Semiotic Models in Artificial Intelligence Problems
Proceedings of the Fourth International Joint Conference on Artificial Intelligence. Mathematical and Theoretical Aspects of A.I. Pages 65-69.
© 1975 International Joint Conferences on Artificial Intelligence
Abstract
The paper shows that the state of art in artificial intelligence requires development of a semiotic system theory which should play in this field the SaJie role that was played by formal systeDIS at its initial stages. Basic problems of semiotic model and systems theory are discussed. Ulustrating exaJDples are given.
- V.N. Pushk:in, "Short-term intel- l:i8ence in large-scale systems", .
- M., "EDergia", 1965 (in Russian).
- V.N. PU$hk:in, "Heuristics am cybernetics", .
- M. , "Znanie", 1965 (in Rus- sian).
- V.N. Pushking, ''Psychology am cybernetics", .M., ''Prosveshenie", 1971 (in Russian).
- D.A. Pospelov and V.N. Pushkin, "Intell:ieence an<l Automata", 11., "Sovet- skoye. Radio", 1972 (in Russian).
- H.Allani, S.Ceccato, E.Naretti, "Classification rules and code of an operational grammar for mechanical trans- lation, No.4, Lornon, part 2, 1961.
- Yu.D. Apresyan, "Russian verb Seman~ ics", M., "Nauka", 1967 (in Rus- sian).
- E.F'. Skorokho<l' ko , 11 Experience in lexicon compilation for information language using a step-by-atep approach for the coding o~ terms", M., VlNl!ri Publ. , 1961 ( in Russ ian) •
- D. A. Pospelov, "System Approach to Intelligence Simulation", In "Metho- dological Problems of System Analysis", .
- M., "J.\ysl 11 1970 (in RUss ian).
- M.M. Bongard, ''Problems of Recog- nition", 14., ''Peyzmathgyz", 1967 ( • in Russian).
- Yu.I. IQ.ykov, "Situative control of large-scale systems", M., ":EDergia", 1974,(in Russian).
- D.A.Pospelov, "Situative . cont- rol principles, 11 USSR Acad. of Sc.Trans- actions, Engineering Cybernetics, No.2, 1971 ( in R1,18s ian) •
- L.S. Zagadskaya aDl Yu. I.KJ.ykov, "Pract leal a,ppl icat ion of the s ituat ive control model", USSR Acad.of Sc.Trans- actions, Engineering Cybernetics, No.2, 1971 (in Russian).
- V.F. Khoroshevsky, Yu. I.Klykov and L.S. Zagadskaya, "$ituatj.ve control a oft..., are", USSR A cad. of Sc,Transact ions, Engineering Cybernetics, No.2, 1973 (in Russian).
- Yu. I. IQ.ykov and T.N. Rybakova, "Model of the game II a I I ' In "Computer Facilities Applications", M., MEI Pub- 1 ic. 1967 (in Rusa ian) •
References were extracted automatically from the PDF and may contain errors.
Full text extracted automatically from the PDF; figures, tables and equations may differ. The PDF is the version of record.
Introduction
.At the beginning of this century the classical works of Dunker have led to development of a psychological theory of thinking which is nav known as the "maze hypothesis". In this theory human haDiling of a creative task is described in the framework of a model in which man faces a maze of possible paths frOD the initial node in the labirinth to a cer tain node which is associated with the desired solution. Each intermediate node characterizes the time of selecting a solution and the corridors, implementa tion o:t the decision made. The spec±ric of human thinking i.s that analyzillg a maze of choices he fiDis a path leadil:s to the goal spend~ not too much time on it. The maze byp othes is came under much fire from psychologists, its fu:o. damentals were experimentally tested and mostly disproved.
Cont rol engineers who worked on au tomating the tasks traditionally regard ed as creative rendered, however, effect ive SUpport to the lzy'pothesis. In the fifties ani :.. ixties the maze theory re~ned unchallenged in all programs for reigned unchallenged in all programs for provmg theorems, behaviour prog rams, arts composition programs, game programs, etc. All heuristic programm ing was built around maze theory con cepts. Therefore the attention of arti ficial intelligence specialists was fo cus ed on search organization aJXl reduc tion, determining the time o:t search cess at ion and ascertaming the advisabi lity of a spec±ric action at a given stage ~ solution. The early successes seemed to corroborate the validity of the initial psychological concept.
In late sixties it became evident ' however, that all problems cannot be sol"V"- ed withm the framework Of the maze modeL One should remember development of one of the most powerful maze programs, the Ge neral Problem Solver of Newell, Shaw and Simon. The core of the program was a lo gic theory machine for proving theorems in proposition calculus. In terms of the 11aze model the routine can be described as follows: the initial spot of the maze 1s the left-hand or right-hand pa;t't of a certain equality which should be proved, the final spot is an equality with iden tical parts. The intermediate solutions are operators appl.ied to the both parts of the equality. For each intermediate result the selection of decisions is dic tated by certain additional considera tions related with the form of this inter mediate result. The routine proved effec tive in proving the theorems but direct application of the logic theory machine to the General Problem solver designed to solve a wide spectrum of problems showed that they are inapplicable for example, to chess playing problems.
