can read the input text. Is the reception the same in both cases?

In a number of characteristics such reception and interpretation of the other language meanings through senses gained by means of symbols of the native language differs in principle from a direct progression from symbols to the meanings required.

First, as has already been stated, a person knowing the input language comes immediately from the symbols of the text to his own meanings of this language, ready for the act of reading. The foreign language receiver identifies from symbols of the output text meanings first of all of his language, and from them he resurrects as sense and not meanings (i.e. he constructs occasionally) those patterns which are meanings for the person having the input language -meanings which are ready, usual communicative units.

Consequently, having progressed from his meanings to senses, the foreign language receiver has not advanced his understanding of the text further than the inout-language speaker who has had the symbols of the text and has only taken up meanings.

Only after this can both receivers progess to the following stage of comprehension - the stage where the sense is reestablished, implied by the transmitter in the input language on the basis of knowledge of those meanings which the transmitter expressed through symbols of the text.

But this is not the only difference.

The set of those hints received in meanings by the person reading the input text as in his native language is itself such that the reader already has the skill to supply precisely through such hints, through such a "perspective" of seeing the intent or plan, generalised patterns of meanings before the patterns of full concrete sense.

When an analagous act of hint patterns is obtained by the foreign language speaker by means of computer translation, much can seem strange to him in these hints: as a rule, what is not usually afforded attention in the native language is emphasised in them (far example, whether the writer was or was not an eye-witness of the described event, as is the case in many languages of the world) and at the same time what must be mentioned, it would seem, quite definitely (for example, indicating the time of an action to a Russian) can be absent among the hints. But since in the total number of hints present there is sufficient to reestablish sense, the foreign language receiver will also understand everything, but initially the understanding process will take up a lot more of his time than the process of understanding a text in his native language.

It is clear that with computer translation an output text will reflect many features of the particular character of the "linguistic vision" of the author of the input text. Therefore it is necessary to acquire the skill of understanding machine translation for each of the output languages (as we become used to the manner of expression in various fields of science) [80].

But then everything which is not reflected in input language meanings only because it is represented by other elements of the text. will be fully preserved in other elements of the text in the output language. The only thing that can be lost is Information based on an association with the external aspect (or sound) of the symbols themselves of the input text.

Thus, in particular, such a computer will give in translation the "between-the-hints" meaning of a poem. but not the rhythm, nor can it give its rhythmic characteristics. A machine translation will however preserve all the implications, puns, and figurative senses, if they are based only on meanings.

The question of the relation of usual sense to meanings in the input language will not be examined here in detail. But it is clear that in general features, since both meanings and senses are usual, meanings must quite definitely be preserved in translation, and exactly as a reader would decide and re-establish from them in his native input language which of the several usual senses is in mind, the receiver will be able to choose the required usual sense of this meaning in the output language from the meaning understood.

The only difference in interpretation here is this : the person with a command of the input language, reading the input text, already knows beforehand the set of usual senses of each meaning, while the reader of the foreign language text cannot know this set of usual senses.

Consequently all usual senses for a given meaning must be given out by a translating computer in the form of a list of descriptions of these senses, while the person interpreting the text has to himself choose the most suitable sense from this list.

If, though, the occasional sense of a symbol does not coincide with any of the usual ones and it is possible to guess at it only by taking into account symbols of the context already introduced, and the content understood on the basis of it, then in this case both the person reading the text in his native language and the person trying to understand it after it has been translated by computer, will be in the same position.

Consequently any metaphoric usages will be preserved in the text in the output language, as will the play of precise synonymous substitutions, the basis for antonyms etc.

In this way such a method of using the computer will be effective for translating literary texts.

If texts which have the same type of content prevail, for example, narrow specialised ones, standard items will be found in them which must be translated regularly in the same manner. Consequently such standard translations of standard items can also be usefully fed into the computer. Then this translation item can be edited so that it is expressed in the style of the output language and not the input language. This edited variant can then be delivered at output.

But such a translation method can only be auxiliary (in usual machine translation "ideology" it is the only one), because everything standard has only static predominance, and elements of a standard item of a narrowly specialised text can sometimes also be used "non-professionally". In these cases it is necessary to return again to an un-edited variant, i.e. to the translation of meanings and not specialised senses in order to retain the possibility of guessing the non-standard occasional sense of the elements of an item.

Consequently the whole process of translation by means of a computer must be conducted in this case within the procedure of dialogue: if something appears incomprehensible to a person he demands from the machine a corresponding word or morpheme, and more detailed information, which would help him to understand the occasional sense.

