Consistency Verification of a Non-monotonic Deductive
System based on OWL Lite
Jaime Ram
´
ırez and Ang
´
elica de Antonio
Technical University of Madrid
Madrid, Spain
Abstract. The aim of this paper is to show a method that is able to detect a partic-
ular class of semantic inconsistencies in a deductive system (DS). A DS verified
by this method contains a set of production rules, and an OWL Lite ontology that
defines the problem domain. The antecedent of a rule is a formula in Disjunctive
Normal Form, which encompasses first-order literals and linear arithmetic con-
straints, and the consequent is a list of actions that can add or delete assertions in
a non-monotonic manner. By building an ATMS-like theory the method is able
to give a specification of all the initial Fact Bases (FBs), and the rules that would
have to be executed from these initial FBs to produce an inconsistency.
1 Introduction
The purpose of this paper is to present a method for verifying the semantic consistency
of deductive systems (DSs) that deal with production rules and an OWL Lite ontology
1
.
One of the most interesting facets of the proposed method is that it is able to deal with
non monotonic reasoning. As the consequent of the production rules is allowed to add
and delete facts about individuals in the Fact Base (FB), the behaviour of the verified
DS is non-monotonic.
Some methods or tools intended to verify the consistency of a DS (mostly rule based
systems) build a model of the DS (Graph, Petri Net, etc.), and execute the model for each
valid input, in order to identify possible inconsistencies during the reasoning process.
This approach in many cases turns to be computationally very costly. Thus, we decided
to follow another approach in which the starting point is one of the inconsistencies that
might be possibly deduced by the verified DS, and the goal is to compute a description
of the initial FBs in which the DS would deduce that inconsistency. This approach takes
some ideas from the ATMS designed by de Kleer [1] since it uses the concept of label as
a way to represent a description of a set of FBs. Other methods for verifying rule-based
systems that follow a similar approach were proposed in [2] [3].
Section 2 of this paper mentions other methods that have also been proposed to
verify the consistency of a hybrid DS or non-monotonic DS. In section 3, some aspects
are explained related to the DSs to be verified. Section 4 describes how to specify the
inconsistencies that are verified by this method. Section 5 explains how this method
1
http://www.w3.org/TR/owl-features/
Ramírez J. and de Antonio A. (2005).
Consistency Verification of a Non-monotonic Deductive System based on OWL Lite.
In Proceedings of the 3rd International Workshop on Modelling, Simulation, Verification and Validation of Enterprise Information Systems, pages 19-28
DOI: 10.5220/0002573400190028
Copyright
c
SciTePress
specifies the way in which a DS can deduce an inconsistency, if possible. In section
6, the procedure for detecting an inconsistency is outlined and illustrated by means of
a simple example. We end with some conclusions and future works derived from our
work.
2 Previous Work
There are not many methods that are able to deal with the verification of non-monotonic
DSs based on rules. Among them, we can highlight Antoniou’s method [4], which anal-
izes DSs expressed in default logic, and Wu & Lee’s method [5], which tests DSs with
production rules under the closed world assumption. While Antoniou’s method is in-
spired on tabular methods for monotonic DSs, Wu & Lee’s method models the DS as
a high level extended Petri net. Analogously, only a few methods have been proposed
that are able to verify hybrid systems, that is, systems that combine two different knowl-
edge representation formalisms, one for representing the problem domain, and another
for representing the inference knowledge. The first work that dealt with hybrid systems
was that of Lee & O’Keefe [6], which characterized a set of new types of anomalies
that appear as a result of considering the subsumption relationship among literals in
different rules. In order to detect those anomalies, Lee & O’Keefe presented a method
with a tabular approach. Finally, we will mention another method to verify hybrid sys-
tems, which was proposed by Levy and Rousset [7], and is the most advanced method
proposed to date to test hybrid systems. This method can be applied to hybrid systems
expressed in the CARIN language [8], which is a language that combines the flexibility
of the Horn clauses with the expressiveness of the ALN CR Description Logic (DL).
