# Rajendra Akerkar

Vestlandsforsking, Norway
10:58:57 1 R. Akerkar: Reasoning in DL
 What is Description Logics (DL) p g ( )
 Semantics of DL
 Basic Tableau Algorithm
10:58:57 2 R. Akerkar: Reasoning in DL
 A formal logic-based knowledge g g
representation language
◦ “Description" about the world in terms of concepts
( l ) l ( ti l ti hi ) d (classes), roles (properties, relationships) and
individuals (instances)
 Decidable fragments of FOL g
 Widely used in database (e.g., DL CLASSIC)
and semantic web (e.g., OWL language)
10:58:57 3 R. Akerkar: Reasoning in DL
Person include Man(Male) and
Woman(Female) Woman(Female),
A Man is not a Woman
A Father is a Man who has Child A Father is a Man who has Child
A Mother is a Woman who has Child
Both Father and Mother are Parent Both Father and Mother are Parent
Grandmother is a Mother of a Parent
A Wife is a Woman and has a Husband( A Wife is a Woman and has a Husband(
which as Man)
A Mother Without Daughter is a Mother g
whose all Child(ren) are not Women
10:58:57 4 R. Akerkar: Reasoning in DL
10:58:57 5 R. Akerkar: Reasoning in DL
 Concepts (unary predicates/formulae with one free variable)
E P F th M th ◦ E.g., Person, Father, Mother
 Roles (binary predicates/formulae with two free variables)
◦ E.g., hasChild, hasHudband
 Individual names (constants)  Individual names (constants)
◦ E.g., Alice, Bob, Cindy
 Subsumption (relations between concepts)
◦ E.g. Female _ Person
 Operators (for forming concepts and roles)
◦ And(Π) , Or(U), Not (¬)
◦ Universal qualifier (¬), Existent qualifier(-)
◦ Number restiction : s > = Number restiction : s, >, =
◦ Inverse role (
-
), transitive role (
+
), Role hierarchy
10:58:57 6 R. Akerkar: Reasoning in DL
 (Inverse Role) hasParent = hasChild
-
◦ hasParent(Bob,Alice) -> hasChild(Alice, Bob)
 (Transitive Role)hasBrother
h B h (B b D id) h B h (D id M k) ◦ hasBrother(Bob,David), hasBrother(David, Mack)
-> hasBrother(Bob,Mack)
 (Role Hierarchy) hasMother _ hasParent  (Role Hierarchy) hasMother _ hasParent
◦ hasMother(Bob,Alice) -> hasParent(Bob, Alice)
 HappyFather _ Father Π >1 hasChild.Woman ppy _
Π >1 hasChild.Man
10:58:57 7 R. Akerkar: Reasoning in DL
Knowledge Base
Tbox (schema)
HappyFather _ Person Π >1
hasChild.Woman Π >1 hasChild.Man
y
s
t
e
m
f
a
c
e
Abox (data)
r
e
n
c
e

S
y
I
n
t
e
r
f
Happy - Fat her ( Bob)
I
n
f
e
(Example taken from Ian Horrocks, U Manchester, UK)
10:58:57 8 R. Akerkar: Reasoning in DL
 ALC: the smallest DL that is propositionally
closed closed
◦ Constructors include booleans (and, or, not),
Restrictions on role successors
 SHOIQ = OWL DL
S=ALCR
+
: ALC with transitive role
H = role hierarchy
O = nomial .e.g WeekEnd = {Saturday, Sunday}
I = Inverse role
Q = qulified number restriction e.g. >=1
hasChild Man hasChild.Man
 N = number restriction e.g. >=1 hasChild
10:58:57 9 R. Akerkar: Reasoning in DL
 What is Description Logic (DL) p g ( )
 Semantics of DL
 Basic Tableau Algorithm
10:58:57 10 R. Akerkar: Reasoning in DL
DL Ontology: is a set of terms and their gy
relations
Interpretation of a DL Ontology: A possible
world ("model") that materializes the
ontology
Ontology:
Student _ People
Student _ -Present Topic Student _ -Present.Topic
KR _ Topic
DL _ KR
10:58:57 11 R. Akerkar: Reasoning in DL
 DL semantics defined by interpretations: I = (A
I
, .
I
),
where
◦ A
I
is the domain (a non-empty set)