The principle of automated translation we have examined lends us to study the structure and semantics of language in some detail. [111]

Moreover, to whatever extent thought elements are eventually fed into a computer certain of the simplest guessing procedures (with the possibility of the receiver of the translated text controlling them) can also be fed into the computer, and then it will change from being an information computer to being a translation computer. The author traced the methods by which such change was effected in 1959. [86]. But the main thing in the principle of computer translation proposed is that the possibility does indeed exist of extracting from a modern computer everything for the creative processes employed by a person when interpreting a text written in an "alien style" in his native language to be based on that maximum amount of information on content reflected in the input text.

It should be noted that when the semantics of a language is being detailed in order to build an information computer much useful information can be obtained from a so-called "sense-text model".

Let us discuss this separately.

Language to describe senses. The works of Yr. D. Apresyan, A.K. Zholkovsky and others [133; 57; 15] will be discussed.

It must be emphasised straightaway that we shall not concern outselves with the details of essential differences between the statements of the authors mentioned concerning the purpose and meaning of their works and the interpretation of the content of their works offered here. For us what is important is the role of these works with regard to the linguistic problems of cybernetics, which we are examining from the standpoint of the concept of the systems approach expounded above and the occasional communication scheme. and also with regard to the use of results in the works mentioned in effecting the described methods of translating by computer.

Since meanings and certain senses are usualised and socialised, i.e. they have been made mutually similar for the basic group of members of the linguistic collective, it is possible in principle to make up a calculation whose initial elements are psychic construct-units of a high level of abstraction i.e. they represent patterns with a small number of internal features. It is possible to gather from such elements, as from the parts of a children's construction set, more complex formations and in particular ones which will coincide reasonably accurately with patterns that are meanings of linguistic signs and are usual senses expressed through these signs. Consequently if special symbols are made to accord with these elements, as constructs and rules for combining them, each meaning and each usual sense will acquire the chance of being described or detailed through combinations of these symbols.

Consequently we will obtain a formal model for calculating usualised units of the content plan of a certain (within that group) natural language. Such a computation is called by some writers (not absolutely successfully as we see it) a "language to describe senses".

From what has been said previously it is clear what the distinction in principle is between the relation of meanings of linguistic signs to senses, for example, usual ones, and the relation of units of a "language of senses" to meanings and these same usual senses which are what they signify.

Constructs of a "language of senses" must always "cover", without anything left out, the meaning of linguistic signs presented by them, or usual senses of natural language, while natural language meanings "cover" only "obvious" parts of usual senses designated by them, and only the minimum necessary number of hints of senses, which is the main reason for the polysemy of linguistic signs.

The question of the technology of describing meanings and usual senses through the meanings of a "language of senses" does not interest us directly. It is clear that it can be based on "integral" principles, i.e. on the enumeration of those components which the described sense can be expanded into, and also on "differencial principles", i.e. the primary or main indication of how one usual sense differs from the remainder (this second tendency dominates in the works of Yu.D. Apresyan and A.J. Zholkovsky and others). Something else is more important for us: to what extent "language", created above all to describe usual senses, is capable of reflecting an actual text's structure determined by links between components of actual sense, i.e. of a plan or intent arising situatively and occasionally.

If this whole total sum of usual sense units of natural language is able to be expressed through a "language of senses" and correspondingly vocabularies are composed for the transfer from natural language units to their "sense description", it would seem that the transfer from actual units to units of sense recording can occur automatically. And since all sense units of a text will be explicated with this, then the sense structure of a text will also prove to be clearly expressed; after this it is easy to reveal the actual sense structure by means of a computer. It seems the creators of a "language of senses" for describing vocabulary units of natural language thus envisage the possibility of revealing the sense structure of a text.

However if we introduce our distinction between the concepts of meaning, usual and occasional sense, it soon becomes clear to us that this path of semantic investigation is not able to provide automatic recognition of the structure of a linked text, if the computer carries out only formalised operations.

Let us assume that it has been possible to make an initial set of meanings of a "language of senses" so that through it it is possible to express explicitly with reasonable accuracy all the usual semantic components of natural language. It is easy to show that even in this case several equivalent precise units, "lekemes" in the "senses language", will correspond to many speech units of natural language (since they represent meaning first of all, and through meaning some particular set of usual senses). The choice of that one which is part of the plan or intent of the speaker or writer, involves turning to the context and the situation, and consequently not to usual, but to occasional senses (of which there is an infinite number); formal rules for the transfer of these occasional senses to units of "senses language" cannot in principle be found.

But let us presume that even this problem has been solved by a direct or indirect method. Shall we obtain a correct reflection of a text's sense structure if we transfer all the linguistic units of a natural text into "senses language".