The method provided is grounded in theoretical results related to the query containment
problem, studied in the database literature. However, the main drawback of this method
is that it is required to evaluate the same query for each valid input, and this process may
imply a very high computational cost. Thus, to the best of our knowledge, no method,
except for ours, can deal with hybrid DSs with non monotonic reasoning. Moreover, our
method does not need to check the DS w.r.t. each possible input, since it begins from
the inconsistency, and then it tries to build the description of a conflictive input working
backwards.
3 Characteristics of the Deductive System to be Verified
The DS is made up of a Knowledge Base (KB) and an inference engine. In turn, the DS’s
KB consists of an OWL Lite ontology and a set of production rules. Let us describe each
part in more detail:
3.1 OWL Lite ontology
OWL language was intended to associate meaning to the web contents. In addition,
OWL allows for defining a shared vocabulary that several agents (which may be en-
dowed with DSs) can process and utilize to exchange information. OWL is a layered
20
language because it specifies a hierarchy of three sub-languages. They are, sorted by de-
creasing degree of expressiveness, OWL Full, OWL DL and OWL Lite. Although OWL
Full and OWL DL are more expressive than OWL Lite, the utilization of OWL Lite for
reasoning purposes is more recommended, as long as the entailment problem for OWL
Full is undecidable, and quite inefficient for OWL DL. An OWL Lite ontology consists
of a set of axioms and a set of facts. The axioms define some class taxonomies where
each class comprises a set of properties, whereas each fact defines an individual. More-
over, an OWL property is either a data-valued property or an object-valued property. In
OWL Lite, the cardinality of the properties must be 0 or 1. Moreover, a property can be
defined to be transitive, symmetrical or inverse of another property. The properties can
be arranged to make up taxonomies of properties, for example, the property HasFather
can be defined as a subproperty of the property HasRelative.
3.2 Production rules
The rule form considered by the method is: (l
11
, l
12
, ..., l
1w
,..., l
m1
, l
m2
, ..., l
ms
)
a
1
, a
2
, ..., a
t
where the antecedent part contains a disjunction of m conjunctions of lit-
erals (l
ij
), and the consequent part contains a list of actions (a
k
). A literal is an atom (or
atomic formula), a negated atom (except when the atom is Different(I1, I2), since it
must not be negated) or a linear arithmetic constraint. The variables must be preceded by
?, and the types of the variables can be determined taking into account the OWL Lite ax-
ioms. The kinds of atoms that can occur in the antecedent part are: SubClass(C1, C2),
Instance(I, C), SubP roperty(P 1, P 2), Dif f erent(I1, I2) and P ROP ERT Y (I1,
V ALUE), where each argument written in capital letters can be a variable or an ob-
ject name. Hence, P ROP ERT Y may be a variable representing any property. The
intended meaning for the literal ¬P ROP ERT Y (I1, V ALU E) is:
¬P ROP ERT Y (I1, V ALUE) V ALU E1(P ROP ERT Y (I1, V ALU E1)
Dif ferent(V ALU E, V ALUE1)). This meaning is established taking into account that the
maximum cardinality for the OWL Lite properties is 1. The linear arithmetic constraints
are intended to represent complex relationships over some data-valued properties on the
real domain. The syntax for these constraints is the same as the syntax specified for the
DL reasoner RACER
2
.
Basically, we admit actions in the consequent part to be addition actions or deletion
actions. By means of an addition action, a rule can add a fact to the FB, whereas by
means of a deletion action, a rule can remove a fact from the FB. Syntactically, an
action can take the form Add(Individual
Atom) or Del(Individual Atom) where
Individual Atom is either Instance(I, C) or P ROP ERT Y (I1, V alue).
3.3 Dynamic aspects
The DS is assumed to contain an inference engine able to execute the rules following a
forward or backward chaining. Furthermore, the DS must support a DL reasoner able to
deal with the entailment problem. As an OWL Lite ontology can be translated into the
description logic SHIF [9], DL reasoners such as RACER can be used to instantiate
2
http://www.cs.concordia.ca/ haarslev/racer/racer-manual-1-7-7.pdf
21
a literal of a rule on demand if possible. If a certain literal cannot be instantiated with
the facts entailed by the ontology, and the DS follows a backward chaining, the rule
inference engine will try to find a rule that deduces that literal.
