Professional Documents
Culture Documents
Main Results
Concluding Remarks
Example
A turkey that walks is alive
Teasing a turkey makes it walk
It is always possible to tease a turkey
A dead turkey remains dead after teasing it
Example
A turkey that walks is alive
Teasing a turkey makes it walk
It is always possible to tease a turkey
A dead turkey remains dead after teasing it
?
Current actions Desired
state state
Outline
1 Introduction
Describing Action Theories
Unwanted Conclusions
2 Main Results
Decomposing Theories
Logical Modularity
Exploiting Modularity
Theory Change
3 Concluding Remarks
Outline
1 Introduction
Describing Action Theories
Unwanted Conclusions
2 Main Results
Decomposing Theories
Logical Modularity
Exploiting Modularity
Theory Change
3 Concluding Remarks
Outline
1 Introduction
Describing Action Theories
Unwanted Conclusions
2 Main Results
Decomposing Theories
Logical Modularity
Exploiting Modularity
Theory Change
3 Concluding Remarks
Outline
1 Introduction
Describing Action Theories
Unwanted Conclusions
2 Main Results
Decomposing Theories
Logical Modularity
Exploiting Modularity
Theory Change
3 Concluding Remarks
Formalizing Domains
Formalizing Domains
In this work. . .
we have chosen Modal Logic
Weak version of Propositional Dynamic Logic (PDL)
Simple and decidable
With a tableaux-based theorem prover: Lotrec
Logical Preliminaries
Ontology
Actions: a1 a2 !!!#"
Atomic propositions: $&%!')(* p1 p2 !!!#"
Literals: +-,./$&%!'0(1243 p5 p 67$&%!'0( "
Action operators
For each a 6@ , a modal operator A aB
A aBC? : “after execution of a, ? is true”
D
aEGF : “a is inexecutable”
H
a I?J def 3KA aBC3L?
M
a NPO : “a is executable”
Complex formulas: Q 1 Q 2 !!!
Ivan José Varzinczak What is a Good Domain Description?
Introduction
Describing Action Theories
Main Results
Unwanted Conclusions
Concluding Remarks
Logical Preliminaries
Ontology
Actions: a1 a2 !!!#"
Atomic propositions: $&%!')(* p1 p2 !!!#"
Literals: +-,./$&%!'0(1243 p5 p 67$&%!'0( "
Action operators
For each a 6@ , a modal operator A aB
A aBC? : “after execution of a, ? is true”
D
aEGF : “a is inexecutable”
H
a I?J def 3KA aBC3L?
M
a NPO : “a is executable”
Complex formulas: Q 1 Q 2 !!!
Ivan José Varzinczak What is a Good Domain Description?
Introduction
Describing Action Theories
Main Results
Unwanted Conclusions
Concluding Remarks
Logical Preliminaries
Ontology
Actions: a1 a2 !!!#"
Atomic propositions: $&%!')(* p1 p2 !!!#"
Literals: +-,./$&%!'0(1243 p5 p 67$&%!'0( "
Action operators
For each a 6@ , a modal operator A aB
A aBC? : “after execution of a, ? is true”
D
aEGF : “a is inexecutable”
H
a I?J def 3KA aBC3L?
M
a NPO : “a is executable”
Complex formulas: Q 1 Q 2 !!!
Ivan José Varzinczak What is a Good Domain Description?
Introduction
Describing Action Theories
Main Results
Unwanted Conclusions
Concluding Remarks
Logical Preliminaries
Ontology
Actions: a1 a2 !!!#"
Atomic propositions: $&%!')(* p1 p2 !!!#"
Literals: +-,./$&%!'0(1243 p5 p 67$&%!'0( "
Action operators
For each a 6@ , a modal operator A aB
A aBC? : “after execution of a, ? is true”
D
aEGF : “a is inexecutable”
H
a I?J def 3KA aBC3L?
M
a NPO : “a is executable”
Complex formulas: Q 1 Q 2 !!!
Ivan José Varzinczak What is a Good Domain Description?
