Taibah University_CS Dept.

CS 362_ Intelligent Systems

CS 362: Intelligent Systems

Slide 1

Unit 6

Knowledge
Representation
 Taibah University_CS Dept. CS 362_ Intelligent Systems

Knowledge Representation
Slide 2

 Issues in Knowledge Representation

 AI Representational Systems
 Associationist Theories of Meaning
 Semantic Networks
 Standardization of Network Relationships
 Frames
 Conceptual Graphs: a Network Language
 Introduction to Conceptual Graphs
 Types, Individuals, and Names
 Generalization and Specialization
 Propositional Nodes
 Conceptual Graphs and Logic
 Taibah University_CS Dept. CS 362_ Intelligent Systems

Review(i) The Pyramid of Knowledge

Slide 3

 The Pyramid of Knowledge

 Taibah University_CS Dept. CS 362_ Intelligent Systems

Review(ii) Knowledge vs. Expert Systems

Slide 4

 Knowledge is of primary importance in expert systems. In fact,

an analogy to classic expression
 Algorithms + Data Structures = Programs
 Knowledge + Inference = Expert Systems
 Knowledge representation is key to the success of expert
systems.
 Expert systems are designed for knowledge representation
based on rules of logic called inferences.
 Knowledge affects the development, efficiency, speed, and
maintenance of the system.
 Taibah University_CS Dept. CS 362_ Intelligent Systems

Review(iii) How is Knowledge Used?

Slide 5

 Knowledge has many meanings – data, facts, information.

 How do we use knowledge to reach conclusions or solve
problems?
Heuristics refers to using experience to solve problems –
using precedents.
Expert systems may have hundreds / thousands of micro-
precedents to refer to.
 Taibah University_CS Dept. CS 362_ Intelligent Systems

Review(iv) Knowledge in Rule-Based Systems

Slide 6

 Knowledge:
• Knowledge is part of a hierarchy.
• Knowledge refers to rules that are activated by facts or other rules.
• Activated rules produce new facts or conclusions.
• Conclusions are the end-product of inferences when done according
to formal rules.
 Metaknowledge:
• Metaknowledge is knowledge about knowledge and expertise.
• Most successful expert systems are restricted to as small a domain
as possible.
• In an expert system, an ontology is the metaknowledge that
describes everything known about the problem domain.
 Wisdom :
• Wisdom is the metaknowledge of determining the best goals of life
and how to obtain them.
 Taibah University_CS Dept. CS 362_ Intelligent Systems

Review(v) Representation and Intelligence

Slide 7

 The question of representation, or how to best capture critical aspects

of intelligent activity for use on a computer.

Three predominant approaches to representation taken by the AI

research community over this time period.
The first theme( weak method problem-solving)
Articulated in the 1950s and 1960s by Newell and Simon in their work with
the Logic Theorist , is known as weak method problem-solving.
The second (Strong method problem solving )
Common throughout the 1970s and 1980s, and espoused by the early expert
system designers, is strong method problem solving .
In more recent years, (Agent-Based)
Especially in the domains of robotics and the internet (Brooks 1987, 1989;
Clark 1997), the emphasis is on distributed and embodied or agent-based
approaches to intelligence.
 Taibah University_CS Dept. CS 362_ Intelligent Systems

Issues in Knowledge Representation (i)

Slide 8

 The representation of information for use in intelligent problem

solving offers important and difficult challenges that lie at the
core of AI.
 Representation Schemes
 Scheme – data/knowledge structure
 Semantic network
• Frames
 Conceptual graphs
 Stochastic methods
 Connectionist (neural networks)
 Implementation media
 Medium – implementation languages
 Prolog, Lisp, Scheme, even C and Java
 Taibah University_CS Dept. CS 362_ Intelligent Systems

Issues in Knowledge Representation (ii)

Slide 9

 knowledge base is described as: (Bobrow 1975)

• A mapping between the objects and relations in a problem
domain
• The computational objects and relations of a program
 The results of inferences in the knowledge base are assumed to
correspond to the results of actions or observations in the
world.
 The computational objects, relations, and inferences available to
programmers are mediated by the knowledge representation
language.
 Taibah University_CS Dept. CS 362_ Intelligent Systems

