A Framework of Cooperative Agents with Implicit Support for Ontologies CIA 2006

advertisement
The Context
OWL as a Content Language
The ALCBI Language
A Framework of Cooperative Agents with Implicit
Support for Ontologies
CIA 2006
Edinburgh, UK
R. Cenk Erdur and İnanç Seylan
Department of Computer Engineering
Ege University
September 13, 2006
The Context
OWL as a Content Language
Outline
1
The Context
2
OWL as a Content Language
3
The ALCBI Language
The ALCBI Language
The Context
OWL as a Content Language
The ALCBI Language
Seagent
A Java-based multi-agent programming framework for the
Semantic Web.
Based on FIPA (Abstract) Architecture.
ACL: FIPA ACL, Content language: OWL.
Supports a HTN planner (in its current form a task executor).
Has predefined plans for (composite) semantic web service
execution.
Has a unit (plan) testing framework (SUnit) to enable the
verification of task executions in a programmatic way. Test
different type of queues, the outcomes of actions, etc.
The Context
OWL as a Content Language
The ALCBI Language
Motivation
This work is a result of self-criticism. Basically, OWL is inadequate:
to be used within an ACL based on speech-act theory where
communication is explained in terms of the mental attitudes
of agents,
to reason about agents’ goals that lead to the selection,
rational decomposition and execution of plans.
OWL can describe the world agents are acting in but what
about an agents’ subjective perception of this world?
The Context
OWL as a Content Language
The ALCBI Language
Inconsidered Problems
We don’t consider in general:
The problems of mental agency in an open, highly
competetive system. Consider cooperative systems first.
Mental agency vs. social agency.
Formal verification of multi-agent systems.
The Context
OWL as a Content Language
Outline
1
The Context
2
OWL as a Content Language
3
The ALCBI Language
The ALCBI Language
The Context
OWL as a Content Language
Schrödinger’s cat is alive
FIPA SL
( a l i v e ( c a t : owner s c h r ö d i n g e r ) )
DAML+OIL
<Cat r d f : ID =”s c h r ö d i n g e r s c a t ”>
<owner>S c h r ö d i n g e r </owner>
<s t a t u s >a l i v e </ s t a t u s >
</Cat>
The ALCBI Language
The Context
OWL as a Content Language
Cats are animals
FIPA SL
( f o r a l l ?x ( i m p l i e s ( cat ?x ) ( animal ?x ) ) )
DAML+OIL
<daml : C l a s s r d f : ID=c a t >
< r d f s : s u b C l a s s O f r d f : r e s o u r c e=#a n i m a l >
</daml>
The ALCBI Language
The Context
OWL as a Content Language
All red things
FIPA SL
( a l l ?x ( red ?x ) )
OWL (Zou)
<o wl : C l a s s r d f : ID =”a l l r e d t h i n g ”>
<o wl : i n t e r s e c t i o n O f r d f : p a r s e T y p e =”C o l l e c t i o n ”>
<o wl : C l a s s r d f : a b o u t=”#Thing ”/>
<o wl : R e s t r i c t i o n >
<o wl : o n P r o p e r t y r d f : r e s o u r c e =”# h a s C o l o r ”/>
<o wl : h a s V a l u e r d f : r e s o u r c e =”#Red”/>
</o wl : R e s t r i c t i o n >
</o wl : i n t e r s e c t i o n O f >
</o wl : C l a s s >
The ALCBI Language
The Context
OWL as a Content Language
Steve believes X
FIPA SL
(B s t e v e X)
DAML+OIL (Botelho and Zou)
<P e r s o n r d f : ID =”s t e v e ”>
<h a s P r o p o s i t i o n >
< B e l i e f r d f : ID =”s t e v e b e l i e f 1 ”>
< b e l i e v e >t r u e </ b e l i e v e >
<S t a t e m e n t >X</S t a t e m e n t >
</ B e l i e f >
</ h a s P r o p o s i t i o n >
</P e r s o n >
The ALCBI Language
The Context
OWL as a Content Language
The ALCBI Language
Conclusions about OWL
OWL is good for expressing ontological knowledge (class and
individual declarations).
One can not express mental attitudes in OWL’s
model-theoretic semantics.
It is not clear how mental attitudes could be conveniently
composed.
It does not allow referential expressions (SWRL and SPARQL).
The Context
OWL as a Content Language
Outline
1
The Context
2
OWL as a Content Language
3
The ALCBI Language
The ALCBI Language
The Context
OWL as a Content Language
The ALCBI Language
Our Methodology
The difficulties are inherent from the expressivity limitations of the
formalism of OWL, namely description logics. Our solution:
We extend it with modal operators of belief and intention in
order to express these mental attitudes in a multi-agent
setting.
We define the rational balance between intentions and beliefs
as a set of axioms.
We define agent interaction semantics by listing the axioms of
communication.