Introduction
Describing Action Theories
Main Results
Unwanted Conclusions
Concluding Remarks
Logical Preliminaries
Example
Actions: shoot, tease
Propositions: loaded, alive, walking
H
Formulas: alive RS3 walking, tease IT ,
loaded UVA shootBC3 alive
Semantics
Definition
b
Semantics
Definition
b
Semantics
Example
H
If > a1 a2 " , and $&%!'0(g p1 p2 " , W W R I , where
Semantics
Example
a1
a1
p1 q p2 p1 s p2 u v)w
p1 p2
r
x
a2 uv w
p1 a2 |C} p2
W : a1 a1 y{z
u v)w
} p1 y{~
a1
p1 str p2
uv w
} p2 yVz
a1 |
a2
Ivan José Varzinczak What is a Good Domain Description?
Introduction
Describing Action Theories
Main Results
Unwanted Conclusions
Concluding Remarks
Semantics
Definition
b b
Definition
Q is a consequence of b the set of global axioms b in all
PDL-models
b (noted PDL Q ) iff for every W , if c , then
c Q .
Semantics
Definition
b b
Definition
Q is a consequence of b the set of global axioms b in all
PDL-models
b (noted PDL Q ) iff for every W , if c , then
c Q .
Outline
1 Introduction
Describing Action Theories
Unwanted Conclusions
2 Main Results
Decomposing Theories
Logical Modularity
Exploiting Modularity
Theory Change
3 Concluding Remarks
Example
A turkey that walks is alive: walking U alive
Teasing a turkey makes it to walk: A teaseB walking
H
It is always possible to tease a turkey: tease IT
A dead turkey remains dead after teasing it
D
alive teaseE. alive
If the gun is loaded, shooting kills the turkey
D
loaded shootEd alive
Teasing does not unload the gun
D
loaded teaseE loaded
Example
H b
tease IT
walking U alive
b A teaseB alive
A teaseB walking alive teaseBPj alive RS3 alive k
b 3 UVA
Example
H b
tease IT
walking U alive
b A teaseB alive
A teaseB walking alive teaseBPj alive RS3 alive k
b 3 UVA
Example
H b
tease IT
walking U alive
b A teaseB alive
A teaseB walking alive teaseBPj alive RS3 alive k
b 3 UVA
Example
H b
tease IT
walking U alive
b A teaseB alive
A teaseB walking alive teaseBPj alive RS3 alive k
b 3 UVA
Example
H b
tease IT
walking U alive
b A teaseB alive
A teaseB walking alive teaseBPj alive RS3 alive k
b 3 UVA
Example
H b
tease IT
walking U alive
b A teaseB alive
A teaseB walking alive teaseBPj alive RS3 alive k
b 3 UVA
Outline
1 Introduction
Describing Action Theories
Unwanted Conclusions
2 Main Results
Decomposing Theories
Logical Modularity
Exploiting Modularity
Theory Change
3 Concluding Remarks
Defining modules
: set of static laws
Given a 6@
a
: effect laws for a
a
: executability laws for a
a
: inexecutability laws for a
H
a a a : domain description for a
n I
a, a, and a
a 4#
a 4#
a #
H
n I : the action theory of a given domain
Defining modules
: set of static laws
Given a 6@
a
: effect laws for a
a
: executability laws for a
a
: inexecutability laws for a
H
a a a : domain description for a
n I
a, a, and a
a 4#
a 4#
a #
H
n I : the action theory of a given domain
Defining modules
: set of static laws
Given a 6@
a
: effect laws for a
a
: executability laws for a
a
: inexecutability laws for a
H
a a a : domain description for a
n I
a, a, and a
a 4#
a 4#
a #
H
n I : the action theory of a given domain
Defining modules
: set of static laws
Given a 6@
a
: effect laws for a
a
: executability laws for a
a
: inexecutability laws for a
H
a a a : domain description for a
n I
a, a, and a
a 4#
a 4#
a #
H
n I : the action theory of a given domain
In our example
If we had an action wait
¡ £¤ ¢ D
PDL
loaded waitE loaded
Definition
Dependence relation [Castilho et al. 99]: ¥¦X§_¨©+ª,.
Example
shoot ¥«3 alive, tease ¥ walking, tease ¥ ¬ alive
From wait ¥«
¬ 3 loaded conclude loaded UVA waitB loaded
In our example
If we had an action wait
¡ £¤ ¢ D
PDL
loaded waitE loaded
Definition
Dependence relation [Castilho et al. 99]: ¥¦X§_¨©+ª,.