Issues in Knowledge Representation (iii)

Slide 10

 There are general principles of knowledge organization that apply

across a variety of domains and can be directly supported by a
representation language.
 Example: class hierarchies are found in both scientific and
commonsense classification systems.
1. How may we provide a general mechanism for representing
them?
2. How may we represent definitions?
3. Exceptions?
4. When should an intelligent system make default assumptions
about missing information and how can it adjust its reasoning
should these assumptions prove wrong?
5. How may we best represent time? Causality? Uncertainty?
Progress in building intelligent systems depends on discovering
the principles of knowledge organization and supporting them
through higher-level representational tools.
 Taibah University_CS Dept. CS 362_ Intelligent Systems

Associationist Theories of Meaning(ii)

Slide 11

 There are many problems that arise in mapping commonsense

reasoning into formal logic.

 For example, it is common to think of the operators ∨ and → as

corresponding to the English “or” and “if ... then ...”.

 However, these operators in logic are concerned solely with

truth values and ignore the fact that the English “if ... then ...”
suggests specific relationship (often more coorelational than
causal) between its premises and its conclusion.
 Taibah University_CS Dept. CS 362_ Intelligent Systems

Associationist Theories of Meaning(iii)

Slide 12

 For example, the sentence “If a bird is a cardinal then it is red”

(associating the bird cardinal with the color red) can be written
in predicate calculus:
∀ X (cardinal(X) → red(X)).

This may be changed, through a series of truth-preserving

operations, Chapter 2, into the logically equivalent expression
∀ X (¬ red(X)→ ¬cardinal(X)).

These two expressions have the same truth value; that is, the
second is true if and only if the first is true.
 Taibah University_CS Dept. CS 362_ Intelligent Systems

Associationist Theories of Meaning(vi)

Slide 13

 Associationist theories:
1. It defines the meaning of an object in terms of a network of
associations with other objects.
2. For the associationist, when humans perceive an object, that
perception is first mapped into a concept.
3. This concept is part of our entire knowledge of the world and
is connected through appropriate relationships to other
concepts.
4. These relationships form an understanding of the properties
and behavior of objects such as snow.
 Taibah University_CS Dept. CS 362_ Intelligent Systems

Semantics of Calculus
Slide 14

 Predicate calculus representation

• Formal representation languages
• Sound and complete inference rules
• Truth-preserving operations
 Meaning(line of reasoning) – Semantics
 Logical implication is a relationship between truth values:
pq

 Associationist theory
 Attach semantics to logical symbols and operators
 Taibah University_CS Dept. CS 362_ Intelligent Systems

Semantics of Calculus(2)
Slide 15

For Example:
“if a bird is cardinal then it is red”
(associating the bird cardinal with the color red)
can be written in predicate calculus:
∀X((cardinal(X) →red(X))
∀X(┐red(X) → ┐cardinal(X))
These two expressions have the same truth value; that is, the
second is true if and only if the first is true.

15
 Taibah University_CS Dept. CS 362_ Intelligent Systems

Semantic Networks(i)
Slide 16

 Using Graphs, by providing a means of explicitly representing

relations using arcs and nodes, have proved formalizing
associationist theories of knowledge.
 A semantic network represents knowledge as a graph, with the
nodes corresponding to facts or concepts and the arcs to
relations or associations between concepts. Both nodes and
links are generally labeled.
 The term “semantic network” encompasses a family of graph-
based representations. These differ chiefly in the names that are
allowed for nodes and links and the inferences that may be
performed. However, a common set of assumptions and
concerns is shared by all network representation languages
 Taibah University_CS Dept. CS 362_ Intelligent Systems

Semantic Networks(ii)
Slide 17

 Semantic network developed by Collins and Quillianin, their

research on human information storage and response times
(Harmon and King 1985).
• classic representation technique for propositional information
• Propositions – a form of declarative knowledge, stating facts
(true/false)
• Propositions are called “atoms” – cannot be further subdivided.
Semantic nets consist of nodes (objects, concepts, situations) and arcs
(relationships between them).
 Taibah University_CS Dept. CS 362_ Intelligent Systems

Semantic Networks(iii)
Slide 18

 Definition
• Represent knowledge as a graph
• Nodes correspond to facts or concepts
• Arcs correspond to relations or associations between concepts
• Nodes and arcs are labeled

 Properties
• Labeled arcs and links
• Inference is to find a path between nodes
• Implement inheritance
• Variations – conceptual graphs
 Taibah University_CS Dept. CS 362_ Intelligent Systems

Semantic Networks Example 1(i)

Slide 19

 Figure 7.1 Semantic network developed by Collins and Quillian in their research on
human information storage and response times (Harmon and King 1985).
 Taibah University_CS Dept. CS 362_ Intelligent Systems

Semantic Networks Example 1(ii)

Slide 20

 Different inferences with

given questions.
 The structure of this
hierarchy was derived from
laboratory testing of
human subjects.
 The subjects were asked
questions about different
properties of birds, such
as, “Is a canary a bird?” or
“Can a canary sing?” or
“Can a canary fly?”.
 Taibah University_CS Dept. CS 362_ Intelligent Systems

Semantic Networks Example 1(iii)

Slide 21

 reaction-time studies indicated that it took longer for subjects to

answer “Can a canary fly?” than it did to answer “Can a canary sing?”
 Collins and Quillian explain this difference in response time by arguing
that people store information at its most abstract level. Instead of
trying to recall that canaries fly, and robins fly, and swallows fly, all
stored with the individual bird, humans remember that canaries are
birds and that birds have (usually) the property of flying.
 Even more general properties such as eating, breathing, and moving
are stored at the “animal” level, and so trying to recall whether a
canary can breathe should take longer than recalling whether a canary
can fly. This is, of course, because the human must travel further up
the hierarchy of memory structures to get the answer.
 Taibah University_CS Dept. CS 362_ Intelligent Systems

Semantic Networks Example 1 (iv)

Slide 22

 The fastest recall was for the traits specific to the bird, say, that
it can sing or is yellow.
 Exception handling also seemed to be done at the most specific
level.
 When subjects were asked whether an ostrich could fly, the
answer was produced faster than when they were asked
whether an ostrich could breathe. Thus the hierarchy
Ostrich → bird → animal seems not to be traversed to get the
exception information: it is stored directly with ostrich. This
knowledge organization has been formalized in inheritance
systems.
 Taibah University_CS Dept. CS 362_ Intelligent Systems

Inheritance systems
Slide 23

 Inheritance systems allow us to store information at the highest level of

abstraction, which reduces the size of knowledge bases and helps prevent update
inconsistencies.
Example:
• If we are building a knowledge base about birds, we can define the traits
common to all birds, such as flying or having feathers, for the general class
bird.
• Allow a particular species of bird to inherit these properties.
• This reduces the size of the knowledge base by requiring us to define these
essential traits only once, rather than requiring their assertion for every
individual.
 Inheritance also helps us to maintain the consistency of the knowledge base when
adding new classes and individuals.
Example:
• Assume that we are adding the species robin to an existing knowledge base.
• When we assert that robin is a subclass of songbird; robin inherits all of the
common properties of both songbirds and birds.
• It is not up to the programmer to remember (or forget!) to add this
information.
 Taibah University_CS Dept. CS 362_ Intelligent Systems

Semantic Network in Natural Language

Understanding(i)
Slide 24

 In the general case, language understanding requires an

understanding of common sense, the ways in which physical
objects behave, the interactions that occur between humans,
and the ways in which human institutions are organized.

 A natural language program must understand intentions, beliefs,

hypothetical reasoning, plans, and goals. Because of these
requirements language understanding has always been a driving
force for research in knowledge representation.
 Taibah University_CS Dept. CS 362_ Intelligent Systems

Semantic Network in Natural Language

Understanding(ii)
Slide 25

 The first computer implementations of semantic networks were

developed in the early1960s for use in machine translation.
 Masterman (1961) defined a set of 100 primitive concept types
and used them to define a dictionary of 15,000 concepts.
 Wilks (1972) continued to build on Masterman’s work in
semantic network-based natural language systems.
 Shapiro’s (1971) MIND program was the first implementation of
a propositional calculus based semantic network.
 Taibah University_CS Dept. CS 362_ Intelligent Systems

Semantic Network in Natural Language

Understanding (iii)
Slide 26

 Quillian’s semantic network (1967)

 Influential program
 Define English in much the same way that a dictionary does: a word
is defined in terms of other words, and the components of the
definition are defined in the same fashion. Rather than formally
defining words in terms of basic axioms
 Each definition leads to other definitions in an unstructured and
sometimes circular fashion
 When look up a word, traverse the network until we are satisfied
that we understand the original word.
 Each node in Quillian’s network corresponded to a word concept,
with associative links to other word concepts that formed its
definition.
 The knowledge base was organized into planes, where each plane
was a graph that defined a single word.
 Taibah University_CS Dept. CS 362_ Intelligent Systems

Three planes representing three definitions

of the word “plant” (i)
Slide 27

 Figure 7.3 below, taken from a paper by Quillian (1967), illustrates three
planes that capture three different definitions of the word “plant:” a living
organism (plant 1), a place where people work (plant 2), and the act of
putting a seed in the ground (plant 3).
 Taibah University_CS Dept. CS 362_ Intelligent Systems

Three planes representing three definitions

of the word “plant” (ii)
Slide 28

1 - A living organism (plant 1),

2- A place where people work (plant 2), and
3- The act of putting a seed in the ground (plant 3).
 Taibah University_CS Dept. CS 362_ Intelligent Systems

Inferences in Semantic Networks

Slide 29

 The program used this knowledge base to find relationships

between pairs of English words.
 Given two words, it would search the graphs outward from each
word in a breadth-first fashion,
 Searching for a common concept or intersection node.
 The paths to this node represented a relationship between the
word concepts.
 Taibah University_CS Dept. CS 362_ Intelligent Systems

Inferences in Semantic Networks Example

Slide 30

 Find the relationship (intersection path) between “cry” and “comfort”

Using this intersection path, the program was able to conclude:

cry 2 is among other things to make a sad sound.
To comfort 3 can be to make 2 something less sad
 Taibah University_CS Dept. CS 362_ Intelligent Systems

Quillian Suggestion
Slide 31

 Quillian (1967) suggested that this approach to semantics might

provide a natural language understanding system with the
ability to:
1. Determine the meaning of a body of English text by building up
collections of these intersection nodes.
2. Choose between multiple meanings of words by finding the
meanings with the shortest intersection path to other words in
the sentence.
Example: it could select a meaning for “plant” in “Tom went home
to water his new plant” based on the intersection of the word
concepts “water” and “plant.”
1. Answer a flexible range of queries based on associations between
word concepts in the queries and concepts in the system.
 Taibah University_CS Dept. CS 362_ Intelligent Systems

Standardization of Network Relationships(i)

Slide 32

 Much of the work in network representations that followed Quillian’s

focused on defining a richer set of link labels (relationships) that
would more fully model the semantics of natural language. By
implementing the fundamental semantic relationships of natural
language as part of the formalism, rather than as part of the domain
knowledge added by the system builder, knowledge bases require less
handcrafting and achieve greater generality and consistency.

