Artificial Intelligence

Module-V:
Structured Representations of Knowledge

Dr. Dwiti Krishna Bebarta

Text Book Referred:

1. Artificial Intelligence, Elaine Rich and Kevin Knight, Tata McGraw -Hill Publications
2. Introduction To Artificial Intelligence & Expert Systems, Patterson, PHI publications
 Structured Representations of Knowledge
Topics to discuss:
• Semantic Nets
• Representing non-binary predicates
• Partitioned Semantic Nets
• Frames as sets and instances
• Slots as full-fledged Objects
• Property Inheritance through tangled hierarchies
• Conceptual Dependency
• Conceptual Dependency Graphs
• Scripts
• Examples in natural language understanding
• Merits and demerits of strong slot filler structures
• Artificial intelligence is a system that is concerned with
the study of understanding, designing and implementing the
ways, associated with knowledge representation to
computers.
• In any intelligent system, representing the knowledge is
supposed to be an important technique to encode the
knowledge.
• The main objective of AI system is to design the programs
that provide information to the computer, which can be
helpful to interact with humans and solve problems in
various fields which require human intelligence.
What is Knowledge?
• Knowledge is an useful term to judge the understanding of
an individual on a given subject.
• In intelligent systems, domain is the main focused subject
area. So, the system specifically focuses on acquiring the
domain knowledge.
Any knowledge representation system should possess
properties such as
• Learning: capability of acquiring new knowledge, behaviors,
understanding
• Efficiency in acquisition: ability of acquiring new knowledge
using automatic methods
• Representational adequacy: ability of representing the
required knowledge
• Inferential adequacy: ability of manipulating knowledge to
produce new knowledge from the existing one
Approaches to knowledge representation
• Relational Knowledge
– Consisting of attributes and associated values
– Each fact is stored in a row of a relational table
– A table is defined as a set of data values organized using
a model of rows and columns
– Example:
• John is a male whose age is 38 years and who is a
graduate with a salary of Rs 80000

Name Age Gender Qualification salary

John 38 Male Post Graduate 80000
James 29 Male PhD 100000
Mike 30 Male Post Graduate 75000
• What is age of John?
• How much salary James earn?
• What is the qualification of Mikes?
• Does a person having a PhD qualification earn more?
Not possible since Inferencing new knowledge is not possible
from this structure
Knowledge represented as Logic
• “every man is mortal” can be represented as
( x) (MAN(x) → MORTAL(x))
If the Fact is John is a man then John is Mortal can easily infer
(MAN(John) → MORTAL(John))
Not possible in relational approach

Knowledge represented as Logic

• Encoded in the form of procedures which carry out specific
tasks based on relevant knowledge
• An interpreter for a PL interprets the program on the basis of
syntax and semantics of the language
• Problem is with completeness and consistency
Semantic Networks:
• Semantic networks (semantic net) are a commonly used
representation in AI.
• In Semantic net, we can represent our knowledge in the
form of graphical networks.
• A semantic network is a graph consisting of nodes that are
connected by edges.
• Nodes represent objects, and the links between nodes
represent relationships between those objects.
• The links are usually labelled to indicate the nature of the
relationship.
• This representation consists of mainly two types of
relations:
– IS-A relation (Inheritance)
– Kind-of-relation
Semantic Relations
Isa:
• Connects two concepts where one concept is a subclass
of the other
• Example: Man is human i.e. Man is a subclass of the
Human
Inst:
• Relates specific members of a concept
• Example: John is a Man i.e. John is an instance of Man
Property Relation
• Represented by dotted lines pointing from concept to its
properties
• Used to represent the properties of every concept such
as {can, has, colour, height, etc}
• Example: living_things can eat and breathe, Animal has
skin and so on
Representing non-binary predicates
• We can represent the generic give event as a relation
involving three things:
– A giver
– A recipient
– An object.
– An Action
John gives an apple to Mike
Object(E, apple)
Action(E, give)
Actor(E, John), recipient(E, Mike)
Isa(John, Human), Isa(Mike, Human)
In FOPL, predicate relation can have n arguments
Give(John, Mike, apple) 3-array relationship