Example
shoot ¥«3 alive, tease ¥ walking, tease ¥ ¬ alive
From wait ¥«
¬ 3 loaded conclude loaded UVA waitB loaded
In our example
If we had an action wait
¡ £¤ ¢ D
PDL
loaded waitE loaded
Definition
Dependence relation [Castilho et al. 99]: ¥¦X§_¨©+ª,.
Example
shoot ¥«3 alive, tease ¥ walking, tease ¥ ¬ alive
From wait ¥«
¬ 3 loaded conclude loaded UVA waitB loaded
In our example
If we had an action wait
¡ £¤ ¢ D
PDL
loaded waitE loaded
Definition
Dependence relation [Castilho et al. 99]: ¥¦X§_¨©+ª,.
Example
shoot ¥«3 alive, tease ¥ walking, tease ¥ ¬ alive
From wait ¥«
¬ 3 loaded conclude loaded UVA waitB loaded
Restriction on models
For all
b
wRa w e : b
b ¬ c
w
p implies b ¬ c p, if a ¥ ¬ p
wf
)c p implies 0c p, if a ¥
¬ 3 p.
w wf
® instead of
PDL
Example b
n ® loaded U¯A waitB loaded
Restriction on models
For all
b
wRa w e : b
b ¬ c
w
p implies b ¬ c p, if a ¥ ¬ p
wf
)c p implies 0c p, if a ¥
¬ 3 p.
w wf
® instead of
PDL
Example b
n ® loaded U¯A waitB loaded
Restriction on models
For all
b
wRa w e : b
b ¬ c
w
p implies b ¬ c p, if a ¥ ¬ p
wf
)c p implies 0c p, if a ¥
¬ 3 p.
w wf
® instead of
PDL
Example b
n ® loaded U¯A waitB loaded
Outline
1 Introduction
Describing Action Theories
Unwanted Conclusions
2 Main Results
Decomposing Theories
Logical Modularity
Exploiting Modularity
Theory Change
3 Concluding Remarks
PC (Consistency):
a a a ¬ ®
b b
n
Motivation
Better control what is going on
PC (Consistency):
a a a ¬ ®
b b
n
Motivation
Better control what is going on
PC (Consistency):
a a a ¬ ®
b b
n
Motivation
Better control what is going on
PC (Consistency):
a a a ¬ ®
b b
n
Motivation
Better control what is going on
PC (Consistency):
a a a ¬ ®
b b
n
Motivation
Better control what is going on
H
tease IT "i[ 43 alive UVA teaseBC "
tease
b tease tease ® alive
n
But alive
¬
! Postulate PS violated
H
tease IT "i[ 43 alive UVA teaseBC "
tease
b tease tease ® alive
n
But alive
¬
! Postulate PS violated
H
tease IT "i[ 43 alive UVA teaseBC "
tease
b tease tease ® alive
n
But alive
¬
! Postulate PS violated
H
tease IT "i[ 43 alive UVA teaseBC "
tease
b tease tease ® alive
n
But alive
¬
! Postulate PS violated
Idea of algorithm
H
For each ?JU a IT
1 find ?[eUVA aBC entailed by the theory
if ?´R©?[e is consistent with
2
Idea of algorithm
H
For each ?JU a IT
1 find ?[eUVA aBC entailed by the theory
if ?´R©?[e is consistent with
2
Example (cont.)
Alternatives for repairing:
:= 1© alive "
add tease ¥ alive
weaken A teaseB walking: alive UVA teaseB walking
H H
weaken tease IT : alive U tease IT
contraction of action theories (addressed later)
Example (cont.)
Alternatives for repairing:
:= 1© alive "
add tease ¥ alive
weaken A teaseB walking: alive UVA teaseB walking
H H
weaken tease IT : alive U tease IT
contraction of action theories (addressed later)
Example (cont.)
Alternatives for repairing:
:= 1© alive "
add tease ¥ alive
weaken A teaseB walking: alive UVA teaseB walking
H H
weaken tease IT : alive U tease IT
contraction of action theories (addressed later)
Theorem
H
a a aI and ¥ satisfy Postulate PS iff
.
n imp* ðï
Theorem
H
Let imp* be the output of Algorithm 1 on input a a aI
n
and ¥ . Then
H
ab a aI has no implicit static law.