It follows from our speech operation scheme that in this case the whole "word" polysemy will be removed, i.e., true, implied, usual senses of the initial text will obtain single, and clear, expression in symbols of "senses language".

However, as we know, naming usual senses in an actual communication act serves only as a means of indicating (directly or indirectly) an occasional sense which as a rule is far more concrete than a usual one and is a pattern representative of external (or imagined) objects in all (or almost all) their irrepeatable individuality. If, though, the machine does not have this plan of directly reflecting external objects, then it also does not have occasional senses. But it is the link between the occasional senses which sets the structure of a linked text, while all the formal indications of linkage serve only as a partial and very approximate manifestation of this actual structure of the text.

So we arrive again at the conclusion that it is necessary, in order to establish a text's content, to "interpret" or "decipher" the creative methods used by the author of a text when choosing ways to express his meaning, and therefore it is necessary either to consciously take the part of a person in the act of revealing actual structure of a text, as in our translation scheme, by means of an information machine, or to look for ways of carrying out creative operations on electronic computers: for this one must straightaway go beyond the boundaries of "pure linguistics" and to form a theory on the psychic bases of not only communicative activity but also thought in general (i.e. "extra-linguistic thought") as Baudonin de Courtenay called it when comparing it to "linguistic thought" [24, vol.11, p.11; 177; 166; 333].

Consequently, "senses language" is found to be an effective means of describing only the usualised part of the semantics of the language permitting mutual correlating (terming), i.e. only those semantics which are reflected in a detailed explanatory dictionary, and not ones which are represented by linguistic units in specific contexts or in acts of meaningful intercourse. Therefore data which is obtained within the framework "sense text" can be extremely useful for creating an information machine, but not a translating computer.

However, it is not possible to conclude from this that a description or detailing of the usual semantics of a language by formal computations has any relation with the study, and the carrying out by machine, of meaningful intercourse in natural language. As is clear from our occasional communication scheme, the transmission of irrepeatable occasional sense through language is only possible when there is reflection for the communicants, the capacity to represent alien ideas, i.e. the knowledge and ability of the other person in the intercourse.

Consequently, if in order to choose an occasional sense which is informative for the receiver it is necessary to know what is, and is not, familiar to him in that area of content which is under discussion, then it is essential, in order to choose the means for expressing this content, to have the backing of fully determined assumptions about how usual, "termined" semantic units are retained in the receiver's memory. But this means that all usual semantics are subject to "realisation", especially if it is a question of recreating acts of meaningful communication by cybernetic machines. For such realisation, and for a description of the results of this awareness, the new innovative work of the creators of the "senses language" is of great importance.

We must also dwell on the problem of the possibility in principle of meaningful intercourse between man and machine as one of the most important problems of modern cybernetics [145]. But before so doing, we must define what should be understood by deductable content which can become the occasional sense of communications.

Conditions of the formal deductability of senses. Having examined the psychic nature of basic linguistic units and the mechanisms not for only the naming ( the nomination) of any content unit by means of a limited set of linguistic signs but also for predication (transmitting elements of the new knowledge), we are now no longer confusing linguistic semantics (i.e. the level of meanings of linguistic signs) with the semantics of the thought content being transmitted (i.e. with the level of occasional and usual senses involved together with units of meaning in hinting or implication relations). These ideas on the nature of communication in natural language have enabled us to conclude that it is impossible in principle to construct purely mathematical models of speech activity, but at the same time they have indicated ways of increasing the efficiency of man-machine intercourse in those particular cases when the hint link in the communicative cycle is finally reduced to particular cases when the -hint link in the communicative cycle is finally reduced to usual contiguity associations. Such reduction is possible when texts, offered to, and generated by, a machine can only be interpreted in usual and consequently accountable senses and in an enumerated composition of transformations of certain senses into others. We shall now dwell in more detail on the question of how great the number or share of such semantic conditions - favourable for machine operation - is, and we shall be primarily interested in the semantics of Economics texts, in that they offer a basic body of information in modern automatic control systems.

Having contrasted sense content with linguistic content, and linguistic content to those speech or text symbols which indicate only linguistic content, we are justified in thinking about sense content, disgressing if necessary from the methods of expressing it textually. Bearing in mind what has been said let us now again turn our attention to one essential feature of sense content in man's psyche, which is important also for the problem of man-machine intercourse the "generative capacities" of senses.

Being a reflection of the properties of phenomena of actual reality, content elements - senses - permit interactions and transformations, and a result of this is new elements of knowledge, new senses, which no longer relate to the number of previously observed phenomena which have developed as a result of reflection, but nevertheless are correlatable with actual phenomena and precisely because of this are of value for the subject and are used by him together with direct reflections. In other words senses can serve as a basis for obtaining deductible knowledge, and the more rarely deductible knowledge comes into conflict with empirically obtained knowledge, the more effective is the thought activity itself. Therefore it is a natural wish to estimate how much this property of the deductability of one piece of knowledge from another, and of certain senses from others is preserved if we intend to have a plan of sense content in cybernetic machines as well.