 Simmons (1973) addressed this need for standard relationships by

focusing on the case structure of English verbs. In this verb-oriented
approach, based on earlier work by Fillmore (1968), links define the
roles played by nouns and noun phrases in the action of the sentence.
Case relationships include agent, object, instrument, location, and
time.
 Taibah University_CS Dept. CS 362_ Intelligent Systems

Standardization of Network Relationships(ii)

Slide 33

 Sentence is represented as a verb node, with various case links

to nodes representing other participants in the action.
 This structure is called a case frame.
 In parsing a sentence, the program finds the verb and retrieves
the case frame for that verb from its knowledge base.
 It then binds the values of the agent, object, etc., to the
appropriate nodes in the case frame.
Example: Using this approach, the sentence “Sarah fixed the chair
with glue” might be represented by the network in Figure 7.5.
Taibah University_CS Dept. CS 362_ Intelligent Systems

Standardization of Network Relationships(iii)

Slide 34
 Taibah University_CS Dept. CS 362_ Intelligent Systems

Frames
Slide 35

 Framesare another representation technique that evolved from

semantic networks (frames can be thought of as an
implementation of semantic networks).

 But compared to semantic networks, frames are structured and

follow a more object-oriented abstraction with greater structure
 Taibah University_CS Dept. CS 362_ Intelligent Systems

Frames
Slide 36

 Similar to classes in Object-oriented

 Quotes from “A Framework for Representing Knowledge”
[Minsky81]
 Proposed by Minsky in 1975
 Here is the essence of the frame theory: When one encounters a
new situation (or makes a substantial change in one’s view of a
problem) one selects from memory a structure called a “frame”.
This is a remembered framework to be adapted to fit reality by
changing details as necessary.
 Taibah University_CS Dept. CS 362_ Intelligent Systems

Frames – What and Why?

Slide 37

[Minsky, 1981]:
A Frame is a collection of questions to be asked about a hypothetical
situation: it specifies issues to be raised and methods to be used in
dealing with them.

To understand a situation, questions like:

 What caused it (agent)?
 What was the purpose (intention)?
 What are the consequences (effects)?
 Whom does it affect (recipient)?
 How is it done (instruments)
 Taibah University_CS Dept. CS 362_ Intelligent Systems

Frame Example (Hotel Room) (i)

Slide 38

 The hotel room and its components can be described by a

number of individual frames.
 In addition to the bed,
 A frame could represent a chair: expected height is 20 to 40 cm,
number of legs is 4, a default value, is designed for sitting.
 A further frame represents the hotel telephone: this is a
specialization of a regular phone except that billing is through
the room, there is a special hotel operator (default), and a
person is able to use the hotel phone to get meals served in the
room, make outside calls, and to receive other services.
 Taibah University_CS Dept. CS 362_ Intelligent Systems

Frame Example (Hotel Room) (ii)

Slide 39

Each individual
frame may be
seen as a data
structure, similar
in many respects
to the traditional
“record”, that
contains
information
relevant to
stereotyped
entities.
 Taibah University_CS Dept. CS 362_ Intelligent Systems

Frame Slots(i)
Slide 40

 A frame is a set of slots (similar to a set of fields in a class in OO)

 The slots in the frame contain information such as:
1. Frame identification information.
2. Relationship of this frame to other frames.
The “hotel phone” might be a special instance of “phone,”
which might be an instance of a “communication device.”
3. Descriptors of requirements for a frame.
A chair, for instance, has its seat between 20 and 40 cm from
the floor, its back higher than 60 cm, etc. These requirements
may be used to determine when new objects fit the
stereotype defined by the frame.
 Taibah University_CS Dept. CS 362_ Intelligent Systems

Frame Slots(ii)
Slide 41

4. Procedural information on use of the structure described.

An important feature of frames is the ability to attach procedural
instructions to a slot.
5. Frame default information.
These are slot values that are taken to be true when no
evidence to the contrary has been found. For instance, chairs
have four legs, telephones are pushbutton, or hotel beds are
made by the staff.

6. New instance information.

Many frame slots may be left unspecified until given a value
for a particular instance or when they are needed for some
aspect of problem solving.
For example,
the color of the bedspread may be left unspecified.
 Taibah University_CS Dept. CS 362_ Intelligent Systems

Frames -example
Slide 42

 In the following ‘Frame’ example, what piece of knowledge can

be inferred?
A.All birds with wings fly
B.Some birds have two wings
C.Some birds with wings do not fly
D.All birds are brown or dark in color
 Taibah University_CS Dept. CS 362_ Intelligent Systems

Conceptual Graphs: a Network Language

Slide 43

 Graphs
 Introduction to Conceptual Graphs
 Types, Individuals, and Names
 The Type Hierarchy
 Generalization and Specialization
 Propositional Nodes
 Conceptual Graphs and Logic
 Taibah University_CS Dept. CS 362_ Intelligent Systems

Conceptual Graphs: a Network Language

Slide 44

 Following on the early research work in AI that developed

representational schemes a number of network languages were
created to model the semantics of natural language and other
domains.