1 imp* n
a a a
n ®³ñ imp* .
Corollary b b
For all ,
a a a iff
.
?ò6@8:9<; n ® ? 1 imp* ?
Theorem
H
a a aI and ¥ satisfy Postulate PS iff
.
n imp* ðï
Theorem
H
Let imp* be the output of Algorithm 1 on input a a aI
n
and ¥ . Then
H
ab a aI has no implicit static law.
1 imp* n
a a a
n ®³ñ imp* .
Corollary b b
For all ,
a a a iff
.
?ò6@8:9<; n ® ? 1 imp* ?
Theorem
H
a a aI and ¥ satisfy Postulate PS iff
.
n imp* ðï
Theorem
H
Let imp* be the output of Algorithm 1 on input a a aI
n
and ¥ . Then
H
ab a aI has no implicit static law.
1 imp* n
a a a
n ®³ñ imp* .
Corollary b b
For all ,
a a a iff
.
?ò6@8:9<; n ® ? 1 imp* ?
½óÆ*¾<ÆSÍì»
b
tease À walking » shoot Àôâ alive
tease b ß®õA teaseB alive
tease ® 3 alive UVA teaseb B3 alive (from tease ¥ ¬ alive)
Thus
teaseb tease tease ® 3 alive UVA teaseBC
n
But
tease ¬ 3 alive UVA teaseBC
PDL
! Postulate PI violated
½óÆ*¾<ÆSÍì»
b
tease À walking » shoot Àôâ alive
tease b ß®õA teaseB alive
tease ® 3 alive UVA teaseb B3 alive (from tease ¥ ¬ alive)
Thus
teaseb tease tease ® 3 alive UVA teaseBC
n
But
tease ¬ 3 alive UVA teaseBC
PDL
! Postulate PI violated
½óÆ*¾<ÆSÍì»
b
tease À walking » shoot Àôâ alive
tease b ß®õA teaseB alive
tease ® 3 alive UVA teaseb B3 alive (from tease ¥ ¬ alive)
Thus
teaseb tease tease ® 3 alive UVA teaseBC
n
But
tease ¬ 3 alive UVA teaseBC
PDL
! Postulate PI violated
½óÆ*¾<ÆSÍì»
b
tease À walking » shoot Àôâ alive
tease b ß®õA teaseB alive
tease ® 3 alive UVA teaseb B3 alive (from tease ¥ ¬ alive)
Thus
teaseb tease tease ® 3 alive UVA teaseBC
n
But
tease ¬ 3 alive UVA teaseBC
PDL
! Postulate PI violated
½óÆ*¾<ÆSÍì»
b
tease À walking » shoot Àôâ alive
tease b ß®õA teaseB alive
tease ® 3 alive UVA teaseb B3 alive (from tease ¥ ¬ alive)
Thus
teaseb tease tease ® 3 alive UVA teaseBC
n
But
tease ¬ 3 alive UVA teaseBC
PDL
! Postulate PI violated
½óÆ*¾<ÆSÍì»
b
tease À walking » shoot Àôâ alive
tease b ß®õA teaseB alive
tease ® 3 alive UVA teaseb B3 alive (from tease ¥ ¬ alive)
Thus
teaseb tease tease ® 3 alive UVA teaseBC
n
But
tease ¬ 3 alive UVA teaseBC
PDL
! Postulate PI violated
Idea of algorithm
For each combination of effect laws
1 find inconsistent consequents
2 mark it as an implicit inexecutability
Idea of algorithm
For each combination of effect laws
1 find inconsistent consequents
2 mark it as an implicit inexecutability
for all a e X a do
?ù a f := ñ 4? i 5=? i UVA aBd÷ i 6
ae
"
a a
imp := imp 12ijP?[ù a f R73ªú k UVA aBC "
for all a e X a do
?ù a f := ñ 4? i 5=? i UVA aBd÷ i 6
ae
"
a a
imp := imp 12ijP?[ù a f R73ªú k UVA aBC "
for all a e X a do
?ù a f := ñ 4? i 5=? i UVA aBd÷ i 6
ae
"
a a
imp := imp 12ijP?[ù a f R73ªú k UVA aBC "
for all a e X a do
?ù a f := ñ 4? i 5=? i UVA aBd÷ i 6
ae
"
a a
imp := imp 12ijP?[ù a f R73ªú k UVA aBC "
for all a e X a do
?ù a f := ñ 4? i 5=? i UVA aBd÷ i 6
ae
"
a a
imp := imp 12ijP?[ù a f R73ªú k UVA aBC "
for all a e X a do
?ù a f := ñ 4? i 5=? i UVA aBd÷ i 6
ae
"
a a
imp := imp 12ijP?[ù a f R73ªú k UVA aBC "
for all a e X a do
?ù a f := ñ 4? i 5=? i UVA aBd÷ i 6
ae
"
a a
imp := imp 12ijP?[ù a f R73ªú k UVA aBC "
Theorem
H
IfH a a aIand ¥ satisfy Postulate PS, then
n
a a aI and ¥ satisfy Postulate PI iff imp ðï .
