Knowledge Representation: Concept, Techniques and The Aanalytico-Synthetic Paradigm

Knowl. Org. 28(2001)No.
1 5
J.C. Binwal, Lalhmachhuana: Knowledge Representation: Concept, Techniques and the Analytico-Synthetic Paradigm
Knowledge Representation:
Concept, Techniques and the
Aanalytico-Synthetic Paradigm
Jagdish Chandra Binwal and Lalhmachhuana
Department of Library and Information Science,

North Eastern Hill University, Shillong, India
Dr. Jagdish Chandra Binwal is a professor in the Department of Library and Information Science and
the Dean of the School of Economics, Management and Information Sciences at the North Eastern
Hill University, Shillong, India. He has also taught library and information science at the Sana'a Uni-
versity, Republic of Yemen during 1996-1998. He served as a university librarian of the North Eastern
Hill University from 1978-1991. He has been a Senior Fulbright Fellow to the Department of Infor-
mation Science, University of Pittsburgh in 1988. He is the author of 3 books and 54 articles. His
areas of research include structure and development of the universe of knowledge and knowledge re-
presentation.
Lalhmachhuana is a research scholar in the Department of Library and Information Science, North
Eastern Hill University, Shillong and is currently working in the Central Library of the university.
His areas of research interests include structure and development of the universe of knowledge, orga-
nisation of knowledge, and information retrieval.
Binwal, J.C., Lalhmachhuana (2001). Knowledge Representation: Concept, Techniques and the
Analytico-Synthetic Paradigm. Knowledge Organization, 28(1). 5-16. 25 refs.
ABSTRACT: Defines the concept of knowledge representation. Explains the major AI-based knowl-
edge representation techniques developed so far. Critically examines the strengths and weaknesses of
such AI techniques. Argues that the analytico-synthetic approach advocated by Ranganathan is really a powerful knowledge
representation technique containing in itself the epistemological foundations required for real progress in the field.
0. Introduction intelligence is concerned with the creation of a know-