There is one important feature of the maze model which probably is respon sible for its popularity ~ong control engineers. In mathematical logic the ana log of a maze model is the well studied formal system model which is the core of proposition calculus, predicate calculus, formal grammars and ma~ other models whereas in formal system theory the out put was always understood as search in a maze of possibilities. This interpreta tion is caused by the fact that formal systems are essentially syntactical sys tems unaffected b,y the semiotics and pro gmatics of phenomena and processes. This is the source of pa.ver of formal systems but this reduces their effectiveness in artificial intelligence problems.
The l.illlited capacity o:t formal sys teJIS bec~e especially apparent in deve lopment of large man-computer dialog systeiiS using a natural language in plan ning actions of an integral robot with a model of environment. In most such prog raJDS the central subsystem was a certain fomal model; as a rule, this VIas first order predicate calculus. ~uch were a number of dialog programs developed by B. Green, the STRIPS system developed in Stanford and a robot activity plann ing system developed by T. Vinograd. To organize work of a formal system, howev er, the entire environmental source in formation should first be translated into a language of correctly constructed formulae of the system used. Following derivation of the result by formal tools the result should be transformed into a form suitable for realization in the environment. It is these transforming systems that are most difficult in all programs of this type. l''orcing the infor mation into the Procrustean bed of' the formal system they take most time in program illplementat ion and. reduce the system effectiveness.
We will deal with another psycholo gical conc~t which leads to semiotic rather than to formal models. It is ray belief that shortly most programs relat ed to description of the environment, behaviour planning in it and man-comput er d ial.og will be bull t around s em iot ic models.
The Psychological Concept
Unlike a maze model where the maze of possibilities is assumed to be spe cified in adva~e am the entire crea tive process is reduced to organizing a search in this maze, the conc~t develop ed in recent years by a Soviet psycholo gist V.I. Pushkin ar.d his followers pro ceeds from the assUJDpt ion that the main creative act in solving a problem is constructing a fragment of the maze in which the path leading to a solution is found with a large probability rather than search for a path in a rea~-made maze.
Consequently, the deductive static model which underlies the maze concept is replaced by and inductive dynamic mo del. This approach was corroborated in numerous e:xperillental studies of human behaviour in solving creative problems and, in particular, in studying human be haviour in chess playing[1], [2),[;], (4]. In contrast to the maze concept this one may be termed model approach. It is built around the following principles :Interest ing for control engineering applications.
- The description of the initial problem situation should be structured and then it can isolate basic concepts and their interrelations important for solution of the problem posed.
- If the objective structure is described in the same language as the problem situation then the maze approach can be used.
- ;. If the objective structures and
the problem situation are described in different languages, then a language should be found to make them co11patible.
- If there is such a language, then it can be u.s ed to construct a frag ment of a maze of possibilities.
- The search in this fragment can be performed in aey way as in maze prog rams.
6. Making the descr:iption o:f the problem situation and objective structu res is ilDp oss ible within the framework of syntactical systems since this pro cess requires a semantic and pragmatic level. ·
Semiotic models
Let us now describe a certain sys tem which will evidently meet the requi rements made by the model concept ot· thinking to computer programs.
ELements of these sets will be referred to as basic concepts, solutions and re lations. Introduce DON inductively the concept of a correctly stated formula, CSF, a concept common for formal sys tems:
1. Any element of A is a CSF. 2. Arsy expression of the form (Clilj'lt) is a C3F.
3. If o{ is a CSF, the ( O;.'lj ~) and (ol'l.jai) are also CSF's.
4. If cl.. and jJ are CSF · s, then (oCljfl) is also a CSF.
5. If d. is a CSF, then (~ 1 f>i) is also a CSF.
6. There are no other CSF's.
Any totality of CSF 's is a text. If there is no CSF of the form ( ol.., pi) among the Cill''s that form the text, then the text forms a fact, otherwise it forms an artifact. Facts are denoted as F'i and artifacts as cpi ·
A geometrical analog of texts are multigra,phs whose vertices are associat ed with elements from the sets R and P and arcs, with the relations 'l.j between appropriate vertices.
