COGNITIVE SYSTEM.

                       Maya I. Otkupshikova,
                       Valentina N. Ignatova


                            ABSTRACT :

     Consideration is being given to the elaboration of models for
     automatic synthesis  of connected texts  in natural (Russian)
     language in accordance with  the linguosemiotic  aspect. The
     models like these are directed toward making linguistic pro-
     cessors  for the system  which stimulate  a person to under-
     stand our text. As for the systems they cannot be only sati-
     sfied with a grammatical level of the synthesis.  The seman-
     tics in the broad sense of the word must be taken into acco-
     unt by them. Our models are the basis of the linguistic pro-
     cessor for the AI "picture --- text" system.

Let us consider models of the systems for modelling the personal
speech behaviour according to the linguosemiotic aspect.

The models for reproducing specific types of the personal speech 
behaviour (speech activity) are cybernetic models. Generalization 
to the speech behaviour can be reduced to three types as follows:

1. In the course of human activity the personal speech behaviour 
[when the textual production and comprehension (perception) take 
place and as a result the wanted actions are made] is accumulated 
or information is given for subsequent activity, management of 
different personal activities is responsibility of them and so on, 
i.e. the usual communication is run;

2. The speech behaviour of the translator from a foreign language 
to a native one and vice versa;

3. The speech activity in studies of native or unnative languages, 
a certain field of knowledge, profession and so on.

The types listed above differ from one another by purposes, 
initial data and results. The difference is based on producing 
different cybernetic models of the speech behaviour (compare to 
expert systems of automatic translations, information systems of 
different types, tutorial systems).

The basis for the different types of speech activity is something 
in common, otherwise the communication between members of the same 
language group would not be possible. First of all the common 
consists in a mechanism of linguistic community (or as it is 
usually said in the language system). Because of this to solve 
problems in modelling requires a general theory with explicative 
intensity for aspects of both the structure of the linguistic 
mechanism and regularities of its functions. It is well known the 
generation of such suitable theory for all modelling problems seems 
to be very difficult. And objectivity of the difficulty consists of
the fact that the language as the linguistic mechanism is a typical 
example of cybernetic "black box". It is really in existence and it 
functions efficiently but members of the communication are not given 
it in the immediate perception. They know only what is available at 
entry and at the output of the "black box". So roughly speaking the 
author knows the true sense of the text that he wants to communicate
the receiver (entry) by means of the given text (output). On the 
other hand in case of the perception the receiver is given the text 
for interpreting its sense.

We cannot immediately observe the language organization and 
function (the linguistic mechanism), among other things, because of 
the fact that there is a great body of linguistic theories, 
hypotheses and points of view which are not specific to so-called 
natural sciences. The second problem relating to the first one 
consists in complications and polyaspected formations of the 
linguistic system taking into account all aspects of automatic 
systems for speech corpora. Such limitations are essential and for 
the "picture --- text" system as well.

The second aspect of the community for all types of the speech
behaviour consists in general semiotic nature of linguistic signs 
which are characterized for speech corpora (texts), they provide 
indispensable components of the speech behaviour and therefore 
components of models for the speech behaviour.

As is known, for describing semiotic nature of linguistic signs 
they use a simple model - so-called Frege-Ostgoff triangle - its 
angles represent denotative, significative and verbal aspects of 
the linguistic sign, as for sides they represent connections of 
these aspects.

Of great importance is the direction of the connections, i.e. to 
find "a way". For our further purposes two "ways" must 
significantly be kept in mind: denotat - significat - verbum and 
verbum - significat - denotat. These "ways" model the cognitive 
process, the process of knowledge as if it be. Being so simple the 
model is quite strong to explain the semiotic nature of the 
language. It should only be added a variety of arguments. 
Practically "the triangle of significance" is used to explain a 
simple linguistic sign conforming very often to a word and not so
often to a combination of words, i.e. elements which may be given 
by final or conditionally final lists. At the same time any type 
of the personal speech behaviour and any type of models for the 
speech behaviour, correspondingly, are concerned not only with 
insulated signs of the finite alphabet but with different 
combinations of them and in practice they may infinitely be 
reproduced in connected texts. Thus we have to show that speech 
corpora (texts) posess qualities of liguistic signs and they may 
semiotically be interpreted by above - mentioned devices. [The 
position that the text has status of the linguistic sign (it is a
main linguistic sign, complex sign) is approved by a lot of
linguists (Dressler, Hartmann, Moskalskaya, Serkova and so on)
while a humber of linguists (Emile Benveniste, Zvegintsev) deny 
the sign nature of the text.