ledge base which, when programmed, results in ma-
Knowledge representation has been the focus of at- chine-based intelligence. In psychology, the represen-
tention for library and information science since its tation system of cognitive theory and models of hu-
origin. Today the notion of knowledge representation man memory are very much concerned with the is-
has pervaded many other disciplines. These include sues of knowledge representation. There are three
computer science, artificial intelligence, linguistics and main approaches to knowledge representation in co-
psychology. In computer science, the issue of know- gnitive theory. First is the computational model re-
ledge representation arises when data structures and presented most systematically by Fodor (1975). Se-
the structure of records and files in databases are to be cond is a combination of a biological approach and a
decided. In linguistics, the problem of knowledge re- computational model as represented by Pylyshyn
presentation arises when dealing with the syntactic (1988). Third is the constructivist tendency as repre-
and semantic structure of natural language. Artificial sented by Piaget (1962).
https://doi.org/10.5771/0943-7444-2001-1-5
Generiert durch IP '172.22.53.54', am 01.05.2022, 00:21:12.
Das Erstellen und Weitergeben von Kopien dieses PDFs ist nicht zulässig.
6 Knowl. Org. 28(2001)No.1
1. Concept of Knowledge Representation Davis, Shrobe and Szolovits (1993) argue that the
notion of knowledge representation can be best un-
Let us now attempt to determine what is meant by derstood in terms of five distinct roles. These are:
knowledge representation. Winston (1984) has descri-
bed it as “a set of syntactic and semantic conventions i) A knowledge representation (KR) is most funda-
that make it possible to describe things.“ Brachman and mentally a surrogate, a substitute for the thing it-
Levesque (1985) state that its concern is “to write des- self, used to enable an entity to determine conse-
criptions of the world in such a way that an intelligent quences by thinking rather than acting, i.e., by
machine can come to new conclusions about its envi- reasoning about the world rather than taking ac-
ronment by formally manipulating these descriptions”. tion in it.
According to what has been stated above, it appears Viewing representation as a surrogate leads
that the notion of knowledge representation is an easy naturally to two important questions. The first
one. It is merely writing down, in some language or question is: How does it correspond to its inten-
communications medium, descriptions that cor- ded referent in the world? It means that there must
respond in some salient way to the world or a state of be some form of correspondence specified between
the world. However, the concept of knowledge repre- the surrogate and its intended referent in the
sentation is not so simple. world. This correspondence constitutes the seman-
Knowledge representation implies some systematic tics of the representation. The second question re-
way of codifying domain knowledge. It views know- lates to fidelity: How close is the surrogate to the
ledge as consisting of “facts” and “heuristics”. Facts real thing? Which attributes of the original does it
represent widely shared, publicly available, and gene- capture and make explicit, and which does it omit?
rally agreed upon knowledge in the domain. Heuri- Perfect fidelity is impossible, both in practice and
stics represent knowledge of good judgement that in principle.
characterise expert level decision-making in a field.
Knowledge needs to be organised in a manner that al- ii) It is a set of ontological commitments, i.e., an
lows easy access, quick processing and is easily built- answer to the question: In what terms should I
upon. It must produce intelligent results, congruent think about the world? It is impossible to repre-
with the way people think (Dubey, 1987). Minsky sent the world in its full detail. It is, therefore, ne-
(1990) is of the opinion that there are many approa- cessary to restrict the attention to a small number
ches to thinking. Consequently, it results in many of concepts which are meaningful and sufficient to
approaches to representing knowledge. According to interpret the world and provide a representation
him, there are two extremes, connectionist and sym- adequate to a certain task or goal at hand. As a
bolic. But, neither is adequate. consequence, a central part of knowledge represen-
There are two basic dimensions to knowledge re- tation consists of elaborating a conceptualisation: a
presentation. These are: knowledge structures and set of abstract objects, concepts and other entities
reasoning mechanisms. Knowledge structures help to as well as the relations that may hold between
organise knowledge into predetermined structures. them. The commitments which are implied by the
That way, they are passive, whereas reasoning mecha- choice of one set of concepts instead of another to
nisms are active. They manipulate the structures to describe a certain phenomenon are called ontolo-
produce useful outputs such as inferences and answers gical commitments (Valente and Breuker, 1996).
(Liebowitz & Beckman, 1998). Further, all representations are imperfect ap-
It is not possible to represent knowledge without proximations to reality, each approximation at-
knowing its properties. Parsaye and Chignell (1988) tending to some things and ignoring others. It in-
have identified five elementary properties of know- volves a set of decisions about how and what to
ledge that can be used to represent objects and their see in the world. In other words, selecting a repre-
interactions. These are: sentation means making a set of ontological com-
mitments. This helps in focussing attention on
a) Naming (Proper nouns)
aspects of the world we believe to be relevant. The
b) Describing (Adjectives)
focussing effect is an essential part of what a repre-
c) Organising (Categorisation and possession)
sentation offers, because the complexity of the na-
d) Relative (Transitive verbs and relationship nouns)
tural world is overwhelming. We need guidance in
e) Constraining (Conditions)
https://doi.org/10.5771/0943-7444-2001-1-5
Knowl. Org. 28(2001)No.1 7
deciding what in the world to attend to and what sentation. We cannot talk about one without dis-
to ignore. cussing the other.
Ontology is, therefore, required to have the fol-
lowing characteristics (Valente and Breuker):
2. Knowledge Representation Techniques
(a) Parsimony;
(b) Clear theoretical basis; There are a number of systems of knowledge repre-
(c) Categories; and sentation available today in the field of Artificial In-
(d) Coherence telligence. These include Logical Systems, Production
Rule Systems, Semantic Networks, Frames, Scripts,
iii) It is a fragmentary theory of intelligent reasoning, and Conceptual Dependency. These systems can be
expressed in terms of three components: broadly grouped into three categories: Logical Sy-
(a) the representation’s fundamental conception of stems, Production Rule Systems, and Structured Ob-
intelligent reasoning; jects Systems comprising Semantic Networks, Fra-
(b) the set of inferences the representation sancti- mes, Scripts and Conceptual Dependency. Many
ons; and other systems are still evolving.
(c) the set of inferences it recommends.
2.1 Logical Systems
There are five distinguishable notions of what
constitutes intelligent reasoning. They have been Logic has been used as a method of knowledge re-