n
Postulate
PS* (No implicit static laws):
b b
Theorem
H H
and ¥ satisfy PS* iff
I
a a aI and ¥
n n
Postulate
PS* (No implicit static laws):
b b
Theorem
H H
and ¥ satisfy PS* iff
I
a a aI and ¥
n n
Postulate b
Theorem
H H
If n I and
H
¥ satisfy PS*, then nJ I and ¥
satisfy PC* iff a a a I and ¥ satisfies PC for all a 67 .
n
Postulate b
Theorem
H H
If n I and
H
¥ satisfy PS*, then nJ I and ¥
satisfy PC* iff a a a I and ¥ satisfies PC for all a 67 .
n
Postulate
PI* (No implicit inexecutability laws):
b b
Theorem
H H
Let n H I and ¥ satisfy PS*. n I and ¥
Postulate
PI* (No implicit inexecutability laws):
b b
Theorem
H H
Let n H I and ¥ satisfy PS*. n I and ¥
Outline
1 Introduction
Describing Action Theories
Unwanted Conclusions
2 Main Results
Decomposing Theories
Logical Modularity
Exploiting Modularity
Theory Change
3 Concluding Remarks
Reasoning Modularly
H
If n I and ¥ satisfy Postulate PS*, then
Theorem b b
n ® iff .
Theorem b b
® ?JUVA aBd÷ iff
a a ® ?JUVA aBG÷ .
n
Reasoning Modularly
H
If n I and ¥ satisfy Postulate PS*, then
Theorem b b
n ® iff .
Theorem b b
® ?JUVA aBd÷ iff
a a ® ?JUVA aBG÷ .
n
Reasoning Modularly
H
If n I and ¥ satisfy Postulate PS*, then
Theorem b b
H H
® ?JU a IT iff
a ® ?JU a IT .
n n
Corollary
PX is a consequence of PS.
Theorem
H
If n and ¥ satisfy Postulates PS* and PI*, then
b b
I
Reasoning Modularly
H
If n I and ¥ satisfy Postulate PS*, then
Theorem b b
H H
® ?JU a IT iff
a ® ?JU a IT .
n n
Corollary
PX is a consequence of PS.
Theorem
H
If n and ¥ satisfy Postulates PS* and PI*, then
b b
I
Reasoning Modularly
H
If n I and ¥ satisfy Postulate PS*, then
Theorem b b
H H
® ?JU a IT iff
a ® ?JU a IT .
n n
Corollary
PX is a consequence of PS.
Theorem
H
If n and ¥ satisfy Postulates PS* and PI*, then
b b
I
Reasoning Modularly
H
If n I and ¥ satisfy Postulate PS*, then
Theorem b
a1 !!!ÿ an Bd÷ iff
b
n ® ?JUVA
a1 an a1 an ® ?JUVA a1 !!!ÿ an Bd÷ .
Theorem b
H
a1 !!!ÿ an IP÷ iff
b
n ® ?JU
H
a1 an a1 an a1 an ® ?JU a1 !!!ÿ an IP÷ .
n
Reasoning Modularly
H
If n I and ¥ satisfy Postulate PS*, then
Theorem b
a1 !!!ÿ an Bd÷ iff
b
n ® ?JUVA
a1 an a1 an ® ?JUVA a1 !!!ÿ an Bd÷ .