In spite of the community of the sign nature of both the natural 
language and artificial languages it might be well point out the 
specific character as follows: the linguistic sign has three- and 
not two- way - character (denotat, significat, word); the 
linguistic sign is characterized by combinations with other signs 
of the same rank (certain artificial languages have the same 
features). So combinations of linguistic sings produce complex
structural and semantic unities - sentences and texts - they 
decrease or lose a number of features of the simple linguistic 
sign and preserve the main features: three-way character, 
linearity, arbitrariness. And sentences decrease (texts completely 
lose) both the feature of reproduction in combinations and the 
feature of assymmetry between "significative" and "signified" that 
is changed according to the following regularity: the more complex 
sign, the less than probable the feature of assymmetry for it.

Thus speech corpora are complex linguistic signs which can be
described by means of the correlation: the denotative aspect - the 
significative aspect - the verbal aspect, that is arranged in the 
known triangle.

It is evident the triplicity of the linguistic sign is considered
to some degree in all types of cybernetic systems for the speech
behaviour. And taking into account correlations of verbal and
significative aspects of the speech corpus it obviously is the
condition that should be made to produce the cybernetic system
with high success. The quest for ignoring the significative aspect 
is only capable of revealing possibilities of formal features of 
words, establishing their relations with other words of the 
sentence in the initial stages of analyses for input texts. As for 
synthesis of the texts, on the whole the sense (the signivicative 
aspect) is an entry for its realization.

The third denotative aspect is certainly neglected by the most of 
projected cybertenic systems but it is obvious that except for 
learning the system of concepts (translating from one language to 
another, for example) it is essential to have a knowledge of the 
world this system described.

The denotative aspect (in other terminology it is the referencial 
aspect) became to take into account of the AI expert and other 
systems including cognitive systems.

The cognitive aspect of the speech activity is an essential
component of it. There are, for example, answers to tasks of the
school teacher: ",     " ("Tell me, 
please, what did you see on the excursion") or "  
   " ("Tell me a connected story about the 
given picture"). The way for producing the speech corpus takes 
place: the denotative aspect - the significative aspect - the 
verbal aspect (the text).

Another way is characterized by many systems for teaching,
translating, reading belles - lettres and scientific works: the 
verbal aspect - the denotative aspect - the significative aspect. 
These two main types of knowledge form the basis for the human 
activity and existence, and both are connected with knowing 
realities of our world.

The "picture --- text" system models the first type representing
the cognitive intellectual system. As the denotat we have scenes 
describing certain situations. The experimental variant of the 
system dealt with situations describing outdoor scenes with objects 
like "" ("a man"), "" ("a house"), "" ("a fir"), 
"" ("a lime"), "" ("a truck"), " " 
("a car"), "" ("a bicycle"), "" ("a hat"), "" 
("a spade"), etc. Every object has a number according to its 
standard and a referencial index that is an indication of the 
object number among other objects with the same standard. As the 
collection of relations the following binary space relations of 
objects were employed: "" ("above"), "" ("under"), "" 
("behind"), "" ("in front of"), "" ("near"), "" 
("to the left of"), "" ("to the right of"). In this case 
specific features of the denotat are evident: these are not real 
scenes which might be observed in the street but some sort of  
imitation of the similar scenes is constructed by a person and 
drawn with the use of a computer. The picture is constructed by 
the simplest graphics editor (needless to say that we don't bring 
forth problems for analyses of arbitrary pictures with an 
"intricate" entry and a process of object understanding). At 
present the editor allows for drawing objects from the fixed 
collection on the display space that is chosen by a user. In 
future it is assumed to provide a means for a new-object entry by 
the user oneself (and the user will be required a name of the 
new-object entry and descriptions of grammatical features of the 
name). Because of working the graphics editor in the system there 
are many objects of the scene constructed and every object is 
characterized by co-ordinates of its own.

Thus we are not concerned with the real world but with the world 
that is artificially constructed and reflected both real objects 
and real space relations though mediately. So the specific 
character of speech corpora is like that and the speech corpora 
can reflect and describe not only the real world. Thus the specific 
character can be discovered in a broad understanding the reality 
that contains not only the whole physical world round us but 
intellectual, social, inner worlds of human being, everything that 
was created by people including speech corpora themselves as well 
as unlimited abilities for describing real systems equally with 
unreal systems according to extended terminology: possible, 
thinkable, probable, fictitious, problematic worlds. This ability 
is realized, for instance, by fairy-tales, mythes, fantastic tales 
as well as belles-lettres works which mediately reflect the real 
world through typified images and often by means of fabricated
topics and characters.

In such a manner the problem is to define the type of the selected 
world and to act according to the choice.

The next stage of the cognitive process is an organization of the 
scene content. [It is evident that natural situations of the world 
(scene) description don't contain the distinctive division of 
stages: significative and verbal aspects closely interact with each 
other but as it takes really place the investigators can only put 
forward more or less convincing hypothesis because it is impossible 
to interpret it in a large measure].

For describing the scene content we take the method of semantics 
representation that can recently be obtained by the works about 
precise descriptions of the language for automatic problems. The 
method of semantics representation has the appearance of 
objective-predicative structures. In our case the predicates 
identify different space relations of objects. It should be 
distinguished binary relations which indicate correlations of two 
different objects and unary relations which point at the object 
positioning in the particular scene with respect to a coordinate 
system. Results for the process of the scene analyzing can be 
represented as a list of objects "participating" in the situation 
and a set M of triplets like xRy where x and y are objects in the 
given scene.

The set M is constructed so that every pair of objects 
"participating" in the situation is given all relations which link 
objects in the situation, i.e.: the set M fixes the whole 
information about the scene that is essential for our further work. 
In the system knowledge about the scene is given as a table where 
the entry is an ordered pair of the scene objects and the result 
is a set of space relations linked these objects. The list of 
studied relations is fixed. Thus the process of the transition 
from the picture to the text has two stages:

1. The picture analysis for constructing the table.

2. The text synthesis according to the constructed table.

The first stage models perception and interpretation of the scene 
by a person (the way from the denotative aspect to the 
significative aspect). The second stage models speech abilities of 
a person to speak about the perceived and interpreted scene (the 
way from the significative aspect to the verbal aspect).

In natural situation, i.e. when a person describes the scene, it 
is possible to have the text variants with the different sequence 
of objects in the text, the different choices of the initial object 
for describing the scene, the different structures of the actual 
segmentation of the text. For the "picture --- text" model (the 
model always consists of generalized and so simplified descriptions 
of objects from life) we select a certain method to generate the 
scene content and the method determines a certain construction of
the given text.

Requirements on the given text are as follows. Every sentence of 
the text conforms to the statement about a connection of one object 
to another. As this takes place, every sentence introduces a new 
object through the object already introduced. Moreover at the start 
of the text there is a sentence introducing the object 
independently of other objects. The mechanism of text-generation 
like this gives a rather good approximation and obtainable texts 
can be considered as a reasonable description of the scene observed. 
Realization of the method of text-generation dictates some rules 
of the choice from the set M of elements for the resulting sequence 
S and the rules for ordering these elements as well.