E is an event of many clauses

Action
Advantages of Semantic nets
• Easy to visualize
• Formal definitions of semantic networks have been
developed.
• Related knowledge is easily clustered.
• Efficient in space requirements
– Objects represented only once
– Relationships handled by pointers
Disadvantages of Semantic nets
• Inheritance (particularly from multiple sources and when
exceptions in inheritance are wanted) can cause problems.
• Facts placed inappropriately cause problems.
• No standards about node and arc values
Partitioned Semantic Nets:
• Hendrix developed the Partitioned Semantic Nets to
represent the difference between the description of an
individual object or the process and the description
(quantification) of a set of objects.
• Divided into one or more networks for description of an
individual i.e. portioning is to allow groups, nodes and arcs
to be bundled together into units called spaces.
• Partitioned semantic nets are a way to group nodes and
links to enhance the expressive and organizational power of
semantic nets. This grouping allows for the distinction
between general statements and specifics.
Statement: Danny the
dog bitten the postman
Peter.

This can be
represented in ordinary
semantic Net.
Knowledge representation using frames
• Introduced by Marvin Minsky (1975)
• Extension to semantic nets
• Each node is represented by a frame
• Frame is data structure use to represent a stereotyped situation
• Used for representing a stereotyped situation
• It consists of a collection of attributes or slots and associated
values that describe some real-world entity
• Similar to the class of OOP
• The related frames are linked together into frame systems and
are organized into hierarchies or network of frames
• Each frame in the network is either a class frame or an instance
frame
• Frame consists of attributes/slots i.e. Described as <slot-name,
value>
• Slots have facets/fillers which describe their properties
 Structure of a Frame

List of Facets in a Frame

 Hospital Frame
Slot name Facet name Facet value
F_name Value Metro Hospital
Country Default Value India
Phone_no Constraint on values 1234567856

Frames in a network of frames are connected using links

given below
1. Ako (A kind of): to two class frames, a class can define its
own slots and also inherits slot-value pairs from its super
class. It gives a sub-typing hierarchy where all instances of
class frame are instances of super class frames.
2. Inst: connects a particular instance frame to a class frame i,e
AIIMS is an instance the class frame hospital
3. Part_of: connects two class frames one of which is contained
in the other class i.e. Ward is part_of the class hospital
Hospital Frame
• F_name hospital
• Country (default value-India)
• Phone_no (Constraint on values (10 digit no.) 3959905901)
• Address (default-New Delhi)
• Director (default-xyz)
• Labs lab(lab frame)
• Wards ward(ward frame)
• Doctors doctor(doctor frame)
Slots as full-fledged Objects:
• Slots as full-fledged objects, also known as frames, are a
knowledge representation system in artificial intelligence
(AI) that uses attribute value pairs and fillers to represent
knowledge, concepts, or entities:
– Slot: An attribute value pair
– Filler: A value that a slot can take, such as a numeric,
string value, or a pointer to another slot.
Slot and filler systems have several advantages, including:
• Quick retrieval: Attribute values can be retrieved quickly
• Easy to describe properties: Properties of relations are easy
to describe
• Monotonic reasoning: Monotonic reasoning can be
performed more effectively than with pure logic
• Non-monotonic reasoning: Non-monotonic reasoning is
easily supported
• Modularity: A form of object oriented programming with
modularity
• Ease of viewing: Easy to view by people .
Property Inheritance through tangled hierarchies:
Tangled hierarchies
• Hierarchies that are not trees
• Usually hierarchy is an arbitrary directed acyclic graph
(DAG)
• Tangled hierarchies requires new property inheritance
algorithm
Exceeds that of the successful path.
Weak slot and filler structure
• A weak slot and filler structure is a data structure in artificial
intelligence (AI) that uses slots to represent attributes and
fillers to represent values. It's used to represent knowledge
and has several advantages, including:
– Quick attribute value retrieval
– Indexed assertions and binary predicates
– Easy description of relation properties
– Effective monotonic reasoning
– Easy support for non-monotonic reasoning