.I
is an interpretation function that maps: is an interpretation function that maps:
 Concept (class) name A -> subset A
I
of A
I
 Role (property) name R -> binary relation R
I
over A
I
I di id l i i
I
l t f A
I
 Individual name i -> i
I
element of A
I
 Interpretation function .
I
tells us how to interpret
atomic concepts, properties and individuals. p , p p
◦ The semantics of concept forming operators is given by
extending the interpretation function in an obvious way.
10:58:57 12 R. Akerkar: Reasoning in DL
 I = (A
I
, .
I
)
 A
I
= {Raj, DL_Reasoning}
 People
I
=Student
I
={Raj}
 Topic
I
=KR
I
=DL
I
={DL_Reasoning}
 Present
I
={(Raj, DL_Reasoning)}
An interpretation that satisifies all axioms in an DL
ontology is also called a model of the ontology.
10:58:57 13 R. Akerkar: Reasoning in DL
Description Logics Tutorial, Ian Horrocks and Ulrike Sattler, ECAI-2002
10:58:57 14 R. Akerkar: Reasoning in DL
Description Logics Tutorial, Ian Horrocks and Ulrike Sattler, ECAI-2002
10:58:57 15 R. Akerkar: Reasoning in DL
 What is Description Logic (DL) p g ( )
 Semantics of DL
 Basic Tableau Algorithm
10:58:57 16 R. Akerkar: Reasoning in DL
 "Machine Understanding" g
 Find facts that are implicit in the ontology
given explicitly stated facts
◦ Find what you know, but you don't know you know
it - yet.
 Example  Example
◦ A is father of B, B is father of C, then A is ancestor
of C.
◦ D is mother of B, then D is female
10:58:57 17 R. Akerkar: Reasoning in DL
 Knowledge is correct (captures intuitions)
C subsumes D w r t K iff for every model I of K C
I
µ D
I
◦ C subsumes D w.r.t. K iff for every model I of K, C
I
µ D
I
 Knowledge is minimally redundant (no unintended synonyms)
◦ C is equivallent to D w.r.t. K iff for every model I of K, C
I
= D
I
K l d i i f l ( l h i t )  Knowledge is meaningful (classes can have instances)
◦ C is satisfiable w.r.t. K iff there exists some model I of K s.t. C
I
=
C;
 Querying knowledge  Querying knowledge
◦ x is an instance of C w.r.t. K iff for every model I of K, x
I
e C
I
◦ hx,yi is an instance of R w.r.t. K iff for, every model I of K, (x
I
,y
I
) e
R
I
R
 Knowledge base consistency
◦ A KB K is consistent iff there exists some model I of K
10:58:57 18 R. Akerkar: Reasoning in DL
Many inference tasks can be reduced to subsumption Many inference tasks can be reduced to subsumption
reasoning
Subsumption can be reduced to satisfiability p y
10:58:57 19 R. Akerkar: Reasoning in DL
 Tableau Algorithm is the de facto standard g
reasoning algorithm used in DL
 Basic intuitions
◦ Reduces a reasoning problem to concept satisfiability
problem
◦ Finds an interpretation that satisfies concepts in p p
question.
◦ The interpretation is incrementally constructed as a
"Tableau" Tableau
10:58:57 20 R. Akerkar: Reasoning in DL
 given: Wife_ Woman, Woman_ Person
question: if Wife_ Person
 Reasoning process
T t if th i i di id l th t i W b t t ◦ Test if there is a individual that is a Woman but not
a Person, i.e. test the satisfiability of concept
C
0
=(WifeΠ¬Person)
0
◦ C
0
(x) -> Wife(x), (¬Person)(x)
◦ Wife(x)->Woman(x)
W ( ) >P ( ) ◦ Woman(x) ->Person(x)
◦ Conflict!
◦ C
0
is unsatisfiable, therefore Wife_ Person is true C
0
is unsatisfiable, therefore Wife_ Person is true
with the given ontology.
10:58:57 21 R. Akerkar: Reasoning in DL
 Transform C into negation normal form(NNF),
i ti l i f t f t i.e. negation occurs only in front of concept
names.
 Denote the transformed expression as C
0
, the p
0
,
algorithm starts with an ABox A
0
= {C
0
(x
0
)}, and