Let us imagine for a start that new senses are obtained from existing elementary senses with logical operations, for example, in the boundaries of Boolean algebra. We shall take as elementary senses those of simple utterances, the one requirement for which, if we start with statements expounded in books on mathematical logic, rests in the obviousness of the relating of the content expressed by them to the meaning "truth" or "falsity", while deductible knowledge we obtain by means of a particular logical operation, for example, implication.

Almost any textbook or monograph on mathematical logic contains such assertions: the truthful meaning of a complex utterance depends only on whether truth or falsity are ascribed (in a table of truthfulness) to the combination, under examination, of truthful meanings of elementary utterances. In particular, in an implication table we find the meaning "false", if the first elementary utterance has the meaning truth and the second has the meaning "false". In the remaining cases however the complex utterance is true, and therefore an assertion of this type must be regarded as true; "it follows from the statement that "an atom of oxygen is lighter than an atom of hydrogen" that "a dolphin is a fish" must be regarded as true.

Examples like this are constantly relished in logics literature, but if we construct a machine to carry out logical operations, there will be no difficulty encountered in automatically obtaining such "deductible knowledge"; however at the same time it is difficult to imagine what benefit a person can obtain from such an act of communication in intercourse with the machine, and what benefit the machine can obtain. Moreover doubts can arise: does what is called a conclusion in mathematical logic have any relation to those logical processes on the basis of which the thinking subject obtains deductible knowledge? If it does not, it is not then clear from what angle to approach the problem of realising in a machine sense contrasts corresponding precisely to those in the human psyche.

In answer to these questions we can first of all direct our attention to certain widespread misunderstandings of correlations between elementary and complex logical utterances, after which the conditions will become clearer where deductible knowledge, obtained even within the framework of Boolean algebra, ceases to be strange and paradoxical and does not contradict our more intuitive ideas concerning the link between deductible knowledge and initial knowledge. Only after this can we continue our analysis of the conditions which are necessary to provide the actual sense content of logical transformations even if these transformations are effected purely formally in the memory of an electronic computer.

First of all let us turn our attention to a certain fact from the history of mathematical logic, which is fundamental to the understanding of its nature but is not, for some reason, taken into account by contemporary logicians when they give "explanatory" examples of the determination of the truthfulness of a complex utterance from elementary ones. Among the few mathematicians acknowledging the basic nature of this fact is A.S. Kuzichev. In his book "Venn's Diagrams" he writes that about a hundred years ago John Venn arrived at the conclusion that "the laws of formal logic do not operate automatically: if notice is not taken of the underlying choice of the universum and the corresponding increased accuracy of concepts they lose their sense". In other words it is necessary in order to form a complex utterance from elementary ones, to establish exactly that they are elements of the same universum; only then are they comparable in sense and a derivative complex utterance also acquires sense. Otherwise the result is the same as in the popular joke: "In the vegetable garden there is an elder tree - and in Kiev there is an uncle".

But how do we understand today the word "universum" (universal set)? This is how the American mathematician A.T. Berztice defined it [20, p.17]; "If all sets examined in the framework of a certain situation, or reasoning or argument, are subsets of a certain set U, then the latter is a universal set (or universum) for the given reasoning or argument".

Thus is follows from this definition that the one fact that the elementary utterances examined take on only the meanings "true" or "false" is not sufficient for the process of forming a complex utterance from a simple one to have any sense or point. These utterances should present elements of the same "situation or argument" as their sense.

If these elements are sets they should be sub-sets of the same higher level set.

This idea can be expressed more exactly through the Boolean concept of C (M), i.e. the set of all the sub-sets of the initial set M; elementary utterances can enter into logical relationships and produce complex utterances not without meaning, if the senses of the elementary utterances are members of the same Boolean expression; here the sense of a component utterance will belong to this very Boolean expression and consequently will have a relation to the problem being solved.

Thus, if we want to carry out even some purely formal logical operations in the computer we must be concerned that the symbols of the elementary utterances should correspond to the sense belonging to the one set of all the sub-sets of the initial set of senses, i.e. to one Boolean expression of senses. Otherwise derivative complex utterances will be senseless and consequently the act of automating logical operations will itself lose practical sense.

Prime sources of the compatibility of Senses of Elementary Utterances.