derived from the fields of mathematical logic, psy- presentation. There are several forms of logical repre-
chology, biology, statistics and economics. The sentation. However, the most common are proposi-
first view, derived from mathematical logic, consi- tional logic and predicate logic. Propositional logic
ders intelligent reasoning as formal calculation ba- represents knowledge in the form of statements such
sed on deduction. The second, rooted in psycho- as ‘Socrates is a man’ and ‘Socrates is mortal’, that are
logy, sees reasoning as a characteristic human be- either true or false. Propositions can, of course, be
haviour. The third, derived from biology, is based compound statements linked together with connecti-
on the theory of stimulus/response behaviour. It ves such as AND (&), OR (V), NOT (7), IMPLIES
takes reasoning to be the outcome of the architec- (!), and EQUIVALENT (").
ture of the machinery that accomplishes it. The Propositional calculus has the properties of com-
fourth view, derived from probability theory, pleteness, soundness, and decidability. However, it
considers intelligent reasoning as that which satis- has its limitations. One cannot deal properly with ge-
fies the axioms of probability theory. The fifth neral statements of the form ‘All men are mortal’,
view, derived from economics, defines intelligent e.g., one cannot derive from the conjunction of this
reasoning by applying the tenets of utility theory. and ‘Socrates is a man’ that ‘Socrates is mortal’. To do
iv) It is a medium for pragmatically efficient compu- this, one needs to analyse propositions into predicates
tation, i.e., the computational environment in and arguments and deal explicitly with quantification
which thinking is accomplished. (Jackson, 1986).
Predicate logic provides a formalism for perfor-
v) It is a medium of human expressions, i.e., a lan- ming this type of analysis. The term ‘predicate logic’
guage in which we describe things about the derives from the fact that propositions are analysed
world. As a language, it should be: into predicate argument compositions. It also uses
(a) Easy; quantifiers. These are of two kinds: existential and
(b) Precise; universal. The statement ‘All feathered creatures are
(c) Expressive; and birds’ will be expressed as:
(d) Functional #x [feathered (x) - bird (x)]
These five roles make it clear that knowledge
On the other hand, the statement ‘Some swans are
representation incorporates a theory of intelligent
black’ will be expressed as:
reasoning. It cannot be viewed in purely epistemo-
logical terms. In fact, epistemology and reasoning $x [swan (x) & black (x)], because it is not a
are inextricably intertwined in knowledge repre- universal statement.
https://doi.org/10.5771/0943-7444-2001-1-5
8 Knowl. Org. 28(2001)No.1
Inferences are derived by using principles of modus iii) The Interpreter that decides which rules to apply
ponens and universal specialisations. (Bonnet, 1985).
Predicate calculus works well in case of problems
When certain facts or knowledge about a situation
having an essentially deductive nature. However, the
are presented to the system, the interpreter starts
majority of problems that humans face are of a much
checking the facts against sets of rules. When the “If”
more inductive nature. Such problems require the
portion of a rule is satisfied by the facts, the action
handling of data that are uncertain. Solutions require
specified by the “Then” portion is performed. In the
complex reasoning processes, such as assuming a
process, the facts base may be modified by adding new
hypothesis to be confirmed gradually, making many
facts to the base. The new facts added to the base can
cross-checks to detect error, and so on.
later on be used to form matches with the “If” portion
Ontological commitments of logical systems are
of rules. Matching of rule “If” portions to the facts
based on the belief that logic is a calculus with syntac-
produces inference chains. These inference chains can
tic rules of deduction. In other words, one can say
be displayed to the user to help explain how the sy-
that the meaning, value or outcome associated with an
stem reached its conclusions.
expression in logical systems depends solely upon its
external form and not upon any extraneous associati- The inference engine applies the rules in two im-
ons or ideas that might attach to the symbols in the portant ways: one is called “forward chaining” and the
mind of someone reading or writing them. Thus, lo- other “backward chaining”. If the inference engine
gical systems involve a fairly minimal commitment to looks first at the established data or facts to decide if
viewing the world in terms of individual entities and these satisfy the left side of a rule (the premise), it is
relations between them. said to work in a forward direction. If, on the other
The theory of intelligent reasoning in these systems hand, it looks first at the aims to be attained as given
is derived from mathematical logic, which proceeds on the right side of the rule (the action part) and then
with the assumption that intelligent reasoning is some tries to satisfy only those rules which have these aims,
variety of formal calculation, typically deduction. In it is said to work backwards (Bonnet, 1985). In other
other words, reasoning intelligently means reasoning words, forward chaining is bottom-up or event-driven
in the fashion defined by first order logic. However, a reasoning and backward chaining is goal-directed rea-
great deal of common sense knowledge cannot be ob- soning or top-down reasoning.
tained using formal logic. Humans often resort to ab- One major issue to be sorted out in a production
duction, which is not permissible in formal logic. Ab- rule system is conflict resolution. It arises when seve-
duction is a legitimate and widely practised problem ral rules are applicable to a given situation. In such a
solving technique. It enables one to make progress in case, the interpreter has to decide whether or not to
solving a problem by assuming informed “guesses” apply them in some particular order and whether to
when little is known with certainty. Similarly, apply all that are applicable or only some selection.
another useful technique is induction. This problem is resolved by adopting conflict resolu-
As a language, logical reasoning cannot be said to tion strategies, some of which are:
be simple. However, it possesses the characteristics of
soundness and expressiveness. (a) Performing the first rule;
(b) Sequencing technique: adopting the rules in the
sequence they are;
2.2 Poduction Rule Systems (c) Performing the most specific;
(d) Most recent rule;
Production rules represent knowledge in a situati- (e) Ordering in decreasing order of the strength of the
on- action couple. It means that whenever a certain si- conclusions or of the premises; and
tuation given as the left side of the rule is encounte- (f) Priority to those that have been most frequently
red, the action given on the right is performed. used.
A system based on production rules usually has
Here one is reminded of the Basic Laws enunciated
three components:
by Ranganathan. They also pertain to conflict resolu-
i) The Rule base; tion. They are to be invoked when there is conflict
ii) Facts base, containing the known facts relevant to between two or more laws. The Basic Laws are:
the domain of interest.
i) Laws of Interpretation
https://doi.org/10.5771/0943-7444-2001-1-5
Knowl. Org. 28(2001)No.1 9
ii) Law of Impartiality 2.3 Structured Objects Systems