Theorem b
H
a1 !!!ÿ an IP÷ iff
b
n ® ?JU
H
a1 an a1 an a1 an ® ?JU a1 !!!ÿ an IP÷ .
n
Outline
1 Introduction
Describing Action Theories
Unwanted Conclusions
2 Main Results
Decomposing Theories
Logical Modularity
Exploiting Modularity
Theory Change
3 Concluding Remarks
Another Tale
Example
If the switch is up, the room is lit up
up light
Toggling the switch changes its position
D
up
D
toggleE up
up toggleEd up
It is always possible to toggle the switch
M
toggle NO
Contraction: Motivation
Contraction: Motivation
Contraction: Motivation
Contraction: Motivation
Contraction: Motivation
Contraction: Semantics
toggle
up q
light up s light
toggle
Contraction: Semantics
toggle
up q
light up s light
toggle
Contraction: Semantics
Contracting static laws
H
W R Iup
light
up s×r light
Contraction: Semantics
Contracting static laws
H
W R Iup
light
up s×r light
Contraction: Semantics
H H
N.B.: We Re Ié¬ toggle IT
Executability laws to be weakened!
Contraction: Semantics
H H
N.B.: We Re Ié¬ toggle IT
Executability laws to be weakened!
Contraction: Semantics
H H
N.B.: We Re Ié¬ toggle IT
Executability laws to be weakened!
Contraction: Semantics
toggle
toggle
up q
light up s light
toggle
Contraction: Semantics
H
W R 1 Rae I§5 Rae X ij w we k 5 w 3 up "h"
Problems:
Don’t link light-worlds
Don’t link all light-worlds
Contraction: Semantics
H
W R 1 Rae I§5 Rae X ij w we k 5 w 3 up "h"
Problems:
Don’t link light-worlds
Don’t link all light-worlds
Contraction: Semantics
toggle
up q
light up s light
Contraction: Semantics
H
W R ø Rae I§5 Rae Xðij w we k 5 wRa w e and w up "h"
Contraction: Semantics
H
W R ø Rae I§5 Rae Xðij w we k 5 wRa w e and w up "h"
Contraction: Syntax
Domain descriptions
H
Simplification: n I
Contraction: Syntax
·
PMA up light $# light up %
·
M M
#P up & light ¶Ïµ:[ toggle NPO('Kµ@ toggle NPO*) %
Contraction: Syntax
·
PMA up light $# light up %
·
M M
#P up & light ¶Ïµ:[ toggle NPO('Kµ@ toggle NPO*) %
Contraction: Syntax
Contraction: Syntax
Soundness
Theorem b b
H H H
If W R I R R , then W R I 1 n I 1 .
Incompleteness
Example
H
, p aBC , a IT
ðï UVA " "
H H
Unique model: W R I- h43 p "h"i ijm43 p "i 43 p "4kp" I
H H H
n p a 0
I ï p UVA aB "i 43 p
b U a IT " I .
H
Syntactically: successful, as a IT .
n e ¬
PDL
p U
Incompleteness
Example
H
, p aBC , a IT
ðï UVA " "
H H
Unique model: W R I- h43 p "h"i ijm43 p "i 43 p "4kp" I
H H H
n p a 0
I ï p UVA aB "i 43 p
b U a IT " I .
H
Syntactically: successful, as a IT .
n e ¬
PDL
p U
Incompleteness
Example
H
, p aBC , a IT
ðï UVA " "
H H
Unique model: W R I- h43 p "h"i ijm43 p "i 43 p "4kp" I
H H H
n p a 0
I ï p UVA aB "i 3 p
b U a IT " I .
H
Syntactically: successful, as a IT .
n e ¬
PDL
p U
Incompleteness
Example
H
, p aBC , a IT
ðï UVA " "
H H
Unique model: W R I- h43 p "h"i ijm43 p "i 43 p "4kp" I
H H H
n p a 0
I ï p UVA aB "i 43 p
b U a IT " I .
H
Syntactically: successful, as a IT .
n e ¬
PDL
p U
Incompleteness
Example
H
, p aBC , a IT
ðï UVA " "
H H
Unique model: W R I- h43 p "h"i ijm43 p "i 43 p "4kp" I
H H H
n p a 0
I ï p UVA aB "i 43 p
b U a IT " I .