When a rule is fired, we assume that all the actions belonging to the consequent of
the rule are executed sequentially.
Facts in a DS are classified into two categories: a deducible fact is a fact that is
obtained from the execution of the DS; and an external fact is a fact that cannot be
deduced by the DS and can only be obtained from an external source.
3.4 Non-Monotonicity and Inconsistency
Rules can introduce new facts in the FB, but they can also delete already existing facts.
This provides the DS’s designer with the capability of building DSs with non-monotonic
reasoning. Consequently, we could find production rules of the form p Add(¬p)
under Open World Assumption (OWA). This kind of rules (when p is assumed to hold)
are not admissible in a monotonic KB, since they are logical inconsistencies. If we
admit rules of the form p Add(¬p), we situate ourselves quite far from the concept
of inconsistency in monotonic DSs as defined in other works, so we are going to clarify
the meaning of inconsistency in this work:
A deductive tree T that deduces a conjunction of facts F and F
is tree consistent
iff:
1. T does not contain a set of contradictory external facts, or
2. the deductive subtree of T that deduces F must not deduce ¬F
in the end, and
vice versa.
This definition is not more than an structural property to be fulfilled by the deductive
trees built by the DS that we want to verify using the method. As the method will simu-
late the DS execution, it will discard any deductive process that implies the creation of
an invalid deductive tree. Let us see an example of an inconsistent set of rules. For the
sake of clarity, a simplified notation for the rules will be employed in this example. Ac-
cording to this notation, ¬p denotes a fact that is contradictory with the fact p w.r.t. the
ontology axioms. Let us take the production rules R1: r, s Del(p), Add(¬p); R2:
t Add(p); R3: ¬p Add(q) under OWA. In the figure 1 we can see the deductive
tree for a conjunction p q that is supposed to be the antecedent of another rule. The
facts p and q are deducible and all the other facts are external. Obviously (see rule R3),
in order to deduce q, ¬p must be deduced beforehand, and after having deduced ¬p it
is not possible to deduce p.
4 Specification of Semantic Inconsistencies
Each semantic inconsistency that must be considered is represented by means of an
Integrity Constraint (IC). The IC form is: x
1
x
2
...x
n
(l
1
(scope
1
) l
2
(scope
2
)
... l
k
(scope
k
)) ⇒⊥. A scope is associated with each literal to specify the kind of data
referenced in the literal (input or output). A literal with input scope states something
about the initial FB, while a literal with output scope states something about the final
FB (resulting after an execution of the DS).
22
Fig.1. Example of an invalid deductive tree
5 Specification of the Contexts
5.1 Describing Fact Bases
The proposed method will construct an object called subcontext to specify how the
initial FB must be and which deductive tree must be executed in order to cause an
inconsistency. There can be different initial FBs and different deductive trees that lead
to the same inconsistency. An object called context will gather all the different ways to
violate a given IC. Consequently, a context will be composed of n subcontexts. In turn,
a subcontext is defined as a pair (environment, deductive tree) where an environment
is composed of a set of metaobjects, and a deductive tree is a tree of rule firings. A
metaobject describes characteristics that one object that can be present in the FB should
have. For each type of OWL Lite object there will be a different type of metaobject. In
order to describe an OWL Lite object, a metaobject must include a set of constraints on
the characteristics of the OWL Lite object. All the metaobjects’s attributes are outlined
below:
Metaclass = (identifier, subclass
of, metaindividuals)
MetaObjectProperty = (identifier,
pairs
of metaindividuals, subproperty of)
MetaDataTypeProperty = (identifier,
pairs
of metaindividual-value, subproperty of)
MetaIndividual = (identifier, instance
of, objectproperties,
datatypeproperties, differentfrom)
MetaValue =(conditions)
Given that certain constraints expressed as arithmetic inequations can restrict the
datatype property values, a different kind of metaobject called condition will represent
these constraints. The attributes of a condition are expression and values. Lets see an ex-
ample of an environment describing a FB in which the formula Instance(?X, P erson)
F eels(?X, T iredness) T emperature(?X, ?temp) ?temp > 36.5 holds. If there
exists an OWL Lite object in the FB for each metaobject in the environment, that sat-
isfies all the requirements imposed on it, then the given formula will hold in the FB.