Art if acts of the form (d. 1 Pi ) will be referred to as terminal. The deriva tion in our system will be uooerstood as search of an ma,pp ing from a set of texts onto a set of terminal art if acts. One specific feature of the model ls that the derivation rules are applied to texts in agreement with certain applicability laws which are formulated in their turn in agreement with the store o:f artifacts accumulated at the preceding derivation. Let us describe the process in more de tail. At the initial stage a certain ini tial text is specified which acts as a tentative axiom. A certain set of deriva t iOD ruJ. e6 n : { 'ii 4 1 fi 2, I, ., I 1i # J is specified. These rules specify mappizlgs of the form {F.:}--.{~·} { ~.J-.(m.1 { 4>.:}-{ cp,} " ) ' 't'd_,
and a particular case of the latter two mappings are the mappings { F'.:\ ... L anQ. { 4>.:}-+L where L, denotes a set o:f ter ainal artifacts. Let ~(ft,;Tj)denote the rule for applicability of the derivation rule 1fi to the text 7j . This rule
may be either binary predicate with valu
es "applicable" or be of a much aore complicated nature (for instance be an attribution function for a fuzey set in the sense of Zadeh). What is illportaut is to emphasize its dependence on the set l., o:f terminal artifacts stored by the system during the derivation. If in that process a new terminal artifact has been obtained, this is introduced into the storage of artifacts. If the role of a tentative axiom is plaYed by a text incorporated into a certain artifact al ready stored in artifact memory then only this artifact is invariable derived from this text.
Introduce now a basic definition. A
semantic system is a six-tuple, A, R, n P M C" where M is the totali-' , .~ ...... ty of rules for obta;i.ning <SF's and elements of the set L .
Depending on the structure of the sets n ,Sdifferent Selliotic systems will result. In the following Section we will describe one af such possible systems.
The Qyromat
This Section will be concerned with a hypothetical unit named gyromat in ana logy with the facility described in a science fiction novel by Stanislaw Lem. That unit could restructure in response to changes in the environment. Our gyro mat which is a certain implementation of a semiotic system is also adaptive in this sense.
A semiotic model of the gyromat r&- ! lies on a standardized natural language. The part Of elements Q is played by concepts Of the language. There are three types of these: concrete concepts, clWiS es and abstract concepts. Concrete con cepts are specified by a set or values at the features<~,,~,_, . .. ,~~~)· These values are taken from certain fixed sets Of values G 'I a l., .... G" · If 'J,i is rep laced in the set or values by }. then th j.s concrete c ooc ep t has no such f eatu re. Classes are specified by certa.in at tribution functions which depend on va lues of the features. The concepts of class ms;y be made specific by specifying the DaJDe of that concept as well as by specifying the set of values of features. The set of names will be denoted as I and ita elements as l j · Abstract con cepts are determined by certain texts.
Elements oi' the set P are impera tives of the natural language. Such as the words axxl combinations "Go", "A path should be traced" and others.
Elements of the set R are metare lations of the type "Go to ··· ", "Object - action", "Cause - effect", "Be aH.multancously", "Have a name", etc.
About the po.ver of the set for a natural language there is a hypothesis that this set is finite. This fact is established for the Italian[5 J and Russian language (6),[?]. To find the power orR texts in a natural language were studied and nt:NI relations were discovered that were in troduced into the li!lt of relations found previously if these new relations were not expressed by super-posit ion of rela tions fouxn previously. This process v1as completed when about 200 various rela tiom were found in the list of basic re lations.
Let us use a simple example to shaw translation of a text in a natural lan guage into that of a semiotic system un derlying the gyro11at structure. The na tural text is: "Nick and his dog left the house ai¥l went to the forest". The following notation is introduced a,- a man, a,- a dog, i~- Nick, ta.- Nick's dog, a,- the forest' "~- the house,
'l1 - to have a name, 'la.- go from, r, - go to, 'Z,- simultaneously, ~s-- after that. Then the initial text takes the form (((( o,"L,L,)'l~(a,_'l,i 1))'l 2 at)'ls(((a, "l, i. 1 )'lt (a,. "1.1 il)) 't~ a 3 )) .
Note, that the retranslation is not uni valent. Thus the text: "Nick and his dog rushed from the house ani ran to the fo rest" has exactly the same structure as the previous one aJld they are not distin guished in the gyromat. If this differ ence is essential, names of relations may be introduced. In this case the words "went", "ran", "rushed", "dragged their feet", etc. will be different names of the relation "go to".
It is easy to see that practically any texts of the natural language may be represented by texts of the gyromat · .If necessary, the set R can be completed with sets of modalities, estimates arxl fixation of emotional states related with the gyromat's Ego. Ref.[ 8 J shows how this js done.