This question now arises: in what way can it be established that senses of elementary utterances belong to a "certain situation or argument", and, consequently, how can one be certain that if these senses have an external expression, for example, in the form of utterances, we are justified in regarding these utterances as symbols of members of the one Boolean expression?

As was noted in one work [117, p.12], the success of the formal analysis of complex phenomena depends first of all on how successfully (whether intuitively or from theoretical presuppositions is not important) the *Indicators have been chosen. For example, the disclosure of the complex structure of animals' reflex reactions only became possible because the physiologist I.P. Pavlov chose a very effective Indicator-the famous saliva gland.

*The author uses a different word for Indicator at this point, one that has a more concrete feel to it. His use of this word will be shown henceforth through an initial capital I in the English.

Naturally a complex problem implies the utilisation of a large number of Indicators but they must be non-independent and combined in hierarchic fashion: a certain "main" indicatory sign reflects the fact that the components examined belong to the one set while other Indicators testify to the division of this set of components into sub-sets, each of which in their turn can be further broken down into sub-subsets, to which indications of the Indicators of the next deeper layer bear testimony etc. But however many levels of examination of a situation are needed to solve a set problem, reflections of its components in the form of utterances, models etc., ultimately reduce the semantics of these utterances and models to an affirmation of the presence of certain physical properties revealed by the Indicator.

In particular, if it is a question of thought, sense semantics, they are a fixture (in the nerves of the brain) of the indications of natural Indicators, the sensory organs.

Here the signals from various sense organs can themselves have a natural physical character, for example, electric impulses, whose frequency is proportional to the intensity of that physical property revealed through each of the sense organs. However by being reflected in the memory these signals must preserve imprints not just of the revealed intensity, but also some particular signs of precisely what organ or of what receptor, the fixed signals are an indicator (we observed this condition in our symbolic notation of reflection processes).

If when a computer's semantic field is being constructed it is to be supplied with receptors, the preservation "of the name Indicator", when information has entered the computer from it, is consequently an essential condition for determining the sense comparability of elements of the knowledge of such a computer about the external environment and, accordingly, the capacity to obtain knowledge about it by means of a logical conclusion.

And finally if a modern computer is envisaged and we intend to solve semantic tasks by means of it, then, as when a formal theory is constructed, such an undertaking will be successful only to whatever extent there will be for the symbols of the theory elements of content compatible in sense relation, even if these elements are placed not in a computer's memory, but in the consciousness of users.

The fullest representation of this principle is found in the Kuznetsov's systemic concept. Here is what he writes on this in one of his recent works [75, p.222]: "In mathematics each term introduced is preceded by a so-called quantum of existence, which makes the use of the corresponding term legitimate. In a real situation the role of quantum of existence is placed on measuring equipment. We can speak of "experimental space", where the number of axes corresponds to the number of scales of the measuring equipment".

It is clear that the need for a set of all the senses of elementary utterances to belong to one Boolean expression is a particular expression of the general principle of the sense of any term or expression of a theory belonging to one and the same "experimental space", while the mutual comparability of various theories is only possible when their "experimental spaces" are "sub-spaces" of a higher-level space.

The more fully the properties of various objects and situations can be reflected through combinations of indications of the one set of measuring equipment, the more probability there is of reducing particular "experimental spaces" to a single, generalising "experimental space", and the more complete deductible knowledge can be obtained in such a space from the knowledge already available.

As was shown in the works of P.G. Kuznetsov and P.O. di Bartini [75; 76; 19], all physical masses can only be expressed in the measurements of time and space, and consequently in this case the comparability of utterances concerning the most varied physical sizes is least likely to prove to be senseless. Bearing this in mind let us examine a concrete example. Let us look at the problem of semantics specific-ant number of automatic control system (ACS) information systems with a view to evaluating the possibilities of using natural language when a person has discourse with a machine.

Features of sense revelation of the properties of economics texts denoters.

It is easy to see that in an impressive number of cases units of an economics text can be correlated with units of external denotative semantics without much difficulty, i.e. with units of actual reality implied by the author of a text. Consequently if we start with our view of senses as a reflection of properties of actual reality in the psyche and also properties which are revealed logically, then the clarity of the external denotative correlation of the textual units, enables us to establish many important sense characteristics and units represented by the units of the text: above all it enables us to check to what extent in economics texts the condition of sense compatability of elementary utterances is realised. For this it is sufficient to show whether senses of these utterances can be formed by means of obvious and generally recognised Indicators of the properties of those denoters and situations which it is important to analyse and examine in order to solve economics problems [136, p.62].

Let us be certain that the semantics of a significant number of Economics texts in the relation examined represent an exception, rately favourable, to the various types of semantics of other texts, if it is a matter of texts reflecting the states of functioning systems of production and economic: industrial and other enterprises, ministries, departments etc.

Let us take as an example the ACS of an industrial enter-price.

Among the main characteristics of objects and phenomena of the external world are their spatial and temporal co-ordinates. In economic systems (with that degree of accuracy which is important for its functioning) there always exist perfected means for fixing the place and time of events, the displacement of objects, the moments of their changes from one state to another, the places, moments, and periods of their interactions etc. So, in particular, the absolute age of the workers of an enterprise, the moments of their change from one age group to another are always physically measured; the moment of the beginning and the end of the manufacturer of a product is fixed, etc.

The belonging of workers to a professional group, to levels and sub-divisions is determined and fixed in a uniform manner by a qualifications committee, and the identification of each worker is carried out by a worker of the licence bureau or by a time-keeper who notes down in a set manner the time spent by workers on production. Efficiency during this time is characterised by the size and quality of production, for which production is seen and frequently measured by employees of the technical control division.

In an enterprise there are also perfected procedures, appliances, and equipment for determining the expense of material and energy on the production of a set product.

It follows from this that it is precisely the properties of actual objects, situations and states of a production -and economic system which are represented by the symbols of economics texts - properties shown, as they are, at the level of physical reality by means of a system of Indicators measuring tools, identifying equipment (including ones which exploit the possibilities of human sensory organs and the logic of human thought) but the results of checking or assessing are fixed or established only by standard symbols.

In a person's psyche all these indications from the Indicators are not simply imprinted; they are distinct according to such parameters as the type of Indicator, from which a particular indication was obtained, or as the link of the indications of quality Indicators with the one indication of a time measure or with a spatial localisation. In the most real case (on the strength of its material nature) some types of properties and interactions between objects of an economic system are possible, while others are excluded, and, thirdly, another group are prepared for the development of others. Consequently, in psychological reflections of these objects with the properties revealed in them, some interactions between the reflections will prove to be possible, others will be excluded, and a third group will be prepared for the rise of data sensors and Indicators although the degree of this homomorphism can vary depending on the number of induced properties, and the amount of accuracy and solving capability of the measuring tools used.

Thus for Economics texts we have the sense level, "experimental space", reflecting the "space" of actual and possible (potential) properties of a certain area of reality. Among the relations between the elements of such a sense field there will appear ones which also have sense, and ones which are senseless and will be automatically rejected when new information is deduced from already existing information. In this case there will not be any place left for paradoxical implications, requiring a complex sense from two elementary ones to be recognised, e.g. "an atom of oxygen is heavier than an atom of hydrogen and dolphins are fish", just on the basis that the truthfulness or falsity of each component will be physically tested. Until that procedure, or that Indicator, or that measurement is shown which makes the comparison of truthfulness and falsity Indicators of component senses non-senseless an answer about the truthfulnes or falsity of a complex sense is simply nor deduced.

At the same time, if one is concerned with an economic system, and the role of its objects is taken by such economic indicators as, for example, the shift production norm of a certain product, the fact of its being carried out before time and the fact of the norm being over fulfilled (i.e. the production during a shift of more than was envisaged by the norm), then all these phenomena and objects on the most physical, external denotative level are situated in a network of fully determined links, they can form more complex phenomena and objects, and they can represent members of one and the same Boolean expression. Therefore deductable knowledge obtained on the basis of the laws of even formal logic cannot in this case be "formally truthful" and factually senseless: a formal test of the deductability of new knowledge will be an effective, although at the same time quite simple, means of deducing a new sense from one already existing, of obtaining new knowledge from experience already acquired, and a means of forecasting what is possible in an actual economic system, and what is in principle impossible, and what enables the number of direct experimental checks of a system's state to be reduced to a minimum without the risk of impairing the process of control of this system.

Let us look at an example. Let it be that in a production line the service for the registration of the fulfilment and over fulfilment of a shift norm by workers has been satisfactorily adjusted. Thus we can meet in documents only such notes as:

1. The norm (during a shift) has been overfulfilled/ the norm has been fulfilled in time

2. The norm has not been overfulfilled/ the norm has been fulfilled in time

3. The norm has not been overfulfilled/the norm has not been fulfilled in time.

In no case do we meet the note: the norm has been overfulfilled and the norm has not been fulfilled in time. This follows from the fact that before overfulfilling a norm during a shift, it has to be fulfilled before the finish of a shift, i.e. on time. In other words, here the implication sense results from the logic of things, and paradoxical deductable knowledge, providing it is not the result of incomplete initial knowledge, cannot be obtained and has no relation to the properties of the implication in itself.

Conditions for realising meaningful discourse with a machine.

Thus we arrive at the conclusion that if we have an actual economic system, and if the relevant economic parameters of this system are checked physically, and if an automated control system is capable, in spite of all the multi-stage transformations of the Indicators' signals and equipment determining these parameters, of preserving the final forms of the signals in such a way that comparable indications remain comparable, and the linked ones remain linked etc. then we can regard such an ACS as containing meaningful sense information capable of deducing new senses from those already existing, and of forecasting the states of the economic system directed on the basis of a comparison of available sense data: relating to this system. And it is only under such a condition when the machine has itself a senses field that this next question becomes important: how should a person be joined to these sensoryon the basis of a certain auxiliary sign system and how should the sense information in an ACS be changed not only through the Indicators' indications of the states of the controlled system of the deductions of new senses from those existing in the sense field of the machine itself, but also on the basis of a person's influences on the machine by means of this sign system's signs.

The discourse methods we have seen between people in their natural language enable us to answer even this question.

The introduction of a language for man to communicate with a machine in contemporary ACS's is greatly simplified if information on an object of control relates to the type of the object's characteristics which include the economic indicators of this object and, consequently, if the texts obtained from the machine and transmitted to the machine are expressers of information with a strict correlation with the physical parameters of external denoters.

Here we can clearly distinguish two main stages of introducing and perfecting such a man-machine language.

First, the introduction - and this is fully realisable in contemporary computers radically easing the interaction of a person with an ACS at the same time must be based on the principles far constructing a quasi-aesthetic language, as we have already noted.

To whatever extent objects, for example economic ones, are reflected in an ACS through its actual, real parameters which are above all numberical and computable, these parameters are to our way of thinking meaningful. Therefore the new parameters which are logically deductable from them (including ones deductable by means of numerical transformations of certain initial numbers) are also intelligible, being interpreted in terms of the properties of the controlled object. If moreover an object of control is not projected, is not created, and functions in an established regime, the changeability of the parameters of its components is based within certain boundaries, as a result of which the laws of the influence of certain parameters on others remain invariable.

And since in the process of intercourse it is necessary to name certain senses, and an enumeration of elementary senses and laws of the change from elementary to derivative in such ACS's can be considered invariable or in any case in a whole variety of senses a sub-net of named unchangeable senses can be isolated, then these senses should, according to our scheme of speech activity, belong to the class of usual senses.

Like any other, a usual sense can he expressed in natural language by means of occasionally chosen hinting meanings, and linguistic signs corresponding to these meanings.

But usual senses can, unlike occasional ones, be expressed in a standardised usualised manner. In this case in natural language the presence of meanings of usual signs of usual senses only affords interest from the point of view of etymology and motivation in the choice of signs of these sense-In the norms of discourse which have formed, the meanings of the signs of usual senses almost lose their functional load and a contiguity association can form directly, from the psyche patterns of a sign to the sense element designated. This frees speakers from the need to choose each time, when naming a given sense, the hinting meanings, and the chain-link "hinting - hinted" in general disappears in the communicative chain.

As we have seen in acts of communication in natural language a "joining" of meanings with sense, i.e. the bringing about of the relation "hinted-hinting", represents a complex creative act: experiments in effecting this act in modern computers have given poor results up to now, which is one of the main obstructions to communication with a machine in natural language (another obstruction is that the distinction of sense operations from operations with conditional signs is not understood).

But if an ACS is first of all intended for controlling the economic indicators of a controlled object, in the first stage of discourse it is only intended to name usual senses, and it is possible to do without the link "hinted-hinting".

It is sufficient to directly link through contiguity association these usual senses with patterns of the signs being used and consequently to build a language from signs having no meanings but only usual senses.

Since signs, if they are associated with senses by contiguity, can be in this case of any sort, this degree of freedom can be utilised in order to adopt as signs those terms which specialists use to name "linguified" senses reflecting the economic indicators of the controlled object.

Thus each economic indicator is represented in such an ACS by two "fields". First, without account being taken of whether communication with such an ACS is envisaged in a convenient man-machine language or not, proper sense information is preserved in the automatic machine: the economically significant parameters of the controlled object's components, and their interactions etc with internal "machine" signs which show which instruments; indicator's, or procedures' indications these parameters are based on. The presence of such information makes an ACS an operating imitation model of the controlled object, and, if necessary, the forecasts of such an ACS and the new parameters obtained in it can directly influence the controlled object and consequently its current state.

Second, if the ACS is such an imitation model, linguistic signs of the type mentioned above must be made to correspond to all of the invariable units of its sense field in the memory. By means of these signs as patterns, speech signs should be easily reproduced and, vice-versa, received speech signs should be identified with corresponding linguistic ones, i.e. they should be identified by means of standards preserved in the memory - linguistic signs.