The environment is defined as {CLASS1, IND1, IND2, OPRO1, DTPRO1, VALUE1,
COND1} where:
CLASS1 = (P erson, , {IND1})
IN D1 = (, {CLASS1}, {OP RO1},
{DT P RO}, )
IN D2 = (T iredness, {OP RO1}, , )
OP RO1 = (Feels, {(IND1, IN D2)}, )
DT P RO1 = (T emperature,
{(IN D1, V ALUE1)}, )
V ALUE1 = ({CON D1})
COND1 = (”?temp > 36.5”, [V ALUE1])
23
CLASS1 is a metaclass, IND1 and IND2 are metaindividuals, OPRO1 is a metaOb-
jectProperty, DTPRO1 is a metaDataTypeProperty, VALUE1 is a metaValue1, and fi-
nally COND1 is a condition. As defining the metaobjects, two consecutive commas
represent an empty field.
5.2 Contexts Operations
We will define the following contexts operations: creation of a context, concatenation
of a pair of contexts and combination of a list of contexts. These operations will be em-
ployed by the method to compute the scenarios, as we will see later on. Before defining
the context operations, the goal object will be defined: A goal g is a pair (l, A) where l
is a literal and A is a set of metaobjects associated with the object names and variables
in l, that specifies the FBs in which the literal l is satisfied without using DL reasoning.
Moreover, a goal (l, A) is external/deducible iff the literal l is external/deducible.
a) Creation: a context with a unique subcontext is created from an external goal g =
(l, A): C(g) = {(E, EM P T Y
T REE)} where the environment E comprises all the
metaobjects included in A.
b) Concatenation of a pair of contexts: let C
1
and C
2
be a pair of contexts and Conc(C
1
,
C
2
) be the context resulting from the concatenation, then: Conc(C
1
, C
2
) = C
1
C
2
.
c) Combination of a list of contexts: Let C
1
, C
2
, ..., C
n
be the list of contexts, and
Comb(C
1
, C
2
, ..., C
n
) be the context resulting from the combination. The form of this
resulting context is: Comb(C
1
, C
2
, ..., C
n
) ={(E
k1
E
k2
... E
kn
, DT
k1
DT
k2
...
DT
kn
) s.t. (E
i
, DT
i
) C
i
}
c.1) Union of environments (E
i
E
j
): this operation consists of the union of the
sets of metaobjects E
i
and E
j
. After the union of two sets, it is necessary to check
whether any pair of metaobjects can be merged. A pair of metaobjects will be
merged if they contain a pair of constraints c
1
and c
2
, respectively, such that (c
1
c
2
)
entail that both metaobjects represent the same OWL Lite object. This will happen if
both metaobjects should have the same value in the identifier attribute according to
the ontology axioms. Finally, if the resulting environment represents an invalid ini-
tial FB, then this environment will also be discarded. This last check will be carried
out with the help of the DL reasoner RACER.
c.2) Combination of deductive trees (DT
i
DT
j
): let DT
i
and DT
j
be deductive
trees, then DT
i
DT
j
is the deductive tree that results from constructing a new tree
whose root node represents an empty rule firing, and whose two subtrees are DT
i
and DT
j
.
6 Computing the Context associated with an Integrity Constraint
The process to compute the context associated with an IC is divided into two steps.
Next, these steps will be explained.