 In this section, we examine a particular formalism in detail, to

show how, in this situation, the problems of representing
meaning were addressed. John Sowa’s conceptual graphs (Sowa
1984) is an example of a network representation language.

 We define the rules for forming and manipulating conceptual

graphs and the conventions for representing classes, individuals,
and relationships.
 Taibah University_CS Dept. CS 362_ Intelligent Systems

Graphs Review
Slide 45

 Graphs are sometimes called a network or net.

 A graph can have zero or more links between nodes – there is no distinction
between parent and child.
 Sometimes links have weights – weighted graph; or, arrows – directed graph.
 Simple graphs have no loops – links that come back onto the node itself.
 A circuit (cycle) is a path through the graph beginning and ending with the
same node.
 Acyclic graphs have no cycles.
 Connected graphs have links to all the nodes.
 Digraphs are graphs with directed links.
 Lattice is a directed acyclic graph.
Taibah University_CS Dept. CS 362_ Intelligent Systems

Simple Graphs
Slide 46
 Taibah University_CS Dept. CS 362_ Intelligent Systems

Introduction to Conceptual Graphs(i)

Slide 47

Conceptual graph
 A finite, connected, bipartite graph
 No arc labels , (why ? ).
Conceptual graphs do not use labeled arcs; instead the
conceptual relation nodes represent relations between
concepts. Because conceptual graphs are bipartite, concepts
only have arcs to relations, and vice versa.
 Nodes
1. concept nodes – box nodes
Concrete concepts: cat, telephone, classroom, restaurant
Abstract objects: love, beauty, loyalty
2. conceptual relation nodes – ellipse nodes
Relations involving one or more concepts
Arity – number of box nodes linked to
 Taibah University_CS Dept. CS 362_ Intelligent Systems

Introduction to Conceptual Graphs(ii)

Slide 48

 Nodes
1. Concept nodes – box nodes
 Concrete concepts: cat, telephone, classroom, restaurant
(are characterized by our ability to form an image of them in our minds)
 Abstract objects: love, beauty, loyalty
(that do not correspond to images in our minds.)

2. Conceptual relation nodes – ellipse nodes

 Relations involving one or more concepts
 Relation Arity are the number of box nodes linked to
 Taibah University_CS Dept. CS 362_ Intelligent Systems

Introduction to Conceptual Graphs(iii)

Slide 49

 Conceptual relation nodes indicate a relation involving one or

more concepts.
 One advantage of formulating conceptual graphs as bipartite
graphs rather than using labeled arcs is that it simplifies the
representation of relations of any arity.
 A relation of arity n is represented by a conceptual relation node
having n arcs. (as shown in Figure 7.14.).
 Each conceptual graph represents a single proposition. A typical
knowledge base will contain a number of such graphs.
 Graphs may be arbitrarily complex but must be finite.
 Taibah University_CS Dept. CS 362_ Intelligent Systems

Conceptual Graphs Example(i)

Slide 50

 For example, one graph in Figure 7.14 represents the

proposition “A dog has a color of brown”.
 Taibah University_CS Dept. CS 362_ Intelligent Systems

Conceptual Graphs Example(ii)

Slide 51

 For example, one graph in Figure 7.15 is a graph of somewhat

greater complexity that represents the sentence
“Mary gave John the book”.
 This graph uses conceptual relations to represent the cases of
the verb “to give” and indicates the way in which conceptual
graphs are used to model the semantics of natural language.
 Taibah University_CS Dept. CS 362_ Intelligent Systems

Types, Individuals, and Names(i)