iii) Law of Symmetry
iv) Law of Parsimony These systems have been developed around the
v) Law of Local Variation concept of structured objects. A structured object is
vi) Law of Osmosis regarded as a prototype, meaning an ideal model of
the object or situation with which objects being stu-
Potentiality of these laws for conflict resolution in died are compared. An object is considered to have a
production rule systems will be known only when number of aspects or attributes. Each attribute can
used in some expert systems. have either default values or possible values. Most re-
As far as ontological commitments are concerned, presentations by structured objects allow the specifi-
rule-based systems view the world in terms of attribu- cation of default values to be used when no other re-
te-object-value triples and the rules of plausible infe- levant information is available and a default value can
rence. Intelligent reasoning process involves a combi- be replaced by an actual value if this becomes known
nation of logic and psychological traditions. Rule ba- at some later stage.
sed systems reflect psychological tradition in the sense The fundamental idea is that the properties of an
that they capture guesses of the type that human ex- object that appear relevant, interesting, and so on, are
perts make, guesses that are not necessarily sound or a function of how one perceives the object and for
true. They reflect logical tradition in the sense that what purpose. As a result, descriptions are viewed as a
propositional and predicate logic are used to derive in- process of comparison in which one specifies a new
ferences. These deductive methods pose problems in entity by saying in what way it is similar to but diffe-
real-life situations where the data as well as inferences, rent from existing objects.
which reflect expert opinion, are usually uncertain. Further, representation by objects allows the two
Thus, several methods that employ measures of the forms of expressing knowledge, i.e., declarative and
uncertainties associated with plausible inferences procedural, to be combined by giving the necessary
(Bonnet, 1985) have been developed to take these un- procedural information with certain attributes of the
certainties into account. One of the methods is objects.
known as the Bayesian approach. The basis of this
method is to assign to each statement a measure of li-
2.3.1 Semantic Networks
kelihood or confidence lying between 1(True) and -
1(False), where O indicates complete uncertainty.
The idea of using semantic networks to represent
Another important method uses the concept of the
knowledge is attributed to Quillian(1969). A semantic
“fuzzy set”. It takes into account the degree of the
network represents knowledge as a pattern of inter-
imprecision of the data rather than the uncertainty of
connected nodes and arcs. The nodes represent either
the data.
concepts, attributes, states, or events and the arcs
As a language, the production rule system is flexi-
show the relationships between the nodes. Many link
ble. It can be easily modified due to its considerable
structures are being used today in semantic nets. Some
degree of modularity. Thus, it can evolve. It is simple
important ones are:
and expressive.
However, production rule systems suffer from a a) “is-a” to represent hierarchy
number of significant disadvantages. First, there is no b) “has” to represent attributes
mechanism for showing context dependency. Second, c) “is in” to represent location
conflict resolution strategies often cause some strange d) “part- of” to represent parts, organs, etc.
side effects in the firing order of the rules. Third, pro- Of all these links, the “is-a” link is the most promi-
duction rule systems fail in reflecting the structure of nent. It generates a hierarchical structure within the
the domain in terms of taxonomic, part-whole, or network. An individual node forms a subset of a ge-
cause-effect relationships between objects and bet- neric node, which forms a subset of another generic
ween classes of objects. Fourth, no discipline is impo- node, and so on. Another property of this link is in-
sed on the ordering of the rules. As a result of these heritance. This means that items lower in the net can
disadvantages, the system tends to become complex, inherit properties from items higher up in the net.
slow, and inefficient as it grows in size. This saves space since information about similar no-
des does not have to be repeated at each node. This
https://doi.org/10.5771/0943-7444-2001-1-5
10 Knowl. Org. 28(2001)No.1
property of inheritance is also applicable to other play the property of inheritance. This is why Minsky
links. has stated that “we can think of frames as a network of
Semantic nets have also been used successfully in nodes and relations” (1975).
Natural Language Processing. Here, a sentence is first Ontological commitments of frames have their
semantically analysed into predicate and arguments. roots in thought on prototypical objects, defaults, and
These form the nodes. Arcs define the relationships hierarchy. The theory of intelligent reasoning in fra-
between the predicate and the arguments (concepts) mes has been derived from psychological traditions.
associated with that predicate. Relationships are re- Minsky (1981) is very clear about this, when he states
presented by specifying the role, such as Agent, Pati- that “this is a partial theory of thinking, ... whenever one
ent, Instrument, Beneficiary, Location, Time etc. encounters a new situation (or makes a substantial chan-
played by an argument in the sentence. ge in one’s viewpoint), he selects from memory a structu-
Ontological commitment is evident in semantic re called a frame; a remembered framework to be adop-
networks to view the world in terms of individual en- ted to fit reality by changing details as necessary”.
tities and relations between them. ‘Hierarchy’ and Reasoning in frames is done by the instantiation
‘Inheritance’ are its other commitments. Intelligent process. This process begins when the given situation
reasoning has its base in the theory of connectionism. is matched with frames already in existence. When a
It is quite simple. All that has to be done is to specify match is found values are assigned to the slots needing
the start node. From the initial node, other nodes are it, resulting in the depiction of a particular situation.
pursued using the links until the final node is reached. Generally, if a given slot characteristic is not present,
However, it lacks the standardisation and formalisati- the slot provides a default value for that characteristic.
on of reasoning process. As a language, it is quite fle- Reasoning process also allows one to move from one
xible. New nodes and links can be defined and added frame to another to match the current situation.
as needed. It also shows a general commitment to As a language, frames are easier to understand.
simplicity and parsimony. They are simple and able to express the situation ade-
quately.
2.3.2 Fames
2.3.3 Scripts
Frames use a relational table approach to knowled-
ge representation. Structured objects form the core of Scripts, which are similar to frames, are another
frames. A frame describes an object by containing all means of knowledge representation. Whereas frames
of the information about that object in “slots”. In one are used to represent all kinds of knowledge, scripts
sense, a frame can be considered a record with fields are used to represent stereotyped events that take pla-
or slots that can be filled with specific values. Marvin ce in day-to-day activities. Some such events are:
Minsky (1975) who originated the frame idea, descri- i) Going to a hotel
bes it as follows: ii) Going to the library
“A frame is a data structure for representing a ste- iii) Going to the bank
reotyped situation, like being in a certain kind of The idea of scripts was introduced by Roger
living room or going to a child’s birthday party. Schank and Robert Abelson (1977). Their main point
Attached to each frame are several kinds of in- is that knowledge of the scripts for many situations is
formation. Some of this information is about how necessary for understanding the way in which the dif-
to use the frame. Some is about what one can ex-
ferent actions comprising an event are interlinked.
pect to happen next. Some is about what to do if
Scripts describe the causal relationships between the
these expectations are not confirmed”.
different episodes and make it possible to draw infe-
Frames are used to represent declarative as well as rences, to guess things that are implied but not stated,
procedural knowledge. Frames containing informati- and, to some extent, to fill in the blank spaces
on about objects are called declarative/factual/situa- (Bonnet).
tional frames. Frames containing slots, which explain Ontological commitments of scripts have their ba-
how to do a thing, are called action-procedures fra- ses in thinking in terms of stereotyped events. Scripts
mes. Like semantic networks, frames, if attached or can only represent stereotyped knowledge. Intelligent
associated one to the other, form a hierarchy and dis- reasoning embedded in scripts has its roots in psycho-
https://doi.org/10.5771/0943-7444-2001-1-5
Knowl. Org. 28(2001)No.1 11
logical traditions. In this sense, scripts are similar to Dependencies or relations are represented by:
frames. The values of the slots are instantiated.
<==> : This means mutual dependence bet-
Scripts, in some respects, have advantages over Fra-
ween two concepts
mes. They allow one to identify preceding and suc-
----> : This means one way dependence bet-
ceeding scenes. An event can be described to the mi-
ween an ACT and a PP or between a
nutest detail. However, knowledge cannot be shared
PP and a PA
across scripts. What is happening in a script is true on-
<== : This means one way dependence bet-
ly for that script. It is difficult to arrange for processes
ween two PPs.
or items of information that are common to a num-
ber of scripts to be shared if these scripts do not form Thus, the conceptual dependency schema for “Ar-
a hierarchy in which, for example, each may represent thur revised his latest book” will be:
a specialisation of the one above (Bonnet).
As a language it is simple and expressive, but limi- Arthur <==> revised <---- book
ted to stereotyped events or situations. (possessed
% &
by)
Latest Arthur
2.3.4 Conceptual Dependency
In other words, we can say that Schank has develo-
Conceptual Dependency, though mainly a theory ped the use of a set of four categories of concepts in a
of natural language processing, is used as a knowledge sentence. These are Objects, Actions, Modifiers of
representation technique for the following reasons: Actions, and Modifiers of Objects. Some of the major
primitive actions are:
a) It enables the development of computer pro-
grammes that can understand natural language Primitive Action Explanation
b) It provides a means of representation which is lan- a) ALTRANS Transfer of abstract relationship
guage independent. (e.g., give)
c) It enables the system to participate in question- b) PTRANS Transfer of the physical location of
answer dialogues. an object (e.g., go)
d) It allows the derivation of inferences from state- c) PROPEL Application of physical force to an
ments. object (e.g., push)
The concept of conceptual dependency was deve- d) MOVE Movement of a body part by its
loped by Schank (1972). The main ideas of this system owner (e.g., kick)
can be summarized as follows (Bonnet): e) GRASP Grasping of an object by an actor
(e.g., throw)
i) Two phrases or sentences having equivalent mea- f) INGEST Ingesting of an object by an animal
nings must have the same internal representation. (e.g., eat)
ii) Every action should be expressed in terms of cer- g) EXPEL Expulsion of something from the
tain primitives. For example, a primitive for body of an animal (e.g., cry)
‘drink’ could be ‘ingest’, which could also be used h) MTRANS Transfer of mental information
for ‘swallow’ and ‘eat’. (e.g., tell)
iii) All the information implied in a phrase or senten- i) MBUILD Building new information out of
ce must be made explicit in the internal represen- old (e.g., decide)
tation. j) SPEAK Producing of sounds (e.g., say)
The meaning of a phrase is represented by a sche- k) ATTEND Focussing of a sense organ toward
ma. This consists of four kinds of nodes or categories: a stimulus (e.g., listen)
a) PP (Picture Production) equivalent to nouns. Similar work has also been done by Wilks (1975)
b) ACT (Action) equivalent to verbs with his theory of preferential semantics. He has de-
c) PA (Picture Aider) equivalent to adjectives (Modi- veloped a set of up to 100 primitive semantic ele-
fiers of PP) ments. He has grouped them into five classes: Entities,
d) AA (Action Aider) equivalent to adverbs (Modi- Actions, Type Indicators, Sorts, and Cases.
fiers of ACT) As far as ontological commitments are concerned,
conceptual dependency believes in semantic primiti-
https://doi.org/10.5771/0943-7444-2001-1-5
12 Knowl. Org. 28(2001)No.1
ves, relations/dependencies among concepts and lan- In this context, Ranganathan’s ideas enshrined in
guage independent representation. Intelligent reaso- his classic work Prolegomena to Library Classification,
ning is derived from psychological traditions. As a as well as in the works of other researchers who have
language, it is simple and expressive. extended his ideas, are quite promising. Epistemologi-
cal foundations for three crucial dimensions of know-
2.3.5 Structured Objects Systems: Evaluation ledge representation, viz., 1) Conceptualisation of
concepts, objects and relations, 2) Classification, and
As a whole, structured objects systems lend them- 3) Inheritance are available in the form of Basic Laws,
selves well to model-fitting approaches. However, Canons, Postulates and Principles.
their representation of hierarchy and inheritance is
not fool-proof, especially when an object tends to in- 3.1. Conceptualisation of Concepts, Objects and Relati-
herit properties from many classes. In addition, the ons:
mechanism of exceptions and defaults is a strong
component of object systems. These systems do not The conceptualisation of concepts, objects and
usually support backtracking. The way inherited pro- their relations involves the application of abstract ca-
perties are cancelled and altered in actual practice, tegories and their formalisation. Broadly speaking,
makes it difficult to properly define one thing in conceptualisation occurs whenever the human mind
terms of another. recognises an entity and, presumably, creates an ab-
straction of that entity which is stored in the memory
as a concept for later use. In the realm of KR, concep-
3. Analytico-Synthetic Paradigm
tualisation involves the identification and analysis of
the several attributes/facets of the ob-
Having analysed the main features of knowledge
jects/entities/concepts constituting the domain in re-
representation languages in the field of artificial intel-
lation to users’ interest and arranging these attribu-
ligence, it will not be out of place to refer to the re-
tes/facets within the domain in some meaningful se-
marks of Peter Jackson (1986) who is of the opinion
quence.
that “knowledge representation languages are really just
Categories are fundamental to all cognitive activi-
high-level programming languages, and that their epi-
ties. Categories can be understood in the sense of ge-
stemological foundations are in fact, quite shallow. A
neral concepts. Eduard Sukiasyan (1998) has aptly de-
more charitable interpretation is that such languages do
provide data and control structures which are more fle- fined a category as “ the most generic notion encompas-
xible than those associated with conventional languages, sing the most universal and essential attributes, proper-
and they are therefore more suited to the simulation of ties, connections and relations of objects and phenomena
human reasoning than anything we had before. These in the real world.” B.C.Vickery (1953) understands
two interpretations are not mutually exclusive, in that “conceptual categories” to mean “concepts of a high
one can hold them both to be true at the same time.” degree of generality with a wide area of application ela-
Thus, it cannot be assumed that contemporary borated by the mind in referring directly or indirectly to
empirical knowledge and utilized by the mind in inter-
knowledge representation issues have somehow been
preting such knowledge.” Thus, categories can be cha-
resolved by the development of highly sophisticated
racterized as “the building blocks of cognition because
tools which, despite the best of efforts, still exhibit
they permit the individual to generalize to new experien-
inherent shortcomings when considered from the epi-
ces the information associated in memory with a particu-
stemological point of view. Further work on the epi-
lar category label”(Jacob,1994).
stemological foundations is, therefore, essential for re-
Categorization, simply put, is the cognitive process
al progress. A knowledge representation language
of dividing the world of human experience into gene-
needs epistemological foundations in the following
ralized groups or broad categories comprising certain
areas:
components sharing immediate similarity in terms of
a) Conceptualisation of Concepts, Objects and Re- attributes within a given context. That this context
lations; may vary, and with it the composition of the catego-
b) Classification; ry, is the very basis for both the flexibility and the
c) Inheritance; power of cognitive categorization. Elin K. Jacob has
d) Intelligent Reasoning and Heuristics; and pointed out that categorization is “the fundamental co-
e) Conflict Resolution. gnitive process of constructing order out of the potential-
https://doi.org/10.5771/0943-7444-2001-1-5
Knowl. Org. 28(2001)No.1 13
ly chaotic environment in which the individual lives by ledge. Ranganathan prefers the term “fundamental
dividing the world of experience into named groups of categories” to describe the category labels represen-
entities whose members bear some relation of similarity ting the relations of commonality between the mem-
to each other.” Without recourse to categorization, the bers of a group of entities, including the particular at-
experience of any one entity would be totally unique, tributes that identify group members as similar enti-
requiring labeling and storage in the memory as a sin- ties. The postulate of fundamental categories which
gular experience identified uniquely by its own set of states that “There are five and only five fundamental ca-
defining characteristics. Behavior based on learning tegories – viz., Time, Space, Energy, Matter, and Perso-
(i.e., the generalization of acquired information) nality” is the basic postulate and all other subsequent
would be impossible. Categorization is thus “the fun- postulates are secondary to it. Ranganathan further
damental cognitive mechanism that simplifies the indi- states that “each facet of any subject, as well as each divi-
vidual’s interaction with the environment by facilitating sion of a facet, is considered as a manifestation of one of
the efficient storage and retrieval of information and the five fundamental categories”. This framework is ba-
thereby reducing the demands for cognitive storage that sed on the assumption that any specific field of know-
would otherwise be placed on human memory.”(Jacob, ledge is formed by the interaction of the five funda-
1994) mental categories in relation to the basic facet. In
Based on the categorical relationships, concepts Ranganathan’s scheme of categories, the “Basic Facet”
constituting the objects of a domain can be hierarchi- is the context-specifying element that provides the
cally structured to express relations between them for environment within which the object/subject is stu-
the purpose of knowledge representation and organi- died. Owing to the difficulty of specifically defining
zation. Categorization may thus be understood to in- the environment, Ranganathan refers the problem to
clude the following processes: a higher level of postulates and suggests that the basic
a) the naming of each possible characteristic of an en- facet (i.e., the context) needs to be postulated. Though
tity/object; the above categories have no claim to any scientific
b) each named characteristic (as an element in relati- justification, and are based entirely on intuition and
on with other elements) forms a component of a belong to the first level of abstraction, they are inte-
structure; gral to the Analytico-Synthetic approach to knowled-
c) Appropriate characteristics are identified and sub- ge classification, representation and organization. The
divided (under a particular category) into varieties concept of categories expounded by Ranganathan, in
which may be ordered (arranged in some meaning- particular, belongs essentially at the highest level of
ful/helpful sequence); and knowledge organization, and thus sets the general di-
d) provision of procedural rules as to the use and rection towards the conception of a universal know-