Let us now take up the rules for derivation of n . These rules be are divided into three subgroups. The first one incorporates rules for updating the facts, or rules of the form (Pi}-(P,:}. These rules relay on formal properties of the relatioJE alld tbeir semantics. Formally the relations can have or have not properties such ~ reflexivity, sym metry or trans it ivity. The presence of such properties may be used to add rela tions to the text. lor instance, the sentence "There are three rooms in the house and two windows in each room" es tablishes the relation of possession between the concepts of house and room and between the concepts of room and win dow. Because of the transitivity the re lations of possession of the rule n will establish thjs relation between the COJ;lcepts of house and window. The use of semantic properties of the relations ·ay be illustrated with the sentence: "In the evening birds fly 101'1 over the ground". lD this phase there are rela tions Such ~ "object - action", "action - time" and "action - site". The rela tion "to be over" between the concepts of a bird and the ground is established through analysjs of the composition aDd structure of relations in the sent ence. For the Russ ian language about 400 rules for facts Updating have been
fown.
A second subgroUp of derivation rul es controls transformation of facts by introduct~on of derivative concepts aDd relations. New relations are introduced in three ways: by elimination of naJDes of conc~ts and relations, by generali zation in terJJS of the features CJ,i aDd by generalization in terms of relation structures. The el.iJIIination of names results in replacement of personified facts related to concrete concepts by facts related to cl~ses. Generalization in terms of features is performed by in troducing new cl~ses using attribution eigenfunctions. This process is well known and described in maizy" papers, e.g. in a monograph(9]. This generalization leads to concepts such as "red", "round and sweet", etc. Generalization in terms of relation structure will be illustrat ed with the following example. Let 01 be the "man" class and iJ' j' = f ,., lo > I F--1"'_) 100 1 be certain names. Introduce the relation a,_- a square and the relations; rz, - "have a name", rc,- "to touch", 'ls - "be behind", 'l, -"be s imul tane ous", 'l,. "be in" · · · Consider the following text {( .·. (( Oj 'lt it )'lit ( ct /lJ i, » rz, ... 'ltt (at'l· i,,.) ... )tr.ra,) , olJ~(aJ'lJij).
It signifies that 100 people are in the square s:iJDul taneously. The symbols are introduced to siJDplify the notation. Then
( ... ( d.{l., rJ..rJ't.(rJt'ltol.J'l.,(olll.J.ol't))'l,(olt T.lrl.t)l'?.t {{ of.,t 'l,t a{l»'l,{olt "t.tolS)), .. ( ol,H 'lL o( JJ) .. . ) .
Thjs signifies that the people are rather dense in the square. The situation is e~ily generalized into a new concept of a "crowd in a square". The concept of a crowd is generated due to the presence of uniform relations between r:J.i aDd because these relations denote ''being simultaneous" and "touching". If the re lation 'la, js replaced by 'l 3 , then with an orderliness of the type (ot. yt. o£. ) c. .. '~' we will have a queue rather than a crowd in the square.
Unlike updating rules, the rules for generalization in terms of features and structures are not specified in advance but formUlated during the gyromat funct ioning as a function af the set of ar tifacts.
Let us now take up the set of rules 8 . Divide the set o:t facts into two subsets, P 1 and ~2. · Let us assume
that the appearance of an artifact (F4 1 Pt.)
for which ft belongs to ~,j is a favor able event whilst tbe appearance of the artifact (Pj, P1J in which Fjbelo~s to P 2. an unfavorable event. Then the ru les 8 have t~ following structure: it the application of Pk results in a favorable artifact, this is entered into the artifact store and the rules 2 which were ~plied to the initial text to obtain F;, are assumed applicable to that initial text, . otherwise their appli cation is banned. But because text trans format ion is a multi-stage process, 8 are of a more complicated nature descri bed in Refs [ 10 ],[ 11J.
The gyromat was used in solving real life probleJOO such a goods handling control in a seaport, solution of game problems, management, etcl12] · [ 13) ,[ 1~.
To conclude let us note that a gy romat model is being successfully test ed in man - computer dialog systeJIS us ing a natural language. Although not a sole implementation of a semiotic sys tem, the gyromat demonstrates the poten t ial of such systems handl i~ art if ic i al intelligence problems.
@inproceedings{ijcai1975p11,
title = {Semiotic Models in Artificial Intelligence Problems},
author = {Poepolov, D.},
booktitle = {Proceedings of the Fourth International Joint Conference on
Artificial Intelligence, {IJCAI-75}},
publisher = {International Joint Conferences on Artificial Intelligence},
editor = {Patrick Winston and Carl Hewitt},
pages = {65--69},
year = {1975},
note = {Mathematical and Theoretical Aspects of A.I},
url = {https://ijcai.org/proceedings/1975/11},
}
TY - CPAPER TI - Semiotic Models in Artificial Intelligence Problems AU - Poepolov, D. T2 - Proceedings of the Fourth International Joint Conference on Artificial Intelligence, IJCAI-75 PB - International Joint Conferences on Artificial Intelligence PY - 1975 SP - 65 EP - 69 UR - https://ijcai.org/proceedings/1975/11 ER -Copied
Machine-readable versions of this paper, for tools and AI agents: