MODELS OF INDEPENDENT SEMANTIC ANALYSIS

	    OF TEXTS IN NATURAL (RUSSIAN) LANGUAGE

	    FOR THE AI "TEXT ---> PICTURE" SYSTEM.

		    Valentina N.Ignatova

		      vig@niimm.spb.su

			   ABSTRACT

  The problem here discussed grew out of research in constructing

  a system that displays a scene according to its description in

  na-tural Russian. The devised models of independent semantic ana-

  lysis of texts propose direct semantic analysis (exclusion is to

  both morphological and syntactical analyses) that is analysis of

  semantic structures constructing the text. The analysis is based

  on the actual segmentation of sentences' theory carried over to 

  the text. The independent semantic analysis which is based on 

  and realized by above mentioned ideas can be applied to solve 

  other problems bound with automatical text processing.
      

			INTRODUCTION.

Methods for transforming some types of knowledge representations
to others, particulary transforming natural languages to graphical
representations there and back, are of great importance for systems
made use of cognitive graphics. As a matter of fact real texts in
natural languages contain frequently graphical illustrations to
clarify their meanings. In certain cases such languages as, for
example, Venn diagram, languages of electric and different circuits
and so on may be treated as graphical languages of their own. Thus
it is possible to bring forth a problem of general methodology for
transforming some types of knowledge representations to others
within the limits of AI systems. Specific problems can be presently
solved by various types of knowledge representations. The work deals
with such a problem, that is visual scene reconstruction by means
of verbal description.

Our models for independent semantic analysis of connected texts in
natural (Russian) language are directed toward constructing
linguistic analyzers for systems which stimulate a man to understand
the text. Such systems can't be satisfied with grammatical level of
analysis only. They should be allowed for both as semantics in the
true sense of the word including text-author's knowledge on described
world, pragmatic direction, and as knowledge of correlation for text
units and given objects of reality, taking into account their relations
(not only grammatical but, for example, space relations). It is connected
with knowledge for every object domain dealing with the realistic and
possible surroundings. Our models are directed toward constructing
systems for analysis of texts of a certain object domain. It is justified
by not only practical reasons but expediency, relative simplicity and
rationality of constructing systems which are intended for special
text processing as it's possible in visible time to have experimental
results by means of available resources. The models are the basis of
the linguistic analyzer devised for the AI "TEXT ---> PICTURE" system.

In so far the systems of the "text-picture" type  consider to be
closely connected with multi-media paradigm and cognitive graphics
and at the same time to be a result of interaction between cognitive
graphics and hyper-text technology as the propounded functional model
for extracting "picturesque" information of connected text in the
natural (Russian) language can be applied to texts in different
natural languages because of identical semantic saturation and thus
can be considered as an universal model.

In the "TEXT ---> PICTURE" system which is devised also with the help
of RFFI and RAAI the texts of narrow object domain (that is a text
describing the scene observed by a man or a robot in the street) are
analysed. Specific character for analyzing the texts like these is that
the system is interested not so much in syntactical or morphological
forming of input texts as the so-called "baze" semantics and ways for
extracting information from the text. And it is allowed a graphical
planner to define a number of objects "participating" in the scene
and indicate space relations between them in order to constructive
display could reproduce the scene described by the connected text in
natural language just so correct as it is possible.

The experimental variant of the system deals 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. In the collection
of relations the following binary space relations between objects were
employed: "above", "under", "in front of", "behind", "near", "to the left of",
"to the right of".

	      INDEPENDENT SEMANTIC ANALYSIS OF TEXTS

		IN THE "TEXT ---> PICTURE" SYSTEM.

The devised models of independent semantic analysis of texts propose direct
semantic analysis (exclusion is to both morphological and syntactical analyses)
that is analysis of semantic structures constructing the text. The analysis is
based on currently central segmentation of sentences' theory carried over to
the text.

On the analogy of the theory Russian text is arranged semantically so that
new information is introduced through the information already introduced.
Except the sentences with the information that is independent of united
semantic complex (set) and as well sentences as "beginners" of connected
semantic structures because a theme is absent in these sentences.

As a rule at the start of the texts describing the space scene there is
a sentence introducing the first object of the scene and then all the others
through various relations.

In that case the first sentence is < THE REMATIC STRUCTURE > (because the theme
is absent in it). The rematic structure can consist of not only a certain
semantic element but several elements for characterizing rem's principal
object and representing as if a "group" ("bush") of it.

We consider patterns of the first-sentences of the connected text describing
the space scene and comment upon them:

1) " " ("A man is going") - The rematic structure consists of
a certain object of the scene. The verb "" is semantically attribute
for the object "" ("a man") and it performs double semantic function:

a) introduces the object "" ("a man") to the description;

b) refines-defines-differentiates the object "" and thus establishes
given object in the scene, that is " " ("a going man")
[in contradistinction to objects " " ("a sitting man"),
" " ("a lying man"), " " ("a standing man")
and so on].

2) "   " ("A man is going with a carrige") - The rematic
structure consists of two elements: " " ("a going man"),
"" ("a carriage") - linked by the SPEC-relation.

[The SPEC-relation is established between the corresponding objects and
this relation is not in line with regular space relation. Interpretation
of these spec-relations is individual for every pair. At the surface level
SPEC-relations are expressed by uncoordinated attributes. It is necessary
for the interpretation simply to have additional knowledge: "How can the
appropriate pair be pictured?" As compared with "  " ("a house
with a chimney"), "  " ("a fence with a gate"), " 
" ("a man with a carriage"), "  " ("a man with a book")
and so on].

3) "      " - The rematic structure consists of
three elements: " " ("a going man"), "" ("a carrige"),
"" ("a hat") - and each of the last two elements is linked with the first
one by the SPEC-relation.

4) "   ,    ,   ,  
 ". ("There is a house with a fence which has a gate and a smoke is
coming from the chimney"). - The rematic structure is complex and it consists
of five semantic elements: "" ("a house"), "" ("a fence"),
"" ("a gate"), "" ("a chimney"), "" ("a smoke"), - (in
contradiction to the pattern 3), structural elements "" ("a fence"),
"" ("a chimney") are replaced by relative word "" ("which") and
owing to the fact these structural elements are linked by the SPEC-relation
consecutively and simultaneously.

The semantic structure of given sentence can be represented schematically
as follows:

	  ------------>  -------------> 
       a house	   SPEC     a fence	SPEC	  a gate
	   |
	    ---------------->  -------------> 
		   SPEC     a chimney	SPEC	  a smoke

The verbs "", "" ("there is"), "" ("is coming") are not
attributes in this case but they establish the enumerated objects in the scene.

Thus the semantic model of the first sentece of the text (and of sentences
describing free objects independent of bound semantic complex as well) can be
represented in general as < THE REMATIC STRUCTURE > that accordingly consists
of the VERB defining or introducing "a new OBJECT" (i.e. the object is not
announced in the text for the present) and the "new" object itself. The OBJECT
in it's turn can consist of both a certain simple element and two or three
simple elements. Moreover the object can be compound (see pattern 4). Then
we have to do with the complex rematic structure. As these texts can be
represented by different variants of the rematic structure it is necessary
to take into account automatic semantic analysis of connected texts.

After the first sentence the semantic structure of connected text from
sentence to sentence can be represented in general as follows:

       RELATION  < THEMATIC STRUCTURE >  < REMATIC STRUCTURE >

For the second sentence of connected text < thematic structure > is conformed
to < rematic structure > (either it's "master" as a substitute of the whole
structure or a separate element of < rematic structure >) of the first sentence.
As for < rematic structure > of the second sentence it is conformed to the
information about a "new" OBJECT (which, as with above-mentioned case, can be
simple, complex or compound). And then according to the text every following
sentence introduces one "new" object or several "new" objects.

In three-dimensional scenes described by texts space relations are expressed
at the surface level by prepositions, prepositional words and adverbs. Thus
relations , ,  are conformed to prepositions "" (on), "" (in),
"" (under), "" (with) in the text. Relations ,  are conformed
to adverbs "" (more right), "" (more left), " ()"
(to the right of), " ()" (to the   left of). Relations ,  are
conformed to prepositions "" (behind), "" (in front of). Relation
   is conformed to prepositional words " ", " "
(near, next to).

It should be noticed that < thematic structure > can be "burdened" with more
precise definition expressed by participial construction at the surface level.
< The rematic structure > can be also "burdened" with a rematic element (or
elements) joined by construction with word "" (which). At the surface
level this "burdening" is conformed to attributive subordinate clause.

Consider for example semantic structure of the sentence of connected text:

   ,  ,  ,     .
relation < thematic structure>	<  r e m a t i c    s t r u c t u r e	    >

As can be seen < rematic structure > can be represented as follows:

" ,                                . "
 (relation) (substitute, i.e. ) 

Semantically "a master" of the < rematic structure > is a name of the object
"" (truck), everything else both its introducing and attributing are
elements of the < rematic structure >.

Such concept for text analysis describing three-dimensional scene is based on
algorithm of independent semantic analysis and it allows us to realize it by
means of a comparatively simple grammar (automation type).

So far as we analyzed semantic structures and relations between every pair
of semantic units were established as a result, we were released from necessity
to use a rather cumbersome system of syntactical and morphological analysis
so the problems were quite simply and compactly tackled. (It is not improbable
that widening will demand for elements of morphological analysis to remove
homonymy in certain cases. But it should not be allowed to do it hurriedly
because the system of analysis can be groundlessly swollen. At the same time
this problem is solved by interactive system. The system provides for different
methods for homonymy ("asked again" procedure, for example).

		  LINGUISTIC STEMS OF THE IMAGE GENERATION

	   IN GRAPHIC PLANNER AND VISUAL SCENE RECONSTRUCTER

Concentrate our attention on opening internal processes which arise
from semantic analysis and synthesis of texts, i.e. opening technique
for constructing and analysis of cognitive images.

1. Initial data of image generation. Linguistic identification of objects.
--------------------------------------------------------------------------

The fild of vision of our system can include approximately thirty
variant types of objects (the fild of vision of our system is broder
with objects of the same name distinguished by referential indices
and some other indications). They can be divided into next groups:

a) "stable" static objects, their names are ""  ""
("a house" or "a small house"), "" ("a chimney"), "" ("sun"),
"" ("moon"), "" ("a lime"), "" ("a fir"), ""
("a bench"), "" ("a hat"), "" ("a book"), "" 
" " ( "sand" or "a heap of sand"), "" ("a spade"),
"" ("a sand-box"), "" (  "") ("a tree" -
 in a case of failing mention "a lime"), "" ("a box"), "
" ("a sitting man"), " " ("a standing man").
These objects are always processed identically (that's the point of
their stability), they are always static (i.e. remain motionless),
they are always described by the same (but for each of them specific)
geometrical functions. They are not subjected to changes in a graphic
planner with the exception of scaling in visual scene reconstruction
(in a case of many objects are in the scene and it is necessary to
locate on display all of them). Their presence in the scene is
established by words "" ("stands") -  , , 
 (about a house, a bench, a standing man), "" ("shines") -
 ,  (about sun, about moon)*; "" ("grows") -
  (about a tree):  (about a fir),  (about a lime);
""  "" ("is" or "there is") for everything else.**

[ * The objects "" (sun) and "" (moon) besides the different
functional expression contain such indication as " " (clour
of sky). As their image-function activates the colour of sky is
located on display automatically ("-" (light-blue) or
"-" (dark-blue), correspondingly)].

[** Exception is to objects "" ("a fir"), "" ("a man")
defined by unary relations with a turn 90  to the right or to the left
(more exact: " " ("with a crown to the right") or "
" ("with a crown to the left"); " " ("with one's head
to the right") or " " ("with one's head to the left"). Such
objects are defined in the text by the word "" (lyes) and their
functions are named " " ("a lying fir"), " "
("a lying man")].

Among them objects with names "" (a chimney), "" (a hat),
"" (a book), "" (sand), "" (a spade) can "participate"
both independently and consisting of spec-functions dictated by special
relations. [Spec-relations are the relations which establish the closest
link between pair of objects and make for spec-functions (conditionally)
to produce more complex  concepts. And then these concepts can
"participate" in the text as semantic unities as follows:
"  " (the house with a chimney), "  "
(the man has a hat on), "   (the man with a book),
"  " (the man with a spade), " ( )
 " (the sand-box (or the truck) with sand). In their turn these
semantic unities can be used in the text independently as a complex
concept: "      ", for example, etc.
(There is a dog next to (near) the man who has his hat on)].

b) objects requiring for more precise definition, i.e. they can have
double positions in the scene (a user can choose, i.e. define more
precisely the position according to inquiry of the system. Among these are
objects with names " " ("a sitting bird"), "" ("a dog"),
"" ("a truck"), " " ("a car"), "" ("a fence"),
"" ("a gate"), " " ("a going man"), " "
("a lying man"), " " ("a lying fir"), ""
("a bicyclist"). These objects can be arranged with the most specific
parts either to the left or to the right. (These are, for example,
the "cab" of a truck ("" ), the "beak" of a sitting bird
(""  ), the "snout" of a dog ("" ), the "head"
of a lying man (""  ) and so on. And also "
" ("a going man") and "" ("a bicyclist") can move either
to the right or to the left. "" (a gate) in the fence can be situated
either to the right of the house or to the left of it. As for "" (the
fence) with regard to the "house" can be situated to the right of-,
to the left of-, in front of the house or the fence can surround the house.
When, "the fence" is situated to the right of-, to the left of- or
in front of the house "a gate" will be automatically planned in the same
manner. In that case if "the fence is around the house" the more precise
definition occurs. [The more precise definitions with regard to "a gate"
can take place in the previous cases as well. But being modelled by our
system they were not included in dialogue because of the danger
to block up the system.]

c) so-called "moving" objects which are named "" ("a smoke"), ""
("a cloud"), " " ("a flying bird"), "" ("a truck"),
" " ("a car"), " " ("a going man"), ""
("a bicyclist"). In the text "moving" indication is expressed by words
(it is customary to assume that the words are verbs in traditional sense):
"" ("is coming") - about a smoke, "" ("is floating") - about
a cloud, "" ("is going") - about a going man, "" ("is flying")
- about a flying bird, "" ("is going or moving") - about a truck, a car,
a bicyclist. As a matter of fact the "text --- picture" system extracts
informations from these "verbs" to reveal either following "the verb"
object is moving or not.

At the present moment considering only static scenes we cannot insist
that "a truck" and "a car" "" (are moving). That is why at entry
of our texts these objects can be introduced by words both ",
" (there is, is) and " ()" (is, are moving). But display
will represent them as static objects (and "a cloud", "a smoke" as well).
Based on analysis of situation we can approve that "a cloud" and
"a smoke" don't stand motionless so it is natural to say
" " (a cloud is floating) and "   "
(a smoke is coming from the chimney) in constructing the text.
[If we say " " (there is a cloud), " "
(there is a smoke) it will be more precise (because our scenes
are static) but less natural from the point of view of mother tongue
users.

As for "a flying bird" and "a going man" - we can say at entry of the
text that "a bird is flying", "a man is going" without any semantic
errors because in this case the words "is flying" and "is going"
are attributes functionally to the objects behind, i.e. these words
"consider" a graphical planner to select the function characterizing
"a flying bird" (in contradistinction to the function characterizing
"a sitting bird") or "a going man" (in contradistinction to the
function characterizing "a standing man", "a sitting man", "a lying
man").

2. Logical structure for description of space scene  consists of:
---------------------------------------------------

d) Formation of concepts (or complex objects). Complex objects
consider to be "  " ("the house with a fence"), " 
" ("the fence with a gate"), "  " ("the house with
a chimney"), "  " ("the chimney with a smoke"), "
  " ("the going man with a carriage"), "  "
("the man with a spade"), "  " ("the man has a hat on"),
"  " ("the truck with sand"), "  "
("the sand-box with sand"), "   " ("the sitting man
with a book"). The objects of every given pair are linked by special
relations*** and they appear together with specfunctions.
The specfunctions can be consider as complex semantic unities and
each pair can appear as a complex object. For example, in the scene
where two or several objects with name "a man" are in different
parts of the scene it is possible to construct the text like this
"     " ("A bicyclist is moving
in front of the man with a spade") or "    
 ( "") " ("There is a dog near the man
who has a hat on") and so on.

[*** Specrelations are always complex functions (i.e. complex
relations) and consist of no less than two binary relations, one of
which is named "NEAR". For example, if objects "a man" and "a hat"
are bound by relations { () (ABOVE(UNDER)) and  (NEAR)}].

Direction for use of complex semantic unities (concepts) is
convenient for constructing texts because a certain object
is defined more precisely (rendered concrete) and differentiated
in the most compact manner when different objects are situated
in the scene and it makes possible to referentiate objects of the
same name and to escape (as users wish) excessive applications of
referencial words: "" (other), "" (the second), ""
(the third), "" (the fourth) and so on. Such semantic unities
in the text are analyzed rather simply, components of complex
concepts can be independent and at entry the texts are as natural
as they can.

e) It is particularly essential for the complex object "a bicyclist"
and the complex relation "between". They were formed by the logical
way: IF ((there is "a bicycle" AND there is "a man" who is represented
as sitting in turn of 90 ) AND ("the man" is (ABOVE and NEXT TO)
"a bicycle")) AND ((vice versa)) THEN the sitting man in turn of 90
ABOVE and NEXT TO a bicycle is named "a bicyclist" and formed by complex
function, correspondingly.

Moreover "a bicyclist is moving" and not "is" or "there is", and the
verb "is moving" becomes an attribute for "a bicyclist", i.e. an
indication of moving object (it is very important for our work with
dynamic scenes).

f) The conditions of direction for use of relation "between" are as
follows:

IF ((an object Y is more right of an object X AND the object Y is
more left of an object Z) OR (the object Y is above the object X
AND the object Y is under the object Z) OR (the object Y is in front
of the object Z AND the object Y is behind the object Z)) THEN one can
confirm that "the object Y is  BETWEEN the objects X and Z".

In contradistinction to unary and binary relations under
consideration the relation "between" is three-dimensional relation
and it binds not two but three objects with each other. That is why
we call it a "complex" relation in contradistinction to simple
relations both unary and binary.

Here is a good example of the text for illustrating both "complex"
relation and "compound concept" (which consists of several (fourths)
"complex concepts"):

"   ,    ,   ,
   ".

"There is a small house with a fence which has a gate and a chimney
which smokes".

The logical form of the sentence is:

((a small house SPEC a fence) AND (a fence SPEC a gate)) AND

((a small house SPEC a chimney) AND (a chimney SPEC a smoke)) OR:

((x spec z1) AND (z1 spec y1)) AND ((x spec y2) AND (y2 spec z2)).

Here the object y2 ("a chimney") is exactly BETWEEN the objects
"a small house" (x) and "a smoke" (z2). And the object y2 ("a gate")
is conditionally BETWEEN the objects "a small house" (x) and
"a fence" (z1).

As a result of automatical analysis (complex function of two pairs
of objects) and (complex function of two pairs of objects) is formed
in visual scene reconstructer, i.e. double superposition of functions
of five objects, function of one of them (named "a small house")
occurs both in the first superposition and in the second superposition.
Complex objects (in contradistinction to compound objects) obtain
the result either simple superposition or pair of functions of
objects which is bound by conjunction of two relations.

Co-ordinates of functions of objects and updating their position
with regard to each other are formed in visual scene reconstructer
according to their relations. The visual scene reconstructer, i.e.
the scene in a whole, takes into account the number of objects
"participating" in the scene, takes a decision for scaling (if it
is needed scaling takes place in order to the whole scene described
by the text was on display) and taking into account complex produces
the scene on display by means of graphical batch - graphics.h
(including a procedure moving.h) after completing dialogue which is
needed for the system to refine certain unary relations (see point b)
about directions of moving objects (see point c) and complex objects
as well (see point d).

The more careful initial material is prepared the more careful
tbe specific character of every object separately and pairs of
objects in complex semantic unities is beared in mind and the
specific character for constructing and analyzing the text as
well. The more considered dialogue of the user with the system,
the more correct the system " - " (the text -
picture) reduces the scene described at entry by the connected
text in Russian (or other natural) language.

It is obligatory to take into account the specific character
of semantic structure of every language for analyzing the
text (for example, in English "rematic structure" precedes
"thematic structure" in most cases [on the contrary in Russian]
and there are certain differences as well).

The more correct all of the above mentioned things will be beared
in mind the less conflicting results will be obtained. So far as
the dialogue with users is "built" into the system, certain possible
contradictions (for example, homonymy and so on) can be removed by
the dialogue.

		 LINGUISTIC PROCESSOR RUN.

The input text comprises description of the scene in Russian
and it is constructed with words of fixed dictionary. The
text is contently considered as a sequence of sentences each
of them introduces one new object or more in the scene. A new
object can be introduced independently or it is bound with
earlier introduced objects by one of the relations: "on", "behind",
"in front of", "more right of", "more left of". In given
variant of the system a significant limitation on the content
of the text is that every object can be introduced in the scene
only once, i.e. the position of earlier introduced object
can,t be defined more precisely in future. It comes from the
fact there is no mechanisn of checking description of the
scene for refutation but a rather simple planner is active.

Semantic analysis of the input text is realized by linguistic
processor. Analyzing the input text it composes many objects
of the scene and displays relations between objects. Analysis
of the text is based on a comparatively simple grammar of
automation type with stack memory. During the analysis
morphological information about words (only verbs and
participles of the same stem are distinguished) is slightly
used and it is taken into account only semantic indications of
some words and units of inner structure. Naturally grammar gives
limitations on the structure of the input text. The correct text
is a sequence of sentences  when every sentence introduces one
new object or more by means of a relation which gives a position
of these objects with regard to earlier introduced objects.
Semantic structure of the sentence appears as RAB where R is
relation, A is old (earlier introduced) object, B is new
(introducing) object. Previous object and new object grammatically
are distinguished with the help of topical articulation and index
is a verb which divides theme (old object) and rem (new object).
Particular emphasis is placed upon the fact that every new object
can be introduced only by one relation. This limitation gives
possibility to be independent of checking description of the
scene for refutation.

A specific kind of sentences is regard to introduction of the
first object of the scene. There is no theme (old object) in
this sentence. Objects of special class i.e. free objects ("sun",
"moon", "a cloud", "flying bird") are introduced in the same
manner. These objects consider not to bind with other objects.
The objects A and B are formed by combination of words where
it is possible to use prepositions, cardinal numbers, ordinal
numerals and the word "which". At the same time attribute to
the old object A prescribes added notable indications and
attribute to the new object gives one new object  more. Note
that the given variant of grammer keeps pronouns and
pronominal words (except words "which", "one of them",
"other") out of  using.

The input text for processing is divided in words: a sequence
of Russian letters from delimiter to delimiter. Proper
information in the dictionary is looked for every word. The
method of search is by complete examination of the dictionary.
At the same time the dictionary is divided in two parts. The
first part consists of lexical units corresponding to
unalterable words. The whole coincidence both a word of the
text and a lexical unit is corrected for them. The second
part consist of stems of alterable words. This time condition
for  search is to correct the first elements: the word must
have lexical unit at the very beginning.

Lexical information for every word is a character string. The
first character string defines "grammatical type" of the word
which is followed by a list of the numbers of references that
the word is bound with.

Consider classes of words as follows:

f are nouns which are followed by a list. The list shows
multitude of references which can be revealed by the given
word.

V are verbs which are followed by a list. The list shows
multitude of references which can be introduced by the
given verb.

Note that morphological analysis of words is not performed
in truth. In particular it is permissible both application
of wrong cases and absence of agreements in initial texts.
At the same time verbs and participles need to be
distinguished structurally. In our system the participle
is a verb which has either letter "" or two letters ""
alongside as its part. It is enough for the fixed dictionary.

r are names of relations. Then it is indicated a number of
the relation designated by the word.

l is an attribute. Then it is indicated multitude of  references
which can be applied the attribute to.

k - word "which".

i - conjunction "and".

n - cardinal number which is followed by word meaning.

c - ordinal numeral which is followed by word meaning.

d - a word is disregarded on analysis.