maintenance of the structure to achieve higher le- ledge representation model.
vels of sophistication.(McIntosh and Griffel, 1969) The most powerful and influential among Ranga-
nathan’s ideas are those relating to facet analysis and
It is in the area of conceptualisation of con- synthesis applicable to real world knowledge structu-
cepts/objects and their relations that Ranganathan has res and their representation. Ranganathan has demon-
made the most fundamental contributions through strated that facet analysis (i.e., breaking down subjects
his postulates and principles concerning categories and into their component parts) and synthesis (i.e., re-
categorisation. To him, KR involves identifying the combining these parts by appropriate connecting
subjects/objects, recognising the nature of relations- symbols to fit the subjects of documents) provide the
hips among the components, and then representing most viable approach to representing the subject con-
them as data structures to be applied in some combi- tents of documents (or for that matter, objects of do-
nations under specific contexts. mains). This powerful and highly flexible method of
The basic approach of Ranganathan is a shift away knowledge organization and representation is known
from typical concepts to categories. To him, the idea as the Analytico-Synthetic (AS) approach. Although
of categories is profoundly related to a very practical the ideas behind this approach have been known to
purpose – that of ordering in a uniform sequence the exist much earlier, their formal enunciation must be
attributes/facets under the various objects/domains. attributed to Ranganathan who systematized and
The term “category” is used by Ranganathan to mean established guidelines for them.
the broadest conceivable classes of phenomena appli- In addition to the postulates concerning categories,
cable for use in the whole, or a large part, of know- Ranganathan also provides specific principles for
https://doi.org/10.5771/0943-7444-2001-1-5
14 Knowl. Org. 28(2001)No.1
achieving an orderly sequence of facets within the ca- res the structuring of each category into a hierarchy,
tegories. These principles for facet sequence, based on subdivided stage by stage (at each level) by the appli-
the analogy of real-life objects and phenomena, seek cation of a series of characteristics/differentiating at-
to determine the helpful sequence of the various facets tributes/sub-facets. It is necessary to group objects in-
occurring in specific domains. In other words, they to classes in order to enable objects in a class to share
seek to achieve an infallible order of objects and their attributes. Each class has its own sets of attributes,
attributes – an area which till today remains a vexed which may be numerous. Out of these, only those re-
issue in KR research. levant attributes are selected. However, the process is
While the AS approach is the outcome of the ef- not as simple as it appears to be. It needs guidance
forts of many researchers from library and informati- from sound theory. Ranganathan’s work in this con-
on science, documentation science, computer science text is fundamental. The following specific Canons
and related fields, the most pathbreaking ideas belong and Principles enunciated by him provide the frame-
to Ranganathan, the chief architect and exponent of work for the classification of knowledge do-
the AS approach. Other scholars who have made si- mains/subjects, concepts/objects, and entities as a step
gnificant contributions to the ideas concerning facets towards their representation:
and categories include Serge Tchakhotine, Z. Dobro-
a) The Canons for Characteristics which seek to re-
wolski, and G. Cordonnier in France; J.E. Holm-
gulate the choice and application of characteri-
strom, E.G. Brisch, J.E.L. Farradane, D.J. Foskett,
stics/attributes in terms of their inherent qualities
B.C. Vickery, Derek Austin, B. Kyle, and others in
which are vital for securing differentiation, rele-
Britain; and Calvin N. Mooers, J.W. Perry, H.P.
vance, ascertainability, and permanence;
Luhn, H.E. Bliss, and Mortimer Taube in the US; and
b) The Canons for Succession of Characteristics
many others. Prominent among present-day scholars
which are prescriptive and aid in determining the
who have further advanced the ideas of facets and ca-
order of succession of characteristics, i.e., attribu-
tegories are I. Dahlberg, Jean Aitchison, Nancy J.
tes chosen for the purpose of classification. The
Williamson, Hemlata Iyer, I.C. McIlwaine, Clare
order so chosen should not result in concomitan-
Beghtol, to name only a few.
ce, but should be relevant to and consistent with
Considerable development of the AS approach has
the intended purpose;
come through two schools of thought: Ranganathan’s
c) The Canons for Array which prescribe certain
group at Bangalore in India and the Classification Re-
conditions be fulfilled in the course of the forma-
search Group in London. Both schools have been ac-
tion, assortment, and arrangement of arrays in a
tive for several decades now though research has re-
sequence which is both helpful and consistent; and
sulted in interesting, yet inevitable, deviations necessi-
d) The Principles for Helpful Sequence which in-
tated by the revolutionary changes in technology and
clude eight guiding principles intended to fulfil the
the fluidity of knowledge structures. Different ap-
demands of the Canon of Helpful Sequence which
proaches can have different implications for categori-
each array of classes should satisfy. These princi-
zation in particular. In the absence of any a priori
ples are based on considerations relating to time
standard method of categorization for achieving a
sequence, evolutionary sequence, spatial contigui-
viable knowledge representation system, surely, the
ty, quantitative measure, complexity, traditional
AS framework expounded by Ranganathan deserves
sequence, literary warrant and, when all else fails,
more serious consideration. Experience has shown,
on alphabetical sequence.
time and again, that Ranganathan’s scheme of catego-
ries provides an adequate and acceptable framework These Canons and Principles provide the basis for
for representing subjects/objects of any degree of the assortment and ordering of the attributes of ob-
complexity, irrespective of domain. jects in some meaningful/helpful sequence.
3.2 Classification 3.3 Inheritance
Subsequent to the initial conceptualisation of con- Following the conceptualisation of concepts, ob-
cepts, objects, their relations and attributes, the next jects, their relations, and their grouping based on spe-
process inevitably involves the grouping of the con- cific attributes; the next logical step involves establis-
cepts/objects/entities/phenomena as a step towards hing hierarchical relationships among them. Inheri-
their systematic representation. Specifically, it requi- tance is achieved by arranging the groups, subgroups,
https://doi.org/10.5771/0943-7444-2001-1-5
Knowl. Org. 28(2001)No.1 15
and isolates in a helpful and filiatory sequence. It in- 4. Conclusion

volves the formation of hierarchies of successive le-
vels. The relationship that emerges is that of superor- Finally, it can be said that the three dimensions re-
dination and subordination expressed as genus-species, lating to the conceptualisation of concepts, objects
whole-part, etc. and their relations, their grouping, and their structu-
In this regard, the Canons for Chain provided by ring form the core of KR. In this context, it is hearte-
Ranganathan prescribe that each chain of classes or of ning to note that Ranganathan’s ideas address the very
ranked isolates in a scheme for classification should sa- foundation of knowledge representation. The authors
tisfy the following two canons: the Canon of Decrea- hope that if present day KR researchers devote more
sing Extension and the Canon of Modulation. These time to ponder over the profundity of the ideas pro-
Canons further seek to achieve the formation of suc- pounded by Ranganathan, then, what was decades ago
cessive levels of hierarchies in an order of decreasing considered a “paradigm shift” in library classification
extension (or increasing intension) without any mis- theory can surely emerge as a potential “paradigm
sing links between the first and the last links. The Ca- set” in knowledge representation. Such is the power
non of Modulation, in particular, implies the concept of Ranganathan’s ideas that they have transcended all
of “Resolving Power” – the power of recognizing the barriers of space, time, technology, and cultures.
classes or the ranked isolates appropriate to the array
of the first order of an immediate universe under con- References
sideration. Within the hierarchy established, Ranga-
nathan further recommends due consideration of the Bonnet, Alain (1985). Artificial intelligence: promise
Canons for Filiatory Sequence which demand that the and performance. London: Prentice Hall Interna-
classes of the universe of subjects (read facets/objects tional, p.108
of categories/domains) should be arranged according Brachman, R. J. and Levesque, H. J.(1985). Readings
to their degree of mutual relationship or affinity reco- in knowledge representation. Los Altos, CA:
gnised as of subordinate or coordinate status. Morgan Kaufman.
Overall, Ranganathan’s AS approach to sub- Davis, Randall; Shrobe, Howard and Szolovits, Peter
ject/object structuring is really a scheme of knowled- (1993). What is knowledge representation? AI Ma-
ge representation. It enables us to create surrogates for gazine, v.14(1), pp.17-33
concrete as well as abstract entities. It is based on a set Dubey, Yogendra P.(1987). Knowledge representation
of ontological commitments provided by postulates and reasoning. In Allen Kent; James Williams
and principles. Theoretical foundations of this onto- (Eds) Encyclopedia of Microcomputers. New
logical commitment are enshrined in Prolegomena. York: Marcel Dekker, v.10, pp.1-18.
The scheme of categorisation of concepts is parsimo- Fodor, J. A.(1975). The language of thought, New
nious, coherent, and corresponds to the syntax in York: Crowell.
which the majority of people arrange ideas in their Jackson, Peter (1986) Introduction to expert systems.
minds. Perhaps, it is more appropriate to say that the New York: Addison-Wesley, p.74.
postulates and principles guiding it conform to the ab- Jacob, Elin K. (1994). Classification and Crossdisci-
solute syntax. As a language, it is precise, expressive plinary Communication: Breaching the Bounda-
and hospitable. As far as intelligent reasoning is con- ries Imposed by Classificatory Structure. In
cerned, it believes in psychological tradition, which Knowledge Organisation and Quality Manage-
takes intelligent reasoning as a particular variety of ment: Proceedings of the Third International
human behaviour. It is based on the assumption that ISKO Conference, 20-24 June 1994, Copenhagen,
human problem solving behaviour could usefully be Denmark. Hanne Albrechtsen and Susanne Oer-
viewed in terms of goals, plans, and other mental nager (Eds.), Frankfurt: Indeks Verlag, p.101-108.
structures. Psychological tradition believes that intel- Liebowitz, J. and Beckman, Tom (1998). Knowledge
ligence is an inherently complex natural phenome- organisations. Boston: St. Lucie Press, p.90.
non. It consists of a variety of mechanisms and phe- McIntosh, Stuart D. and Griffel, David M.
nomena, for which complete and concise descriptions (1969).Computers and Categorisation. In DRTC
may not always be possible. Annual Seminar 7, Bangalore, p.201-220.
Minsky, Marvin (1975). A framework for represen-
ting knowledge. In The Psychology of Computer
https://doi.org/10.5771/0943-7444-2001-1-5
16 Knowl. Org. 28(2001)No.1
Vision, edited by D.H. Winston. Chapter 6. New Schank, Roger and Abelson, Robert (1977). Scripts,
York : McGrawHill. plans, goals and understanding. Hillsdale, NJ:
Minsky, Marvin (1981). A framework for represen- Lawrence Erebaum Assn.
ting knowledge. In Mind Design, edited by J. Star, Susan Leigh (1998). Grounded classification:
Hangeland. Cambridge, Mass: MIT Press, pp.95- grounded theory and faceted classification. Library
128. Trends, v.47(2), p.222.
Minsky, Marvin (1990). Logical vs analogical or sym- Sukiasyan, Eduard (1998). Classification Systems in
bolic vs connectionist or neat vs scruffy. In Artifi- their Historical Development: Problems of Typo-
cial intelligence at MIT: expanding frontiers, edi- logy and Terminology. In Structures and Relati-
ted by P. Winston. Cambridge, Mass: MIT Press. ons in Knowledge Organisation: Proceedings of
Parsaye, K. and Chignell, M.(1988). Expert systems the Fifth International ISKO Conference, 25-29
for experts. New York: John Wiley. August 1998, Lille, France, edited by Widad Mu-
Piaget, J. (1962). Play, dreams and imitation in child- stafa el Hadi, Jacques Maniez and Steven A. Pol-
hood. London: Routledge and Kegan Paul. litt. Wurzburg: Ergon-Verlag, p.72-79.
Pylyshyn, Z. W.(1988). Computation and cogniti- Valente, Andre and Breuker, Joost(1996). Towards
on:Issues in the foundations of cognitive science. principled Core Ontologies. WWW.KSI.CPSC.
Behavioural and Brain Sciences, v.3, pp. 111-69. VCALGARY.CA/KAW/KAW/96/VALENTE/
Quillian, M R. (1969). Semantic memory. In Semantic DOC.HTML.
Information Processing, edited by M. Minsky. Vickery, B.C. (1953). Systematic Subject Indexing.
Cambridge, Mass : MIT Press, pp. 227-270 Journal of Documentation, 9(1), p.54.
Ranganathan, S. R.(1967). Prolegomena to library Wilks, Y. (1975). A preferential pattern seeking se-
classification, 3rd ed. Bombay: Asia Publishing mantics for natural language inference. Artificial
House. Intelligence,v.6(1), pp. 53-74
Schank, R C.(1972). Conceptual dependency: A theo- Winston, P. H. (1984). Artificial intelligence. 2nd ed.
ry of natural language understanding. Cognitive Reading, Mass : Addison-Wesley.
Psychology. v.3.
https://doi.org/10.5771/0943-7444-2001-1-5

Knowledge Representation: Concept, Techniques and The Aanalytico-Synthetic Paradigm

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Knowledge Representation: Concept, Techniques and The Aanalytico-Synthetic Paradigm

Uploaded by

Copyright:

Available Formats

Knowl. Org. 28(2001)No.

Department of Library and Information Science,

0. Introduction intelligence is concerned with the creation of a know-

ii) Law of Impartiality 2.3 Structured Objects Systems

3.2 Classification 3.3 Inheritance

and isolates in a helpful and filiatory sequence. It in- 4. Conclusion

You might also like