Upload
amey-kerkar
View
282
Download
7
Tags:
Embed Size (px)
Citation preview
WEAK SLOT -AND FILLER -STRUCTURES
Amey D.S.Kerkar,Asst.Professor, Computer Engineering Dept.
Don Bosco College of Engineering,Fatorda-Goa.
Inheritable knowledge• The relational knowledge base determines a set of
attributes and associated values that together describe the objects of knowledge base.
E.g. Player_info(“john”,”6.1”,180,right_throws)• The knowledge about the objects, their attributes and
their values need not be as simple as shown.
• One of the most powerful form of inference mechanisms is property inheritance.
Player Height Weight Bats_throws
John 6.1 180 Right_throws
Sam 5.10 170 right_right
Jack 6.2 215 Bats_throws
• Property Inheritance
• Here elements of specific classes inherit attributes and values from more general classes in which they are included.
• In order to support property inheritance objects must be organized into classes and classes must be arranged in generalization hierarchy.
Here,
Lines ==attributes and boxed nodes== object/values of attributes of an object.
This structure is also called as slot and filler structure. These structures are the devices to support property inheritance along isa and instance links.
Mammal
Person
Owen
Nose
Red Liverpool
isa
instance
has-part
uniform
colour team
person(Owen) instance(Owen, Person)
team(Owen, Liverpool)
Here,
Lines ==attributes and boxed nodes== object/values of attributes of an object.
This structure is also called as slot and filler structure. These structures are the devices to support property inheritance along isa and instance links.
Mammal
Person
Owen
Nose
Red Liverpool
isa
instance
has-part
uniform
colour team
• Advantage of slot and filler structures:
1. monotonic reasoning can be performed more effectively than with pure logic and non monotonic reasoning is easily supported.
2. Makes it easy to describe properties of relations.
e.g. “does Owen has-part called nose?”
3. Form of object oriented programming and has advantages such as modularity and ease of viewing by people.
Slot and filler structures
Attribute= slot and its value= filler
Weak slot and filler structure Strong slot and filler structure
FramesSemantic nets ScriptsConceptual Dependency
Weak slot and filler structures: are “Knowledge- Poor” or “weak” as very little importance is given to the specific knowledge the structure should contain.
Semantic nets• In semantic nets information is represented as:
– set of nodes connected to each other by a set of labelled arcs.
• Nodes represent: various objects / values of the attributes of object .
• Arcs represent: relationships among nodes.Mammal
Person
Jack
Nose
Blue Chicago Royals
isa
instance
has-part
uniform
color team
• In this network we could use inheritance to derive the additional info:
has_part(jack, nose)
Intersection Search
One way to find relationships among objects is to spread the activation(links) out from two nodes and find out where it meets
Ex: relation between :
Red and liverpoolMammal
Person
Owen
Nose
Red Liverpool
isa
instance
has-part
uniform
color team
• Representing non binary predicates:
1. Unary –
e.x. Man(marcus) can be converted into:
instance(marcus,Man)
2. Other arities-
e.x. Score(india,australia,4-1)
3 or more place predicates can be converted to binary form as follows:
1. Create new object representing the entire predicate.
2. Introduce binary predicates to describe relation to this new object.
TEST4
isa
Game
Australia
India
4-1
Visiting _team Score
Home _team
Score(india,australia,4-1)
Ex. 2. “john gave the book to Mary”
give(john,mary,book)
EV 1
instance
Give
John
Mary
Book
BK1
instance
Objectagent
beneficiary
Making some important distinctions
1. “john has height 72”
2. “john is taller than Bill”
John 72
height
John Bill
H1 H2
height height
72
Value
greater_than
Partitioned semantic nets• Used to represent quantified expressions in
semantic nets.
• One way to do this is to partition the semantic net into a hierarchical set of spaces each of which corresponds to the scope of one or more variable.
• “the dog bit the mail carrier” [partitioning not required]
d
Dogs
b
Bite
m
Mail-Carrier
isa isa isa
assailant victim
• “every dog has bitten a mail carrier”
x: dog (x) y: mail-carrier(y) bite(x, y)
• How to represent universal quantifiers?
– Let node ‘g’ stands for assertion given above
– This node is an instance of a special class ‘GS’ of general statements about the world.
– Every element in ‘GS’ has 2 attributes:-
• Form - states relation that is being asserted.
• connections - one or more, one for each of the universally quantified variables.
– ‘SA’ is the space of partitioned sementic net.
• “every dog has bitten a mail carrier”
SA
S1
d
Dogs
b
Bite
m
Mail-Carrier
isa isa isa
assailant victimg
GS
isaform
• “Every dog in the town has bitten the constable”
SA
m
Constables
isaS1
d
Dogs
b
Bite
isa isa
assailant
g
GS
isa
form
victim
• “Every dog in the town has bitten every constable”
SAConstables
S1
d
Dogs
b
Bite
isa isa
assailant
gGS
isa
form
victimc
isa
• More examples of sementic nets:
• “ Mary gave the green flowered vase to her favourite cousin”
EV 1
instance
Give
Mary
cousin
vaseObjectagent
beneficiaryColour_pattern
Green flowered
favourite
• “every batsman hits a ball”
SA
S1
b
Batsman Hits
b
Balls
isa isa isa
action Acts_ong
GS
isaform
h
• “Tweety is a kind of bird who can fly. It is Yellow in colour and has wings.”