Frames are a type of weak structure where each frame

contains slots and associated values that describe an entity.
Strong slot and filler structure
In artificial intelligence (AI), a strong slot and filler structure is a
way to represent knowledge about objects and their
relationships using attributes (slots) and their associated values
(fillers):
• Rigid rules
• Object types and relations
• Common situations
– Strong slot and filler structures represent knowledge about common
situations.
– In contrast, weak slot and filler structures provide little domain
knowledge. Frames are a type of weak structure where a frame
contains slots describing an entity.
Conceptual Dependency Theory
• Conceptual Dependency theory is based on the use of
limited number of primitive concepts and rules of
formation to represent any natural language statement.
• Sentences are represented as a series of diagrams
depicting actions using both abstract and real physical
situations.
– The agent and the objects are represented.
– The actions are built up from a set of primitive acts
which can be modified by tense.
– Building blocks – entities, actions, conceptual cases,
conceptual dependencies and conceptual tenses.
Entities
• Picture producers (PP) – actors/physical objects including
human memory that perform different acts.
• Picture aiders (PA) – supporting properties/attributes of
producers.
Actions
• Primitive actions (ACTs) – actions.
• Action aiders (PA) – supporting properties/attributes of
primitive actions.
Conceptual Primitive Actions
Actions in NL generally specified by the verbs
Most of the verbs have been categorized as given below.
Conceptual Primitive Actions
• ATRANS: give, take, purchase (two way ATRANS)
• PTRANS: go or walk, fly
• PROPEL: push, pull, throw
• MOVE: kick, throw, hit
• GRASP: catch, clutch
• INGEST: eat, drink, smoke (actor, object, direction)
• MTRANS: tell, read, speak, sing (recipient, object, actor)
• MBUILD: decide, describe, imagine, consider, answer
• EXPEL: weep, cry, sweat
• SPEAK: say, tell, sing
• ATTEND: listen, watch, see, look
Conceptual cases
All actions involve one or more of these
• Objective case
• Directive case
• Instrumental case
• Recipient case
Conceptual dependencies
• Semantic rules to form dependency structure
• Relationship between an actor and an event
• Relationship between primitive action and an instrument
⇔ two-way dependency
One-way relation as per direction
Two way relation i.e. one belongs to the set defined by the other
Conceptual tense
Rule-1: represents relation between an actor and the event

ATRANS
Conceptual Dependency:
3. PP PP one object belongs to the set of other object
1. John is a doctor John doctor
4. PP PP (one object provides information about other
object)
1. John’s car John pos by Car
2. Mary is in Delhi Mary loc Delhi
5. PP ⇔ PA (to represent state/modifier of an object)
1. John is fat John ⇔ weight > 50
2. Mary is poor Mary ⇔ income<2000
6. PP PA (relation between an object and an attribute
defined)
– Smart John John smart
– Charming Mary Mary charming
 Rule-7: it shows a relationship between an ACT and its
physical source and destination locations of ACT. R indicates
relation between source and destination
ACT

John went to school from home

Mary travelled from Delhi to Mumbai by train

p School
John ⇔ PTRANS
Home

p Mumbai
Mary ⇔ PTRANS
v Delhi
Train
Mike ate custard with spoon

p i
Mary ⇔ INGEST Spoon
o
Custard
INGEST: eat, drink, smoke (actor, object, direction)
SCRIPTS
Frame like structures used to represent commonly occurring
experiences.
 Example
– Going to movie
– Shopping in a supermarket
– Eating in a restaurant
– Visiting a dentist
 Components
– Actors
– Roles
– Props
– Scenes
A Super Market Script Structure

• Write a SCRIPT
structure for a
Restaurant
• Write a SCRIPT
structure for visiting
a doctor in a hospital
• Write a SCRIPT
structure for going to
watch a movie
Weak slot filler structures Strong slot filler structures
Weak slot filler structures Strong slot filler structures are very
are very general rigid rules
Individual semantic networks and exhibit powerful theories of how AI
frame systems may have specialized programs can represent and use
links and inference procedures, but knowledge about common situations.
there are no hard and fast rules
about what kinds of objects are good
to use for Knowledge Representation
Frames are a type of weak structure Conceptual Dependency, Scripts
where each frame contains slots and which tells what type of objects and
associated values that describe an relations are permitted
entity.

Semantic Nets, which provide only a provides both a structure and a

structure into which nodes specific set of primitives, at a
representing information at any level particular level of
can be placed granularity(modules), out of which
representations of particular pieces
of information can be constructed.