The Context
OWL as a Content Language
The ALCBI Language
Formula
Definition
Let C and D be concepts, R a role name, x and y object names
and m a (possibly empty) sequence of modal operators from
{B, I } × AG. Then axioms of the form C = D (terminological),
R(x, y ) and C (x) (assertional) are atomic formulas. If ϕ and ψ are
formulas then so are mϕ, ¬ϕ, and ϕ ∧ ψ.
The Context
OWL as a Content Language
The ALCBI Language
Example Formulas
Bi (Door (x) ∧ isOpen(x, ”true”))
- i believes that the door x is open.
(Parent = Mother t Father ) ∧ Bspoiled (Parent v MoneySource)
- Parent is a mother or a father, objectively. But a spoiled kid
sees his/her parents as a money source.
Ij (TurnOffAction(to4347) ∧ light(to4347, l654))
- j intends to do the turn off action on light l654.
IhappyAgent (Person v Happy )
- happyAgent intends that everyone is happy.
Bi (Agent(i) ∧ Tired(i)) ∧ Ii (Sleeping (i))
- Agent i believes that it is tired and intends to sleep.
Bj Ii (Sleeping (i))
- Agent j believes that agent i intends to sleep.
The Context
OWL as a Content Language
The ALCBI Language
Frame
Definition
A frame for ALCBI is a pair F = (W , S) such that a) W is a set
of worlds; and b) S is a collection of structures on W consisting of
a) an accessibility relation Bi , which is a function Bi : W → 2W
W
and b) a function Ii : W → 22 for every i ∈ AG.
The Context
The ALCBI Language
OWL as a Content Language
Model
Definition
A model for ALCBI is a pair M = (F, KI ), where F is a frame for
ALCBI and KI is a function over W . The K-interpretation KI
consists of a domain ∆KI and an interpretation function ·KI . ∆KI
is the union of non-empty domains ∆KI (w ) for all worlds w ∈ W .
The interpretation function ·KI associates with each w a structure
·KI (w ) =
∆KI (w ) , RKI (w ) , C KI (w ) , X KI (w )
where ∆KI (w ) is the domain of w , RKI (w ) are binary relations on
∆KI (w ) , C KI (w ) subsets of ∆KI (w ) , and X KI (w ) are objects in
∆KI (w ) .
The Context
OWL as a Content Language
The ALCBI Language
Satisfiability
Definition
For a model M = (F, KI ) and a world w ∈ W , a formula ϕ is
satisfied (written as |=M
w ϕ) in the following way:
|=M
w
|=M
w
|=M
w
|=M
w
|=M
w
|=M
w
C =D
C (x)
R(x, y )
Bi ϕ
Ii ϕ
¬ϕ
iff
iff
iff
iff
iff
iff
C KI (w ) = D KI (w )
x KI (w ) ∈ C KI (w )
(x KI (w ) , y KI (w ) ) ∈ R KI (w )
Bi (w ) ⊆ kϕkM
kϕkM ∈ Ii (w )
6|=M
w ϕ
The Context
The ALCBI Language
OWL as a Content Language
Relations between Mental Attitudes
From Herzig et. al.
Ii ϕ
Ii ϕ
(Ii Bi ϕ ∧ Bi ¬ϕ)
Bi Ii ϕ
Bi ¬Ii ϕ
→
→
↔
↔
↔
Bi ¬ϕ
Ii Bi ϕ
Ii ϕ
Ii ϕ
¬Ii ϕ
(A1)
(A2)
(A3)
(A4)
(A5)
The Context
OWL as a Content Language
The ALCBI Language
Decidability of the Satisfiability Problem
Modal operators are only allowed in front of terminological
axioms, assertional axioms and other formulae.
Decidability = DL component decidability & Modal
component decidability
The formula Ii ϕ is translated into multi-modal formula
¬I i,1 ¬(I i,2 ϕ ∧ I i,3 ¬ϕ) where Ii,1 , Ii,2 and Ii,3 are normal modal
operators. Hence in such a translated frame, S contains the
three accessibility relations for intention Ii, , Ii, , Ii, along
with Bi for each i ∈ AG.
The algorithm generates the set of possible worlds first. Then
tests the satisfiability of DL axioms in each world.
The Context
OWL as a Content Language
The ALCBI Language
Current Position and Goals
The ALCBI reasoner implementation is a front end to the
Pellet DL reasoner (can use other DL reasoners through a
Java interface).
Code is 100% in Java.
The reasoner is at alpha stage (literals, final syntax decisions,
better documentation).
The goal is to create generic plans for plan selection that uses
the reasoner.
Implement an application from the Linux package
management domain.
Integrate dynamic logic characteristics to formally represent
actions and reason about them?
The Context
OWL as a Content Language
The ALCBI Language
Thank You
You can download the ALCBI reasoner version 0.1a from:
http://aegeants.ege.edu.tr/inanc/alcbi/
Download