apply consistency-preserving transformation
rules (tableaux expansion) to the ABox as far rules (tableaux expansion) to the ABox as far
as possible.
 If one possible ABox is found, C
0
is satisfiable.
f f d d ll h h  If not ABox is found under all search pathes,
C
0
is unsatisfiable.
10:58:57 22 R. Akerkar: Reasoning in DL
10:58:57 23 R. Akerkar: Reasoning in DL
Clash Clash
10:58:57 24 R. Akerkar: Reasoning in DL
 An ABox is called complete if none of the
expansion rules applies to it.
 An ABox is called consistent if no logic
l h i f d clash is found.
 If any complete and consistent ABox is
found the initial ABox A is satisfiable found, the initial ABox A
0
is satisfiable
 The expansion terminates, either when
finds a complete and consistent ABox or finds a complete and consistent ABox, or
try all search pathes ending with complete
but inconsistent ABoxes.
10:58:57 25 R. Akerkar: Reasoning in DL
 Embed the TBox in the initial ABox concept
 C_D is equivalent T_ ¬C U D (T is the
"top" concept. It imeans ¬C U D is the super
t f ANY t ) concept for ANY concepts)
 E.g.
Given ontology: Mother Woman Π Parent ◦ Given ontology: Mother _ Woman Π Parent,
Woman _ Person
◦ Query: Mother _ Person y
◦ The intitial ABox is : ¬Mother U(Woman Π Parent)
Π (¬Woman U Person) Π (Mother Π ¬Person)
10:58:57 26 R. Akerkar: Reasoning in DL
Search
10:58:57 27 R. Akerkar: Reasoning in DL
 Another explanation of tableaux algorithm
is that it works on a finite completion tree
whose
i di id l i th t bl d t d ◦ individuals in the tableau correspond to nodes
◦ and whose interpretation of roles is taken from
the edge labels. g
10:58:57 28 R. Akerkar: Reasoning in DL
 Similar tableaux expansions can be
d i d f i DL designed for more expressive DL
languages.
 A tableau algorithm has to meet three  A tableau algorithm has to meet three
requirements
◦ Soundness: if a complete and clash-free ABox
is found by the algorithm the ABox must is found by the algorithm, the ABox must
satisfies the initial concept C
0
.
◦ Completeness: if the initial concept C
0
is
i fi bl h l i h l fi d satisfiable, the algorithm can always find an
complete and clash-free ABox
◦ Termination: the algorithm can terminate in
finite steps with specific result.
10:58:57 29 R. Akerkar: Reasoning in DL
 What is Description Logic (DL) p g ( )
 Semantics of DL
 Basic Tableau Algorithm
10:58:57 30 R. Akerkar: Reasoning in DL
 Rich literatures in the past decade.
◦ Blocking (Subset Blocking, Pair Locking, Dynamic
Blocking) Blocking)
◦ For more expressive languages: number
restriction, inverse role, transitive role, nomial,
data type
◦ Detailed analysis of complexities.
10:58:57 31 R. Akerkar: Reasoning in DL
SHIQ Expansion Rules
10:58:57 32 R. Akerkar: Reasoning in DL
 F. Baader, W. Nutt. Basic Description Logics. In the Description
Logic Handbook edited by F Baader D Calvanese D L Logic Handbook, edited by F. Baader, D. Calvanese, D.L.
McGuinness, D. Nardi, P.F. Patel-Schneider, Cambridge
University Press, 2002, pages 47-100.
 Ian Horrocks and Ulrike Sattler. Description Logics Tutorial,  Ian Horrocks and Ulrike Sattler. Description Logics Tutorial,
ECAI-2002, Lyon, France, July 23rd, 2002.
 Ian Horrocks and Ulrike Sattler. A tableaux decision procedure
for SHOIQ. In Proc. of the 19th Int. Joint Conf. on Artificial
Intelligence (IJCAI 2005), 2005.
 I. Horrocks and U. Sattler. A description logic with transitive
and inverse roles and role hierarchies. Journal of Logic and
Computation, 9(3):385-410, 1999.
10:58:57 33 R. Akerkar: Reasoning in DL