A problem of linguistic analysis is to transform information
of initial description of the scene into inner presentation.
The inner presentation of the scene is a great number of
objects of the scene and for every of them is showed "a master"
(the object which is bound the given object with) or it is
marked "a master,s" abscene (the object is introduced
independently). In other words: the inner presentation is
a great number of triplets like ARB where A is the next
object, B is "a master" of the next object, R is relation
indicating position A with regard to B. The process of
analysis consists in constructing multitude like that
according to the initial text.

Grammar of analysis is a variant of automation grammar. It
is based on resulting number of states, start state and
procedure for transforming from one state to the other. The
system reads the next word. Depending on the state and
grammatical class of the word certain actions are produced
and the other state is developed. Possible actions of the
system are to built an basic object ("master,s" object), to
built a new object then to refine the basic object and to
find the basic object in provided multitude, to specify
relation bound the basic object and the new object, to
refine the new object. More precise definition is given
by attributes, participal constructions, subordinate
clauses with word "" (which). Moreover it is possible
to distinguish objects by means of referencial indexes
prescribed by ordinal numerals.

On analysis grammatical correctness of constructions is not
tested but some semantic contradictions are revealed
(uncorrect more precise definitions of objects, for example).
It stands to reason that the system can't analyze the
construction which is not provided for the grammar but
grammatical rules of agreement are not corrected by provided
constructions. Misunderstandings can appear on using
unmonosemantic constructions which can be considered not
so manner as the user supposed. The "asked again" procedure
can be used for removing unmonosemanticness.

Results of the linguistic analyzer, run - a great number of
objects of the scene with relations prescribed their
positions with regard to the "master,s object" - serve as
an "entrance" information for the next part of the system,
i.e. for visual scene planner.

		     IN CONCLUSION.

Thus we have carefully considered the system,s both
procedures for analyzing the text and auxiliary procedures
(building objects, search of the given object in the scene,
correctness of attributes consistency and so on).

On analysis messages are produced about possible refutations,
structure violations and inquiries for correcting positions of
the objects revealed by dialogue as well.

It was necessary to look for well-known linguistic facts
again because of 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 analyzer for the AI system "text - picture".

Such approach is not trivial because of universalty of the
problems; it is compact (rather cumbersome technique of
morphological and syntactical ana-lysis and synthesis of the texts is
omitted) and unpretentious for realizing (it is realized by
grammar of automation type); for the most part it is
composed "matrix knowledge" which provides efficiency
of the graphical planner and the visual scene reconstructer.

The independent semantic analysis which is based on and
realized by above mentioned ideas can be applied to solve
other problems bound with automatical text processing.

Simiral systems can be applied to various domains dealing
(including teaching foreign languages); facing the problem
of drawing a scheme, a map, a plan and so on matching its
textual description; in different robot controlling tasks,
text intellectual games and in some other applications
actualized as in the country as abroad.

		     ACKNOWLEGEMENTS

The works are followed by support of Russian AI association
and by financial support of Russian Fond of Fundamental
Investigations.

		       REFERENCES

1.  ..    ( 
     ). .  .  ... .,
   1987.

2.  ..,  ..   
      ,   
   . "-     ".
   -  -.  . -  . ...
   ., 1989. . 149 - 152.

3. Ilyin G.M., Ignatova V.N. The system, describing a scene which
   a robot observes and displaing the scene according to text in
   natural language. Interna-tional conference "Artificial
   intelligence - industrial application". Leningrad,
   15 - 19.04.1990, pp. 169-173.

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

5.  ..,  ..   
      " ---- ".  . "III  
     -92". -    2- .
   . 1992. . 1, . 147-149.

6.  ..      "".
   . "   ", N 3, 1994, . 7-12.

7.  ..,  ..    
    " ---- ".    
   -  " -96", 
       . , . 1996. . III. . 397-402.

8.  ..,  ..,  ..  
         
     . . "  
   "  5. . , . 1997.

9. 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.

10. 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.

11. 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.

12. 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.