Slide 52

 In conceptual graphs, every concept is a unique individual of a

particular type.
 Each concept box is labeled with a type label, which indicates
the class or type of individual represented by that node.
For example:
 A node labeled dog represents some individual of that type.
 Types are organized into a hierarchy. The type dog is a subtype of
carnivore, which is a subtype of mammal, etc.
 Taibah University_CS Dept. CS 362_ Intelligent Systems

Types, Individuals, and Names(ii)

Slide 53

 Boxes with the same type label represent concepts of the same
type; however, these boxes may or may not represent the same
individual concept
 Each concept box is labeled with the names of the type and the
individual.
 The type and individual labels are separated by a colon, “:”.
 Taibah University_CS Dept. CS 362_ Intelligent Systems

Types, Individuals, and Names(iii)

Slide 54

 Type
 A class, a concept
 Types are organized into hierarchy
 Individual -- Concrete entity
 Name – Identifier of type and individual
 Conceptual Graph
 Concept box with type label indicating the class or type of individual
represented by a node
 Label consists of ( type, :, and individual) boy: Ali
 Unnamed individual labeled as marker: #<number> boy: #12354
 Marker can separate an individual from name
Taibah University_CS Dept. CS 362_ Intelligent Systems

Types, Individuals, and Names Example 1

Slide 55
 Taibah University_CS Dept. CS 362_ Intelligent Systems

Types, Individuals, and Names Example 2

Slide 56

 For example:
“Her name was McGill, and she called herself Lil,
but everyone knew her as Nancy”
 Taibah University_CS Dept. CS 362_ Intelligent Systems

Types, Individuals, and Names(iv)

Slide 57

 As an alternative to indicating an individual by its marker or

name,
 we can also use the generic marker * to indicate an unspecified
individual.
 This is often omitted from concept labels; a node given just a
type label, dog, is equivalent to a node labeled dog:*.
 In addition to the generic marker, conceptual graphs allow the
use of named variables.
 These are represented by an asterisk followed by the variable
name (e.g.,*X or *foo).
 This is useful if two separate nodes are to indicate the same, but
unspecified, individual.
 Taibah University_CS Dept. CS 362_ Intelligent Systems

Types, Individuals, and Names Example 3

Slide 58

 For example:
“The dog scratches its ear with its paw”.
Although we do not know which dog is scratching its ear, the
variable *X indicates that the paw and the ear belong to the same
dog that is doing the scratching.
 Taibah University_CS Dept. CS 362_ Intelligent Systems

Generalization and Specialization(i)

Slide 59

 The theory of conceptual graphs includes a number of

operations that create new graphs from existing graphs.
 These allow for the generation of a new graph by either
specializing or generalizing an existing graph, operations that are
important for representing the semantics of natural language.
 The four operations are :
1. Copy
2. Restrict
3. Join
4. Simplify.
 Taibah University_CS Dept. CS 362_ Intelligent Systems

Generalization and Specialization(ii)

Slide 60

1. The copy rule allows us to form a new graph, g, that is the

exact copy of g1.
2. Restrict allows concept nodes in a graph to be replaced by a
node representing their specialization.
 There are two cases:
1. If a concept is labeled with a generic marker, the generic
marker may be replaced by an individual marker.
2. A type label may be replaced by one of its subtypes, if this
is consistent with the referent of the concept. In Figure
7.22 we can replace animal with dog.
 Taibah University_CS Dept. CS 362_ Intelligent Systems

Restriction Example
Slide 61

 For example we can replace animal with dog.

 Taibah University_CS Dept. CS 362_ Intelligent Systems

Generalization and Specialization(iii)

Slide 62

 The join rule lets us combine two graphs into a single graph.
 If there is a concept node c1 in the graph s1 that is identical to a
concept node c2 in s2, then we can form a new graph by
deleting c2 and linking all of the relations incident on c2 to c1.
 Join is a specialization rule, because the resulting graph is less
general than either of its components.
 Taibah University_CS Dept. CS 362_ Intelligent Systems

Join Example
Slide 63

 The Join of g1 and g3

 Taibah University_CS Dept. CS 362_ Intelligent Systems

Generalization and Specialization(iv)

Slide 64

The simplify rule.