Bird
Tweety
Wings
instance
has-partyellow
fly
colour
action
• Represent following using sementic nets:-
Tom is a cat. Tom caught a bird. Tom is owned by John. Tom is ginger in color. Cats like cream.The cat sat on the mat. Acat is a mammal. Abird is an animal. All mammals are animals.mammals have fur.
Frames• Another kind of week slot and filler structure.
• Frame is a collection of attributes called as slots and associated values that describe some entity in the world (filler).
• Consider,Room
Hotel room
isa
Hotel bed
contains
Hotel Chair
contains
Location
Chair
Sitting_on
4
20-40 cmsisa
use
legs
height
Room No 2
instance
Hotel Room
isa : Room
contains: Hotel Bed
contains: Hotel Chair
Hotel Chair
isa: Chair
use: sitting_on
location: Hotel Room
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
Frame System for Hotel Room
Frame structure for Hotel Room
Frame structure for Hotel Chair
Frame structure for all remaining attributes
Person Jack_Roberts
isa: Mammal instance: Fielder
cardinality: 6,000,000,000 height: 5-10
* Handed: right balls: right
batting_avg: 0.309
Adult__Male team: Chicago cubs
isa: Person uniform_color: blue
Cardinality: 2,000,000,000 Fielder
* Height: 5- 10 isa: ML_Baseball_Player
cardinality: 376
ML_Baseball_Player batting_avg: 0.262
isa: Adult_Male
cardinality: 624 ML_Baseball_Team
* height: 6-1 isa: Team
* bats: equal to handed cardinality: 26
* batting-avg: 0.252 team_size: 24
* team: manager:
*uniform_color:
Individualframe
• Meta Class: special class whose elements themselves are classes.
– If X is meta class and Y is another class which is an element of X, then Y inherits all the attributes of X.
• Other ways of relating classes to each other
1. Mutually disjoint: 2 classes are mutually disjoint if they are guaranteed to have no elements in common.
2. Is covered by: relationship is called as ‘covered-by’ when we have a class and it has set of subclasses, the union of which is equal to the superclass.
ML_Baseball_Player
isa
Fielder Pitcher
isa
Catcher
isa
American Leaguer
isa
National Leaguer
isa
Jackinstanceinstance
ML_Baseball_Player
is covered-by: { Pitcher, Catcher, Fielder,American leaguer, National leaguer}
Pitcher
isa: ML_Baseball_Player
mutually_disjoint-with: {Catcher, Fielder }
Catcher
isa: ML_Baseball_Player
mutually_disjoint-with: {Pitcher, Fielder }
Fielder
isa: ML_Baseball_Player
mutually_disjoint-with: {Pitcher, Catcher}
.
.
.
.
Tangled Hierarchies
• Hierarchies that are not trees
• Usually hierarchy is an arbitrary directed acyclic graph.
• Tangled hierarchies requires new property inheritance algorithm.
•FIGURE A
• Can fifi fly?
• The correct answer must be ‘no’.
– Although birds in general can fly, the subset of birds , ostriches does not.
– Although class pet bird provides path from fifi to bird and thus to the answer that fifican fly, it provides no info that conflicts with the special case knowledge associated with class ostrich, so it should hove no effect on the answer.
isa
isaisa
isa
Ostrichfly :no
fififly : ?
Birdfly :yes
Pet-Bird
FIGURE B• Is Jack Pacifist?
– Ambiguity
• One way to solve ambiguity is to base the new inheritance algo based on path length:– Using BFS start with the frame for which slot value is needed.
– Follow its instance links, then follow isa links upwards .
– If the path produces a value it can be terminated, as can all other paths once their length exceeds that of the successful path.
instanceinstance JackPacifist : ?
Quakerpacifist: true
Republicanpacifist: False
• Using this technique our answers to fifi problem is :’no’ and for jack problem we get 2 values hence ‘contradiction’.
• Now consider following hierarchies:FIGURE C
• Our new algo gives answer: fifi can fly. i.e. Fly: yes.
isa
instanceinstance
isa
White-PlumedOstrich
fififly : ?
Birdfly :yes
Pet-Bird
Plumed Ostrich
isa
Ostrichfly: no
isa
FIGURE D
• Here our new algo reaches Quaker and deduces pacifist:true and stops without noticing further contradiction.
instanceinstanceJack
Pacifist : ?
Quakerpacifist: true
ConservativeRepublican
isa
Republicanpacifist: false
• Solution to the problem is to base our algo not based on path length but on inferential distance.
• Class 1 is closer to class2 than class 3 if class1 has an inferential path through class2 to class3.
• For figure A answer is no
• For figure B Contradiction
PLEASE REFER PROPERTY INHERITANCE ALGO
Rich and Knight pg. 204-205 3rd edition.
Class 2Class 1 Class 3