H
Syntactically: successful, as a IT .
n e ¬
PDL
p U
Completeness: Modularity
Theorem b
H H
IfH and ¥ satisfy Postulate PS*, H then iff
b b
n I n I 1 32
H
W RI 1 32 , for every W R I such that W R I .
R R
Levi identity
Revise by glued U{A toggleBC amounts to
D
1 Contract by glued D toggleEGFK
2 Expand by glued toggleEGF
Problem: we can contract by domain laws only
D M
M ¡
glued
toggleEGFK54 glued ¶ toggle NPOK
N!
glued 687 toggle 90: not defined
What is the negation of
an effect law?
an executability law?
Levi identity
Revise by glued U{A toggleBC amounts to
D
1 Contract by glued D toggleEGFK
2 Expand by glued toggleEGF
Problem: we can contract by domain laws only
D M
M ¡
glued
toggleEGFK54 glued ¶ toggle NPOK
N!
glued 687 toggle 90: not defined
What is the negation of
an effect law?
an executability law?
Related Work
Modularity
[Pirri & Reiter 1999]: deterministic actions without
ramifications in Situation Calculus
[Amir 2000]: object-oriented concepts in Situation Calculus
[Zhang et al. 2002]: EPDL approach/normal form
[Lang et al. 2003]: computational complexity
[Kakas et al. 2005]: elaboration tolerance, concurrent
actions
[Ghilardi, Lutz & Wolter, KR’06]: uniform interpolation and
conservative extensions in ;=<
Related Work
Theory change
[Li& Pereira 1996]: motivations
[Liberatore 2000]: meta-results
[Eiter et al. 2005/06]: update in action languages
Summary
Claim
Consistency is not enough to evaluate a domain
description
The dynamic part of an action theory should not influence
the non-dynamic one (otherwise: problems)
Contribution
Fine-grained postulates of modularity
Algorithms to check/give hints on modularity
Satisfaction of modularity
More efficient reasoning
Important for updating theories [Herzig et al. ECAI’06]
Our results apply to every approach allowing for , ,
and
Ivan José Varzinczak What is a Good Domain Description?
Introduction
Main Results
Concluding Remarks
Summary
Claim
Consistency is not enough to evaluate a domain
description
The dynamic part of an action theory should not influence
the non-dynamic one (otherwise: problems)
Contribution
Fine-grained postulates of modularity
Algorithms to check/give hints on modularity
Satisfaction of modularity
More efficient reasoning
Important for updating theories [Herzig et al. ECAI’06]
Our results apply to every approach allowing for , ,
and
Ivan José Varzinczak What is a Good Domain Description?
Introduction
Main Results
Concluding Remarks
Summary
Contribution (cont.)
Semantics of action theory contraction
Domain-independent
Does not require extra information (preferences/epistemic
entrenchment relation/. . . )
Fully automatic
Completeness result: highlights importance of modularity
Summary
Future work
Fine tune contraction of effect laws
Contract by any formulas (not just laws)
Postulates about effect laws? about causation?
Summary
Future work
Fine tune contraction of effect laws
Contract by any formulas (not just laws)
Postulates about effect laws? about causation?
Example
Suppose a passport control system in an airport
Such a system is composed of many software components
One of them an ontology (knowledge base) about
passengers
All passengers must be controlled
Example (Ontology)
A passenger has a passport
European citizens have European passports
Foreigners have non-European passports
Someone with double citizenship is a foreigner and a
European
BC KC N
C
Passenger FHG passport I Ð » CL
Æ 2Citizen Fgå passport I Â EU JÊâ EU Å
EUcitizen Fgå passport I EU »
D
N
C
CE
Foreigner F*å passport I â EU »
C
C Æ 2Citizen Fgå passport I ä
2Citizen F Foreigner J EUcitizen
M
N
Æ 2Citizen F´â Passenger
Hence. . .
!
if we have 2Citizen j BINLADENk ,
this individual is not obliged to be controlled!
BC KC N
C
Passenger FHG passport I Ð » CL
Æ 2Citizen Fgå passport I Â EU JÊâ EU Å
EUcitizen Fgå passport I EU »
D
N
C
CE
Foreigner F*å passport I â EU »
C
C Æ 2Citizen Fgå passport I ä
2Citizen F Foreigner J EUcitizen
M
N
Æ 2Citizen F´â Passenger
Hence. . .