6.1 First Step
The first step can be considered as a pre-processing of the set of rules. In the second step,
a backward chaining simulation of the real rule firings is carried out without making
24
calls to the DL reasoner (except for consistency checks in the union of environments,
see 5.2, or in the updates of the set of assumed individuals, as we will see in the second
step). However, in a real execution of the DS some literals in the rule antecedents may be
instantiated thanks to these DL reasoner calls. In order to fill this gap in the simulation,
some new rules derived from the ontology axioms and already existing rules are added
to the set of rules. In particular, the generation of new rules is related to the presence of
deduced literals in the rules. Sometimes, a rule adds a new fact to the FB that matches
a literal in another rule’s antecedent; in that case, we will say in the simulation that
the first rule can be chained with the literal. In other cases, a rule adds a new fact to
the FB that does not match directly a literal in a rule antecedent, but it actually does
it indirectly, because the new fact allows the DL reasoner to deduce another fact that
does match the literal. For the purpose of simulating properly this kind of inference
situations, the set of rules must be pre-processed. Next, we will explain how the new
rules are computed from the ontology axioms and already existing rules:
Deducing the literal ¬Instance(ID1, C):
According to the syntactical restrictions explained in 3.2, no rules can deduce directly
this kind of literals, but DS actually can indirectly deduce it these ways:
1. R(ID1, ID2), ¬subclass(C, Domain(R)) ¬Instance(ID1, C)
2. R(ID2, ID1), ¬subclass(C, Range(R)) ¬Instance(ID1, C)
Where R is any property. Thus, in each rule whose antecedent contains a conjunction
c where the deducible literal ¬Instance(ID1, C) occurs, the conjunction c will be
replaced with the new conjunctions:
Substitute(c, ¬Instance(ID1, C)”, R(ID1, ID2), ¬subclass(C, Domain(R))”) and,
Substitute(c, ¬Instance(ID1, C)”, R(ID2, ID1), ¬subclass(C, Range(R))”)
where the function Substitute(c, s1, s2) returns the conjunction resulting from replac-
ing the string s1 with the string s2 in the conjunction c.
Deducing the literal Instance(ID, C):
Following an analogous reasoning to the previous replacement, in each rule whose an-
tecedent contains a conjunction c where the deducible literal Instance(ID1, C) oc-
curs, two new conjunctions must be added:
Substitute(c, Instance(ID1, C)”, R(ID1, ID2), subclass(C, Domain(R))”) and,
Substitute(c, Instance(ID1, C)”, R(ID2, ID1), subclass(C, Range(R))”) .
Furthermore, given that any individual a that is instance of a class A is also instance
of any superclass of A, then another conjunction must be added:
Substitute(c, Instance(ID1, C)”, Instance(ID1, C1), subclass(C1, C)”)
Deducing transitive object properties:
If the object property R is defined to be transitive, then in each rule whose antecedent
contains a conjunction c where the deducible literal R(ID1, ID2) occurs, the new con-
junction must be added:
Substitute(c, R(ID1, ID2)”, R(ID1, ?X), R(?X, ID2”) st. the variable X does not occur
in the conjunction c.
Deducing symmetric object properties:
If the object property R is defined to be symmetric, then in each rule whose antecedent
contains a conjunction c where the deducible literal R(ID1, ID2) occurs, the new con-
25
junction must be added:
Substitute(c, R(ID1, ID2)”, R(ID2, ID1)”) .
Deducing inverse object properties:
If the object property R
1
is defined to be inverse of the object property R, then
in each rule whose antecedent contains a conjunction c where the deducible literal
R
1
(ID1, ID2) occurs, the new conjunction must be added:
Substitute(c, R
1
(ID1, ID2)”, R(ID2, ID1)”) and vice versa.
6.2 Second Step
Basically, the second step can be divided into twophases. In the first phase, the AND/OR
decision tree associated with the IC is expanded following a backward chaining simu-
lation of the real rule firings. The leaves of this tree are rules that only contain external
facts in their antecedents. At this point, the difference between a deductive tree and an
AND/OR decision tree should be explained. While a deductive tree can be viewed as
one way and only one way for achieving a certain goal (that is, for deducing a bound
formula or for firing a rule), an AND/OR decision tree comprises one or more deduc-
tive trees, therefore it specifies one or more ways to achieve a certain goal. During the
first phase, metaobjects are built corresponding with each variable of a rule/IC that is
being processed and each referenced OWL Lite name, and these metaobjects are propa-
gated from a rule to another one. In this propagation, some constraints are added to the
metaobjects due to the rule literals, and some constraints are removed from the metaob-
jects due to the rule actions, because any constraint deduced by an action is not required
to be satisfied by the initial FB any more. In addition to the metaobjects, a set of as-
sumed individuals (SAI) is propagated and updated. The aim of SAI is to warrant that
the expanding deductive tree fulfills the second condition of the T ree Consistency
definition (see 3.4). The first condition of the T ree Consistency definition is checked
in the union of environments (see 5.2) during the next phase.
Figure 2 shows an example with a rule R1 and an IC, as well as the deductive tree
expanded by the proposed method for this IC. In this example, the T property literals
are deducible, whereas the rest of literals are external. This figure also shows the names
of the metaobjects built for the variables and the OWL Lite names, as well as the two
propagations of metaobjects through the two goal-action chainings. We will follow the
trajectory of metaindividual I2 from the IC, where it is created for the variable ?X, to
the rule R1. In the IC, I2 is created as (, , {OP R1}, , ) (see the format of the metaobjects
in section 5.1). Then, the reference to OP R1 is removed from I2 in the first chaining,
because the action deduces a pair of the object property T in which I2 is involved. Next,
in the rule R2, I2 is required to appear in two pairs, one of the object property R, and an-
other of the object property T ; therefore I2 is updated to I2 = (, , {OP R1, OP R2}, , ).
Now, I2 is involved in another chaining, this time from R2 to R1, and in this chain-
ing the reference to OP R1 is removed from I2 due to the simulation of the action
effect. Finally, in the rule R1, a reference to the object property R and a constraint
stating that the individual I2 is an instance of class A are added to I2. For the ex-
ample of the figure 2, a SAI is created in the IC, so that SAI = {SubP roperty
(DP R1, DP R2), DP R1(I1, V 1), V 1 > 5, T (I2, I3), Dif f erent (I3, a)}. Then,
in the first chaining the action removes T (I2, I3) from SAI, and when SAI gets to
26
Fig.2. Deductive tree of the case study
the conjunction of R2, it is updated so that SAI = {SubP roperty (DP R1, DP R2),
DP R1(I1, V 1), V 1 > 5, V (a, I3), T (I2, I5), Different(I3, a)}. Finally, in R1,
SAI = {SubP roperty(DP R1, DP R2), DP R1(I1, V 1), V 1 > 5, V (a, I3),
Different(I3, a), Instance(I2, C1)}. As we can just see in this paragraph, the ex-
ample of the figure 2 does not raise any inconsistency propagating SAI. However, if the
external literal V (a, a)(I) was added to IC, then in the antecedent of R2, SAI would
be {SubP roperty(DP R1, DP R2), DP R1(I1, V 1), V 1 > 5, V (a, I3), T (I2, I5),
Different(I3, a), V (a, a)}, which is inconsistent because it forces the object prop-
erty V to have two pairs (a, I3) and (a, a) st. Dif f erent (I3, a). If SAI turns to be
inconsistent w.r.t. the ontology axioms, the T ree Consistency property does not hold
for the current deductive tree, and then the current rule must be discarded.