To construct the sequence like this it is necessary to define the 
rules for the choice of the specific subset from the set M. With 
this in mind it is essential for every object except one to define 
the only triplet where the object takes the position of the left 

The procedure for selecting the sequence S, i.e. to select 
"the content" from the set M, constructs a content graph, according 
to it the text is synthesized. When synthesizing there is a number 
of unique manipulations:

-selection of the verb designated as a unary predicate according 
to the object type, for example, " ", " ";

-introduction of ranks for objects and relations which define their 
importance or intimate connection with objects, correspondingly;

- introduction of definitions for  some triplets xRy, a type of 
"     " ---- "  ";

- assignment of referencial indices;

- generation of new objects, for example, "". But both 
the mechanism for distribution of text elements (the actual 
segmentation) and the mechanism of the sense repetition have 
special linguistic characters, the scheme demonstrates their work 
for generating connected texts.

Connectivity of any text just as the natural text and so the
artificial one (if it is called upon to imitate the natural text) 
follows four kinds of conditions which directly involve the sense 
structure of the text: conditions of the sense repetition; 
conditions of the sense prediction; conditions of the sense 
distribution in the text and logical relations of the text parts. 
The sense distribution plays a distinctive role, i.e. it is a 
composition of the text information. The actual segmentation, being 
a fundamental category of the text composition, provides the sense 
distribuition that in turn is responsible for the text connectivity. 
As any text phenomenon the actual segmentation is represented by 
sentences of the text. Elements of the actual segmentation are 
theme and rheme, they are not in accordance with syntactic units. 
The theme is an element of the sentence for connecting with earlier 
contexts; the rheme can be defined as a main purpose of the 
communication within the limits of the sentence, it consists in 
reporting a receiver unknown and/or essential information (from 
author's point of view), at the same time the rheme is an element 
of the sentence that has potential connections with following

Themo-rhemic structures of different degrees of complication
functionate in the texts. Fundamental elementary structures are as 

a) the canonic structure of "close connection". Schematically it 
is represented like this: the theme1 - the rheme1; the theme2 (of 
equal idea to the rheme1) - the rheme2; the theme3 (= the rheme2) 
- the rheme3 and so on. For example, "   .  
 .   ". [For comparision with the 
"picture-text" system: " .    
.    ".]

As a modification of the "close connection" structure may be
thought of the structure that is produced by means of the
quantifier word as plurality with the first rheme: the theme1
- "" (the rheme1); the theme2 (" ") the rheme2; the
theme3 (= " " the rheme1) - the rheme3 and so on.
"    .    .
      ". [For comparision with the
system: "  .     .
 .     .

b) the parallel structure: the theme1 - the rheme1; 
the theme1 - the rheme2; the theme1 - the rheme3 and so on.
For example, "   .    
.   ". [For comparision with the "picture-
text" system: " .   .   ",
converted by the system with automatic editing to "  

c) the reverse parallel structure which realizes relations
between different themes and one and the same rheme: the
theme1 - the rheme1; the theme2 - the rheme1; the theme3
- the rheme1 and so on. The construction is characteristic
of not frequently found text instances of predicate substitution
by means of special substitutes the type of: ",  (),
, ". "  .  - ". [For comparison
with the system: "   .  
,   " or "  .  

d) the "bunched" structure: the theme1 - the rheme1, the
rheme2, the rheme3 ... These constructions are specific to
a type, for example, of compositions with homogeneous
predicates which can be believed to be separate sentences. " 
   .    .   ". 
And so on.

In natural texts the given structures can be obviously used 
together with each other. The texts where only one structure of 
the actual segmentation is represented are rare and if they are 
long enough they can be interpreted as a little bit artificial but 
sufficiently understandable texts.

As was shown above the "picture ---- text" system takes into
account all types of the given structures (both in the pure state 
and in combination with each other) [8], [9], this fact enables to 
offer strong possibilities of the system and provides the texts 
synthesized not only compact but natural form (from language 
bearer's point of view).

Let us consider an example of the text synthesized by the system:

 " .   .    ,   
 ,   ,    .     
.     .    ,  
   .     ,   .  
  ,     .   -
   .       . 
.      ,   .  
   .      -
  ,    .     
.     .   ."

The text makes allowance for possibilities of the system almost
without exceptions.

We can see quite well using language mechanism of substitution with 
the sense repetition of the theme and the rheme. The substitution 
is realized by means of "which-word" constructions and above all 
it is possible to do "sticking" two simple sentences in a 
complicated sentence.

It should be noted that the specific character of the actual
segmentation in Russian [to be expressed by means of words, order 
(for neutral speech styles)] and the triadic composition of 
elements in the sentence (xRy) as well provide a means to carry out 
the first experiences without expansive morphologo - syntactical 

                     In conclusion.

In was necessary to look for well-known linguistic facts again 
because of the specific character of our problem, i.e. 
cognitive-graphical approach and desire to develop and use 
hyper-text technologies that brought to unexpected but rather 
effective results and allowed us to move on making the linguistic 
processor for the AI "picture --- text" system.

Similar systems can be applied to various domains (including
teaching fo-reign languages), facing the problem of drawing a 
scheme, a map, a plan and so on matching its textual description 
in natural language, in different robot controlling tasks and in 
some other applications.


Our works at the models and the linguistic processor for
automatic synthesis of connected texts in natural (Russian)
language are followed by support of Russian AI Association
and Russian Fond of Fundamental Investigations.


1.  ..,  ..,  ..,  ..
    ().   . .: 
     . - : , 1975.
   . 24 - 25.

2.  ..,  ..,  ..  
   .  . . 25 - 26.

3.  ..      
      ( - ):
   -. . . - ., 1967.

4.  ..,  ..,  ..  
    ,  . "".
   - . -. ., 1977. . 22 - 28.

5.  ..,  ..  " 
   ---- "       
    () . . " 
    ".     12- 
    -12. . 1982. . 524 - 526.

6.  ..,  ..    
   -      .
    . . 527 - 529.

7.  ..,  ..    
   - ()    , 
   . "-   . . -
   . -, 1990, (-   
   ; .25). . 171 - 184.

8. G.M.Ilyin and V.N.Ignatova. A system for describing a scene
   which a robot observes and for displaing the scene according to
   text in natural language. "Industrial applications of artificial
   intelligence". North-Holland. Amsterdam. 1991. Pp. 130 - 134.

9.  ..,  ..  " ---- ".
   . "   "., Software and Systems.,
   N2, 1992, . .48-52.

10.  ..      "".
    - "   ", N 3, 1994, . 7 - 12.

11.  ..    ( 
    -  ). .  .  ... .,

12.  ..,  ..,  .. 
      .  ."  
    ". . 5. . , -. 1997.

13.  ..,  ..   
       " ---- ".
    . "  ", . 4,  
     "-    "
    "-96".   . - "", . 1996.
    . 40 - 53.

14. .., ...     
    " ---- ". -96.    
      " -96". 
    -    . . 1996.  III. . 397 - 402.

15. G.M.Ilyin and Valentina N.Ignatova. Logic-linguistic processors
    for the AI "Picture ---- text" system. Formalizated description
    of data for constructing matrix knowledge. The First
    International Conference on Problems of Dynamic Objects
    Logic-Linguistic Control "DOLLC'97", June 16-20, 1997.
    Saint Petersburg. Russia. Abstracts. Pp. 44 - 47.

16. Valentina N.Ignatova. Logic-linguistic stems of image generation
    in graphic planner and visual scene reconstructer for the AI
    "Picture ---- text" system. The First International Conference
    on Problems of Dynamic Objects Logic-Linguistic Control
    "DOLLC'97", June 16-20, 1997. Saint Petersburg. Abstracts.
    Pp. 40 - 43.

17. G.M.Ilyin and Valentina N.Ignatova. Logic-linguistic processors
    for the AI "Picture ---- text system. Formalizated description
    of data for constructing matrix knowlege. The First International
    Conference on Problems of Dynamic Objects Logic-Linguistic Control
    "DOLLC'97", June 16-20, 1997. Saint Petersburg. Pp. 59 - 68.

18. Valentina N.Ignatova. Logic-linguistic stems of image generation
    in graphic planner and visual scene reconstructer for the AI
    "Picture ---- text" system. The First International Conference
    on Problems of Dynamic Objects Logic-Linguistic Control
    "DOLLC'97", June 16-20, 1997. Saint Petersburg. Pp. 69 - 78.