In the first stages the easiest thing is to realise speech signs in printed-out simultaneously perforated form, which will simplify the problem of identifying them, although intercourse with a computer in this case will only be written or textual. According to the degree of perfection of the recognition systems of sound patterns communication with an ACS can be translated into a sound, or oral, intercourse.

Consequently the means of communication will be a language consisting of terms accepted in economics texts, and in this sense a person will recognise such communication as sufficiently natural. The "artificiality" of such communication will be expressed in restrictions on the types of questions which can be put to the computer which in practice will be evident in the limited number of types of answers which it can give out.

In conclusion let us note the ways of changing from the first to the second stage of working out a language of man-machine intercourse in ACS's controlling the economic parameters of objects. At this stage the language will be approximated to natural language not only externally from the point of view of the perception of the man communicating with the computer, but also internally, i.e. according to the principles for forming and analysing economics texts in the computer's internal information mechanisms.

It can easily be seen from the above that sooner or later in the content of an ACS which is capable of forecasting and obtaining a deductable sense, the deduction of a new sense will eventually be effected: this will be not only by means of rigidly determined logical schemes and not only with probability amendments which do not in essence lead us beyond the boundaries of determinatory modelling, but also by comparing the characteristics of one economic object with the characteristics of others as patterns between which there is only a loose regulating relation according to the degree of resemblance. The introduction of these regimes is only suitable when answers are sought to those problems which can no longer be solved by a determined quantitative or logical method.

Consequently these loose regimes for solving and forecasting do not cancel out more strict solutions - they simply augment them. They cannot compete with the strict ones if the latter can be effected, hut on the other hand, when a solution, having been determined, is not realisable in general, the machine is able to produce a looser, second grade solution: at least this happens - it does not in general decline to make a solution.

In conclusion I would like to formulate one purely external feature of systemological conception described in this book covering the solution of complex tasks with, an the one hand, a strict formal structuralist approach, and on the other with an intuitive one which relates more to the field of art than science.

A person having rich experience and natural intuition is capable of solving the most complex problems. But how he does this and whether he has done all that is possible in each particular case is unknown, even to himself. If in fact a direct solution to the problem can be replaced with an analysis of an auxiliary structural model constructed from elements whose properties have already been studied as a preliminary, there will be an almost certain guarantee that the solution obtained is a true one. This is because any other person with a corresponding formalism has an opportunity to recheck each stage of the solution and of assuring himself of its correctness (or of finding a mistake and correcting it). The basic advantage of such a method for finding solutions (which many nowadays consider to be the only valid scientific one) is its objectivity: all results will be identical quite apart from the individual features of the many solution methods (if of course simple mistakes in calculations or proofs do not occur).

However if a problem does not permit formalisation, then it proves to be scientifically (in the sense above) insoluble, and the number of problems formalised as we have seen, is, in principle, extremely limited. What does the systemic approach give?

First, solving through formal structural modelling is one of the primary stages of the systemic approach to a problem.

Second, one can succeed in assessing through systemolog-ical concepts the degree to which the essential conditions for successfully applying structural modelling are observed, so that "pseudo-exact" solutions which are not interpreted in terms of the components of a situation leading to the setting of the problem, are with the systemic approach excluded in actual practice- while with a purely structural approach the determination of the level of "formalisability" of a problem is only possible on the basis of intuition and experience.

Third, after a problem has been solved at the level of formalisation of the situation component, the researcher or constructor can draw on his knowledge of the laws of adaptation, the principles of the agreement of structure, substance, function material, and an object's determinant, and explain or forecast through this many precise properties of the object which can no longer be discussed on the level of formal modelling. It is true that here the effectiveness of drawing additional information through methods of systemology depends on the individual features of the person doing the solving, and his experience and erudition, and consequently the results obtained by several researchers or constructors can in principle vary. But this variation is different from that with the purely intuitive approach. It is after all a question of a difference in supplementary, more accurate, information, obtained above the maximum which is in principle attainable at the stage of formal solution. Consequently only with the systemic approach does the possibility arise of ensuring, that the quality of a solution is objectively at no lower a level then that obtained by "strict methods", and subjectively it represents something higher which only a person's creative possibilities make him capable of producing. It is precisely this principle which will in time also be realised in cybernetic automatic machines when they produce creative acts, albeit elementary ones. Without the conscious introduction of this principle into the theory and practice of modern cybernetics there is little hope of real success in the field of the creation of an artificial intellect and the utilisation of natural language for man to converse with a machine.

bibliography






| contents | | main page | | bibliography |