   

p g What is Description Logics ( ) (DL) Semantics of DL Basic Tableau Algorithm Advanced Tableau Algorithm

R. Akerkar: Reasoning in DL

10:58:57

2

g g A formal logic-based knowledge representation language

 

Decidable fragments of FOL g Widely used in database (e.g., DL CLASSIC) and semantic web (e.g., OWL language)

in terms of concepts (classes), roles ( ( l ) l (properties, relationships) and ti l ti hi ) d individuals (instances)

R. Akerkar: Reasoning in DL

10:58:57

3

Akerkar: Reasoning in DL 10:58:57 4 .Person include Man(Male) and Woman(Female). Woman(Female) A Man is not a Woman A Father is a Man who has Child A Mother is a Woman who has Child Both Father and Mother are Parent Grandmother is a Mother of a Parent A Wife is a Woman and has a Husband( which as Man) A Mother Without Daughter is a Mother g whose all Child(ren) are not Women R.

10:58:57 R. Akerkar: Reasoning in DL 5 .

g. transitive role (+). Mother E P F th M th And(Π) .g. Not (¬) Universal qualifier ( Existent qualifier() Number restiction :     Inverse role (-). Female  Person Operators (for forming concepts and roles) ◦ ◦ ◦ ◦ ◦ E. Or(U).     Concepts (unary predicates/formulae with one free variable) Roles (binary predicates/formulae with two free variables) Individual names (constants) Subsumption (relations between concepts) ◦ E. Akerkar: Reasoning in DL 10:58:57 6 . Cindy ◦ E.g. Role hierarchy R. Bob... Father. Alice. hasHudband ◦ E. Person.g. hasChild..

Alice) -> hasChild(Alice.Man R.David).Mack) ◦ hasMother(Bob. Mack) -> hasBrother(Bob. Bob) ◦ h B h (B b D id) h B h (D id M k) hasBrother(Bob. hasBrother(David. Akerkar: Reasoning in DL 10:58:57 7 . (Inverse Role) hasParent = hasChild(Transitive Role)hasBrother ◦ hasParent(Bob. Alice)   (Role Hierarchy) hasMother  hasParent  HappyFather  Father Π hasChild.Woman ppy Π hasChild.Alice) -> hasParent(Bob.

U Manchester.Man Abox (data) Happy-Father(Bob) (Example taken from Ian Horrocks.Knowledge Base Tbox (schema) Inference Sy ystem Interf face HappyFather  Person Π  hasChild. UK) R.Woman Π hasChild. Akerkar: Reasoning in DL 10:58:57 8 .

g WeekEnd = {Saturday.g.g. Akerkar: Reasoning in DL 10:58:57 9 . >=1 hasChild R. ALC: the smallest DL that is propositionally closed ◦ Constructors include booleans (and. not).Man hasChild Man  N = number restriction e. or. Sunday} Q = qulified number restriction e.e. >=1 hasChild. Restrictions on role successors  SHOIQ = OWL DL S=ALCR+: ALC with transitive role H = role hierarchy I = Inverse role O = nomial .

    p g (DL) What is Description Logic ( ) Semantics of DL Basic Tableau Algorithm Advanced Tableau Algorithm R. Akerkar: Reasoning in DL 10:58:57 10 .

