Knowledge Representation

1
Knowledge Representation

• Structured Objects

2

• Structural knowledge is important
• e.g. viral meningitis is meningitis
• Representation is analogous to graphs or records.
• Motivation
• Grouping of knowledge intuitively
• All knowledge about an entity is stored together
• Having intuitive access paths
• Inferencing via
• Generalisations of situations
• Properties and interrelationships
• Inheritance of properties
• Same benefits of object-oriented design and
  programming e.g. Saves on storage

3

• Structured Objects can be grouped into
• Semantic Networks
• Frames

4
Semantic Networks

• Concentrate on categories of objects and the
  relations between them

BIRTHDAY-PARTY
ELEMENT-OF
JOHNS-B-P
food
date
CAKE
AUGUST-3
guest
place
JOHNS-B-P
MARY
5

• LinkType Semantics Example
• A B A ? B
  Cats ? Mammals
• A B A ? B
  Bill ? Cats
• A B R(A,B)
  Bill 12
• A B ?x x ? A ? R(x,B)
  Birds 2
• A B ?x ? y x ? A ?
  Birds Birds
• y ? B ? R(x,y)

Subset
Member
R
Age
R
Legs
Parent
R
6
Example
Person
member
member (is-a)
profession
Mary
lecturer
married-to
Joe
profession
engineer
lives-in
lives-in
is-a
Kingston
city
7
Characteristics of Semantic Nets

• Limited Expressiveness
• Cannot express negation or disjunction
• Quantification Complex, using partitioned nets
• Simple and easy to understand.
• Syntax is not clear and consistent.
• Semantics are intuitive and dependent on
  implementation
• Inheritance is captured
• Inferencing
• Intersection Search
• e.g. What is the relationship between Joe and
  Mary?

8
Frames

• A data structure representing a stereotype
  situation.
• The pulling together of procedural and
  declarative knowledge.
• Has slot names and slot fillers
• Usually arranged in a hierarchy
• Frames lower down inherit slot fillers from
  frames higher up.
• Properties high up are fixed
• Properties with values lower down overwrite
  information higher up

9
Animals
Alive
T
Flies
F
Subset
Subset
Birds
Mammals
Legs
2
Legs
4
Flies
T
Subset
Subset
Subset
Penguins
Cats
Bats
Legs
2
Flies
F
Flies
T
Member
Member
Member
Opus
Bill
Pat
Name Opus
Name Bill
Name Pat
Friend
Friend
10
Types of Slots

• Attribute-value slots
• Primitive data types
• ltname Jumbogt
• Pointers to other frames
• ltowner e56gt
• Attribute slots with value restrictions
• ltowner (a person)gt
• ltmother (an elephant with ltsex femalegt) gt
• Object hierarchy slots
• super-class/subclass slots
• member-of/instance slots
• Procedure slots
• Used for calculations
• instead of storing the value e.g. salary
• Used to propagate changes when a slot value is
  changed

11

• Example
• lt e1 ltmember-of american-person
• dog-ownergt
• ltname Mickey Mousegt
• ltaddress Disneylandgt
• ltowns e3gt
• ltpersonality unpleasantgt gt
• lt e3 ltmember-of doggt
• ltname Plutogt
• ltowned-by e1gt gt
• lt dog-owner ltsuperclass persongt
• ltowns (a dog)
  gt
• ltmust-have (a
  dog-licence) gt
• ltpersonality pleasantgt
  gt
• ltdog ltsuperclass pet

12
Reasoning in Frame Systems

• Matching
• Find a matching frame
• Difficult because an instance frame may have
  values from several class frames.
• Potential inefficiency
• Inheritance
• If the value of the slot is not found in the
  instance frame, search up the hierarchy.
• E.g. Does Bill fly? No
• Allows for Default Reasoning
• e.g. Does Opus fly?
• No, since instance value overrides class value.

13

• Allowing multiple inheritance can cause conflicts
  in reasoning.
• Example
• Does Tweety fly?

Bird Move Fly
Ostrich Move Walk, Not fly
Cartoon Bird
Tweety

• Depends on the path taken for the search.
• Strategy 1 Use path length (BFS)
• Tweety does not fly.

14

• Problem with BFS Does Tweety fly?

Bird Move Fly
IS A
IS A
Ostrich Move Walk, Not fly
Cartoon Bird
IS A
Plumed Ostrich
IS A
White-Plumed Ostrich
instance
instance
Tweety

• Strategy 2 Use Inferential Distance
• Tweety does not fly.

15

• Can still get a contradiction

Republican Pacifist false
IS A
Quaker Pacifist true
Conservative-Republican
Instance
Instance
Dick Pacifist ?
16
Advantages

• More knowledge about the nature of the entities
  involved.
• More than logic or production rules
• Can represent highly structured knowledge
• Easy
• To maintain
• To add new objects
• Default Reasoning Possible
• The drawing of plausible inferences on the basis
  of less than conclusive evidence in the absence
  of evidence to the contrary. (Moore 1985)

17
Disadvantages

• Precise notion of meaning is absent
• Translation work has been done, but has
  difficulties with default reasoning and
  procedural attachment.
• Hard to represent rules (p,q ? r)

18
Problems

• Draw a semantic net for the following
  information Tweety is a canary who is a bird and
  all birds are animals. All animals breathe,
  typically birds have wings and their method of
  travel is by flying. A penguin is a bird whose
  method of travel is by walking. Discus how the
  queries How does Tweety travel?, Does Tweety
  have wings? and What is the link between
  Tweety and Penguins? will be executed.

19

• Birds are usually, covered with feathers,
  fly and reproduce by laying eggs. There are three
  groups of birds, flightless birds, songbirds and
  scavengers. Flightless birds do not fly. Song
  birds eat bugs and seeds while scavengers eat
  meat. Sparrows and canaries are both flightless
  birds, the difference between them being that
  canaries are found in tropical countries, while
  sparrows are found in North America. Penguins are
  flightless birds they eat fish and are found in
  the South Pole. Opus, Tweety and Beaky are all
  birds Opus is a penguin, Tweety is a canary and
  Beaky is a rather unusual bird as he is a mix
  between a penguin and a canary.
• Represent the information above using
  frames. What would be the results of the
  following queries?
• How does Opus reproduce?
• Does Beaky fly?