In the second phase, the AND/OR decision tree is contracted by means of context op-
erations, so that metaobjects in external goals and conditions related to metaobjects
in external goals are inserted in the subcontexts of the context associated with the
IC. Basically, the creation operation is employed to work out the context associated
with an external goal; the combination operation is employed to work out the con-
text associated with a conjunction of literals from the contexts associated with the
literals; and the concatenation operation is employed to work out the context associ-
ated with a disjunction from the contexts associated with the formulas involved in the
disjunction. Let us see the context associated with the IC in the example of the fig-
ure 2: C(IC) = {SUBC1} = {({C1, I1, I2, I3, I5, I5
, I6, DP R1, DP R2, OP R3,
OP R5 , V 1 , CON D1}, tree(R1, [tree(R2, [EMP T Y
T REE])]))} where:
C1 = (A, , {I2})
I1 = (, , , {DP R1}, )
I2 = (, {A, R1)}, {OP R5}, , )
I3 = (, , {OP R3}, , {I6})
I5
= (, , , , {I5})
I5 = (, , {OP R5}, , {I5
})
I6 = (a, , {OP R3}, , {I3})
/ I6 = I4 + I3
/
DP R1 = (, {(I1, V 1)}, {DP R2})
DP R2 = (D, , )
OP R3 = (V, { (I6, I3)}, )
OP R5 = (R, {(I2, I5),
(I2, I5
)}, )
/ OP R5 = OP R2 + OP R4 /
V 1 = ({CON D1})
COND1 = (
′′
?U > 5
′′
, {V 1})
27
The two phases of the second step are explained in detail for a frame-like knowledge
representation formalism called CCR-2 in [10].
7 Conclusion and Future Work
In this paper, a formal method to verify the consistency of the reasoning process of a
DS has been presented. It is noteworthy that the DS to be verified encompasses a KB,
endowed with an OWL Lite ontology and a set of production rules, which permits the
representation of non-monotonic reasoning and arithmetic constraints. So far, most of
the efforts dedicated to the consistency verification of DSs have focused on the veri-
fication of a set of rules ignoring the domain knowledge. One of the few works that
has dealt with the verification of both the set of rules and the domain knowledge was
proposed in [7]. That work explains how to verify DSs whose domain knowledge is
expressed in a rich language based on a DL. In our approach, however, we have chosen
to sacrifice expressiveness (OWL Lite instead of OWL DL) in favour of efficiency, so
that the proposed method can be applied to large systems; one of our next steps will be
to show this empirically. We are currently studying more deeply if the proposed pre-
processing rules in 6.1 is exhaustive, so as to ensure the completeness of the simulation
in the second step (see 6.2). Besides, we are working on an extension of the proposed
method that verifies the reasoning module of a deliberative agent cohabiting a dynamic
environment with other agents. In this dynamic environment the truth value of some
external facts may change during the reasoning process as a result of the reception of
new messages or stimuli coming from the environment of the verified agent.
References
1. de Kleer, J.: An assumption based TMS. Artificial Intelligence 28 (1986) 127–162
2. Rousset, M.: On the consistency of knowledge bases: The COVADIS system, Proceedings
ECAI-88, Munich, Alemania (1988) pp. 79-84.
3. Ginsberg, A.: Knowledge-base reduction: A new approach to checking knowledge bases for
inconsistency and redundancy, Proceedings of the AAAI-88 (1988) pp. 585-589.
4. Antoniou, G.: Verification and correctness issues for nonmonotonic knowledge bases. Inter-
national Journal of Intelligent Systems 12 (1997) 725–738
5. Wu, C.H., Lee, S.J.: Knowledge verification with an enhanced high-level petri-net model.
IEEE Expert 12 (1997) 73–80
6. Lee, S., O’Keefe, R.M.: Subsumption anomalies in hybrid knowledge based systems. Inter-
national Journal of Expert Systems 6 (1993) 299–320
7. Levy, A.Y., Rousset, M.: Verification of knowledge bases on containment checking. Artificial
Intelligence 101 (1998) 227–250
8. Levy, A.Y., Rousset, M.: CARIN: A representation language combining horn rules and
description logics, Proceedings ECAI’96 (1996)
9. Horrocks, I., Patel-Schneider, P.F.: Three theses of representation in the semantic web. In:
Proc. of the Twelfth International World Wide Web Conference (WWW 2003), ACM (2003)
39–47
10. Ram
´
ırez, J., de Antonio, A.: Knowledge base semantic verification based on contexts propa-
gation, Notes of the AAAI-01 Symposium on Model-based Validation of Intelligence (2001)
http://ase.arc.nasa.gov/mvi/abstracts/index.html.
28