Akerkar: Reasoning in DL 11 .Topic KR  Topic DL  KR 10:58:57 R.DL Ontology: is a set of terms and their gy relations Interpretation of a DL Ontology: A possible world ("model") that materializes the ontology Ontology: Student  People Student  Present Topic Present.

I is an interpretation function that maps:  Concept (class) name A -> subset AI of I  Role (property) name R -> binary relation RI over I  I di id l name i -> iI element of I Individual l t f  Interpretation function . R.I). .p p ◦ The semantics of concept forming operators is given by extending the interpretation function in an obvious way. p . DL semantics defined by interpretations: I = (I. properties and individuals. where ◦ I is the domain (a non-empty set) ◦ .I tells us how to interpret atomic concepts. Akerkar: Reasoning in DL 10:58:57 12 .

R. Akerkar: Reasoning in DL 10:58:57 13 . DL_Reasoning} PeopleI=StudentI={Raj} TopicI=KRI=DLI={DL_Reasoning} PresentI={(Raj.I) I = {Raj. DL_Reasoning)} An interpretation that satisifies all axioms in an DL ontology is also called a model of the ontology.     I = (I. .

ECAI-2002 R. Ian Horrocks and Ulrike Sattler. Akerkar: Reasoning in DL 10:58:57 14 .Description Logics Tutorial.

Ian Horrocks and Ulrike Sattler.Description Logics Tutorial. ECAI-2002 R. Akerkar: Reasoning in DL 10:58:57 15 .

Akerkar: Reasoning in DL 10:58:57 16 .    p g (DL) What is Description Logic ( ) Semantics of DL Basic Tableau Algorithm Advanced Tableau Algorithm R.

but you don't know you know ◦ A is father of B. ◦ Find what you know. D is mother of B. B is father of C. Akerkar: Reasoning in DL 10:58:57 17 . then A is ancestor ◦ of C. then D is female  R.  g "Machine Understanding" Find facts that are implicit in the ontology given explicitly stated facts Example it .yet.

t.yI)  RI ◦ A KB K is consistent iff there exists some model I of K  Knowledge base consistency R.r.   Knowledge is correct (captures intuitions) ◦ C subsumes D w.r.t. xI  CI ◦ hx. K iff for every model I of K.  Querying knowledge ◦ x is an instance of C w. every model I of K. K iff there exists some model I of K s.t.r.t. CI  . K iff for. (xI. K iff for every model I of K. K iff for every model I of K.yi is an instance of R w.t.r. CI = DI Knowledge i meaningful ( l K l d is i f l (classes can h have instances) i t ) ◦ C is satisfiable w. CI µ DI wrt K Knowledge is minimally redundant (no unintended synonyms) ◦ C is equivallent to D w. Akerkar: Reasoning in DL 10:58:57 18 .r.t.

Akerkar: Reasoning in DL 19 .Many inference tasks can be reduced to subsumption reasoning Subsumption can be reduced to satisfiability p y 10:58:57 R.

◦ The interpretation is incrementally constructed as a "Tableau" Tableau R. Akerkar: Reasoning in DL 10:58:57 20 .  g Tableau Algorithm is the de facto standard reasoning algorithm used in DL Basic intuitions ◦ Reduces a reasoning problem to concept satisfiability problem ◦ Finds an interpretation that satisfies concepts in p p question.

  given: Wife Woman. therefore Wife Person is true with the given ontology.e. (¬Person)(x) ◦ Wife(x)->Woman(x) ( ) >P ( ) ◦W Woman(x) ->Person(x) ◦ Conflict! ◦ C0 is unsatisfiable. Woman Person question: if Wife Person Reasoning process ◦ T t if th Test there is a individual th t i a W i i di id l that is Woman b t not but t a Person. test the satisfiability of concept C0=(WifeΠ¬Person) ◦ C0(x) -> Wife(x). Akerkar: Reasoning in DL 10:58:57 21 . R. i.