!
if we have 2Citizen j BINLADENk ,
this individual is not obliged to be controlled!
BC KC N
C
Passenger FHG passport I Ð » CL
Æ 2Citizen Fgå passport I Â EU JÊâ EU Å
EUcitizen Fgå passport I EU »
D
N
C
CE
Foreigner F*å passport I â EU »
C
C Æ 2Citizen Fgå passport I ä
2Citizen F Foreigner J EUcitizen
M
N
Æ 2Citizen F´â Passenger
Hence. . .
!
if we have 2Citizen j BINLADENk ,
this individual is not obliged to be controlled!
BC KC N
C
Passenger FHG passport I Ð » CL
Æ 2Citizen Fgå passport I Â EU JÊâ EU Å
EUcitizen Fgå passport I EU »
D
N
C
CE
Foreigner F*å passport I â EU »
C
C Æ 2Citizen Fgå passport I ä
2Citizen F Foreigner J EUcitizen
M
N
Æ 2Citizen F´â Passenger
Hence. . .
!
if we have 2Citizen j BINLADENk ,
this individual is not obliged to be controlled!
BC KC N
C
Passenger FHG passport I Ð » CL
Æ 2Citizen Fgå passport I Â EU JÊâ EU Å
EUcitizen Fgå passport I EU »
D
N
C
CE
Foreigner F*å passport I â EU »
C
C Æ 2Citizen Fgå passport I ä
2Citizen F Foreigner J EUcitizen
M
N
Æ 2Citizen F´â Passenger
Hence. . .
!
if we have 2Citizen j BINLADENk ,
this individual is not obliged to be controlled!
BC KC N
C
Passenger FHG passport I Ð » CL
Æ 2Citizen Fgå passport I Â EU JÊâ EU Å
EUcitizen Fgå passport I EU »
D
N
C
CE
Foreigner F*å passport I â EU »
C
C Æ 2Citizen Fgå passport I ä
2Citizen F Foreigner J EUcitizen
M
N
Æ 2Citizen F´â Passenger
Hence. . .
!
if we have 2Citizen j BINLADENk ,
this individual is not obliged to be controlled!
Our results. . .
can be applied in DL, too
Thank you!
Merci beaucoup !
Danke schön!
Choukran!
¡Muchas gracias!
Muito obrigado!
Example
l
loaded U{A shootBC3 alive
loaded R alive k U{A shootB alive
ðï L
jP3 o
H
hasGun U shoot IT "i[ 43 hasGun U{A shootBC "
nJ
b
® 3 hasGun O loaded UVA shootB3 alive
nJ
b
¬ ® 3 hasGun O loaded UVA shootB
but ¬ ® 3 hasGun O loaded UVA shootBC3 alive
Example
l
loaded U{A shootBC3 alive
loaded R alive k U{A shootB alive
ðï L
jP3 o
H
hasGun U shoot IT "i[ 43 hasGun U{A shootBC "
nJ
b
® 3 hasGun O loaded UVA shootB3 alive
nJ
b
¬ ® 3 hasGun O loaded UVA shootB
but ¬ ® 3 hasGun O loaded UVA shootBC3 alive
Example
l
loaded U{A shootBC3 alive
loaded R alive k U{A shootB alive
ðï L
jP3 o
H
hasGun U shoot IT "i[ 43 hasGun U{A shootBC "
nJ
b
® 3 hasGun O loaded UVA shootB3 alive
nJ
b
¬ ® 3 hasGun O loaded UVA shootB
but ¬ ® 3 hasGun O loaded UVA shootBC3 alive
Example
l
loaded U{A shootBC3 alive
loaded R alive k U{A shootB alive
ðï L
jP3 o
H
hasGun U shoot IT "i[ 43 hasGun U{A shootBC "
® jP3 hasGun
b
R loadedk UVA shootBC3 alive
but
ný ® jP3 hasGun R loadedk UVA shootBC
Example
l
loaded U{A shootBC3 alive
loaded R alive k U{A shootB alive
ðï L
jP3 o
H
hasGun U shoot IT "i[ 43 hasGun U{A shootBC "
® jP3 hasGun
b
R loadedk UVA shootBC3 alive
but
ný ® jP3 hasGun R loadedk UVA shootBC