 If a graph contains two duplicate relations, then one of them may be
deleted, along with all its arcs. This is the simplify rule.
 Duplicate relations often occur as the result of a join operation, as in
graph g4.
 Taibah University_CS Dept. CS 362_ Intelligent Systems

Propositional Nodes
Slide 65

 In addition to using graphs to define relations between objects

in the world, we may also want to define relations between
propositions. Consider, for example, the statement “Tom
believes that Jane likes pizza”. “Believes” is a relation that takes
a proposition as its argument.
 Conceptual graphs include a concept type, proposition, that
takes a set of conceptual graphs as its referent and allows us to
define relations involving propositions.
 Propositional concepts are indicated as a box that contains
another conceptual graph.
 These proposition concepts may be used with appropriate
relations to represent knowledge about propositions. Figure
7.24 shows the conceptual graph for the above assertion about
Jane, Tom, and pizza.
Taibah University_CS Dept. CS 362_ Intelligent Systems

Propositional Nodes Example(i)

Slide 66
 Taibah University_CS Dept. CS 362_ Intelligent Systems

Propositional Nodes Example(ii)

Slide 67

 The experiencer relation is loosely analogous to the agent

relation in that it links a subject and a verb. The experiencer link
is used with belief states based on the notion that they are
something one experiences rather than does.

 Figure 7.24 shows how conceptual graphs with propositional

nodes may be used to express the modal concepts of knowledge
and belief. Modal logics are concerned with the various ways
propositions are entertained: believed, asserted as possible,
probably or necessarily true, intended as a result of an action, or
counterfactual (Turner 1984).
 Taibah University_CS Dept. CS 362_ Intelligent Systems

Conceptual Graphs and Logic

Slide 68

 Using conceptual graphs, we can easily represent conjunctive

concepts such as “The dog is big and hungry”, but we have not
established any way of representing negation or disjunction. Nor
have we addressed the issue of variable quantification.
 We may implement negation using propositional concepts and a
unary operation called neg. neg takes as argument a proposition
concept and asserts that concept as false. The conceptual graph
of Figure 7.25 uses neg to represent the statement “There are
no pink dogs”.
 Taibah University_CS Dept. CS 362_ Intelligent Systems

Conceptual Graphs and Logic Example

Slide 69

 Conceptual graph of Figure 7.25 uses neg to represent the

statement “There are no pink dogs”.
 Taibah University_CS Dept. CS 362_ Intelligent Systems

Conceptual graphs and predicate calculus(i)

Slide 70

 Conceptual graphs are equivalent to predicate calculus in their

expressive power. As these examples suggest, there is a
straightforward mapping from conceptual graphs into predicate
calculus notation. The algorithm, taken from Sowa (1984), for
changing a conceptual graph, g, into a predicate calculus
expression is:
1. Assign a unique variable, x1, x2, . . . , xn, to each of the n
generic concepts in g.
2. Assign a unique constant to each individual concept in g. This
constant may simply be the name or marker used to indicate
the referent of the concept.
 Taibah University_CS Dept. CS 362_ Intelligent Systems

Conceptual graphs and predicate calculus(ii)

Slide 71

3. Represent each concept node by a unary predicate with the

same name as the type of that node and whose argument is
the variable or constant given that node.
4. Represent each n-ary conceptual relation in g as an n-ary
predicate whose name is the same as the relation. Let each
argument of the predicate be the variable or constant assigned
to the corresponding concept node linked to that relation.
5. Take the conjunction of all atomic sentences formed under 3
and 4. This is the body of the predicate calculus expressions.
All the variables in the expression are existentially quantified.
 Taibah University_CS Dept. CS 362_ Intelligent Systems

Conceptual graphs and predicate calculus Example

Slide 72

 For example, the graph of Figure 7.16 below may be

transformed into the predicate calculus expression

∃ X1 (dog(emma) ∧ color(emma,X1) ∧ brown(X1))