If one possible ABox is found. C0 is unsatisfiable. f d d ll h h If not ABox is f found under all search pathes. negation occurs only in front of concept i ti l i f t f t names. i. and apply consistency-preserving transformation rules (tableaux expansion) to the ABox as far as possible. R. Akerkar: Reasoning in DL 10:58:57 22 . the p algorithm starts with an ABox A0 = {C0(x0)}. Denote the transformed expression as C0.e.    Transform C into negation normal form(NNF). C0 is satisfiable.

R. Akerkar: Reasoning in DL 10:58:57 23 .

Akerkar: Reasoning in DL 10:58:57 24 .Clash R.

l hi f d If any complete and consistent ABox is found. R.    An ABox is called complete if none of the expansion rules applies to it. either when finds a complete and consistent ABox or ABox. An ABox is called consistent if no logic clash is found. Akerkar: Reasoning in DL 10:58:57 25 . try all search pathes ending with complete but inconsistent ABoxes. found the initial ABox A0 is satisfiable The expansion terminates.

Akerkar: Reasoning in DL 10:58:57 26 . It imeans ¬C U D is the super concept f ANY concepts) t for t ) E. ◦ Given ontology: Mother  Woman Π Parent Parent.g. Woman  Person ◦ Query: Mother  Person y ◦ The intitial ABox is : ¬Mother U(Woman Π Parent) Π (¬Woman U Person) Π (Mother Π ¬Person) R.   Embed the TBox in the initial ABox concept CD is equivalent T ¬C U D (T is the "top" concept.

Akerkar: Reasoning in DL 10:58:57 27 .Search R.

R. Another explanation of tableaux algorithm is that it works on a finite completion tree whose ◦ i di id l i th t bl individuals in the tableau correspond t nodes d to d ◦ and whose interpretation of roles is taken from g the edge labels. Akerkar: Reasoning in DL 10:58:57 28 .

  Similar tableaux expansions can be designed for more expressive DL d i df i languages. A tableau algorithm has to meet three requirements ◦ Soundness: if a complete and clash-free ABox is found by the algorithm. the ABox must algorithm satisfies the initial concept C0. the algorithm can always fi d an find complete and clash-free ABox ◦ Termination: the algorithm can terminate in finite steps with specific result. ◦ Completeness: if the initial concept C0 is i fi bl h l ih l satisfiable. Akerkar: Reasoning in DL 10:58:57 29 . R.

Akerkar: Reasoning in DL 10:58:57 30 .    p g (DL) What is Description Logic ( ) Semantics of DL Basic Tableau Algorithm Advanced Tableau Algorithm R.

inverse role. R. nomial. transitive role. Advanced techniques ◦ Blocking (Subset Blocking. Akerkar: Reasoning in DL 10:58:57 31 . Dynamic Blocking) ◦ For more expressive languages: number restriction.  Rich literatures in the past decade. Pair Locking. data type ◦ Detailed analysis of complexities.

Akerkar: Reasoning in DL 10:58:57 32 .SHIQ Expansion Rules R.

France. D. on Artificial Intelligence (IJCAI 2005). R. 2002.    F. Description Logics Tutorial. Cambridge University Press. 2005.F. D. Baader. pages 47-100. of the 19th Int. Lyon.L. Nutt. A tableaux decision procedure for SHOIQ. In the Description Logic Handbook. Baader D Calvanese D L Handbook McGuinness. A description logic with transitive and inverse roles and role hierarchies. 1999. F. 2002. Joint Conf. Journal of Logic and Computation. edited by F Baader. Ian Horrocks and Ulrike Sattler. D. ECAI-2002. In Proc. I. W. Basic Description Logics. Sattler. 9(3):385-410. July 23rd. Calvanese. Ian Horrocks and Ulrike Sattler. Nardi. Patel-Schneider. Horrocks and U. P. Akerkar: Reasoning in DL 10:58:57 33 .