An ontological approach to the construction
of problem-solving models
ABSTRACT knowledge and reasoning) fulfils an important function: it
Our ongoing work aims at defining an ontology-centered must allow problem-solving methods to be specified in
approach for building expertise models for the Common- terms which are independent of particular application do-
KADS methodology. This approach (which we have mains, thus facilitating re-use of these generic methods.
named "OntoKADS") is founded on a core problem- Even today, the extra-ontological status of this component
solving ontology which distinguishes between two concep- does not seem to have progressed. In fact, methods which
tualization levels: at an object level, a set of concepts en- recommend the use of ontologies all resort to a syntactic
able us to define classes of problem-solving situations, and ploy - transformation rules in PROTÉGÉ and bridges in
at a meta level, a set of meta-concepts represent modeling UPML - to link domain knowledge and reasoning [٦].
primitives. In this article, our presentation of OntoKADS Here, we re-examine this presupposition. We show that
will focus on the core ontology and, in particular, on roles - recent progress in the field of formal ontologies enables
the primitive situated at the interface between domain one to account for this component in semantic terms,
knowledge and reasoning, and whose ontological status is within a coherent ontological framework.
still much debated. We first propose a coherent, global,
In our previous work [۱٤], we suggested drawing a distinc-
ontological framework which enables us to account for this
tion between two types of role: roles played by objects (e.g.
primitive. We then show how this novel characterization of
Physician, Student) and those played by concepts (e.g. Hy-
the primitive allows definition of new rules for the con-
pothesis, Sign). After having likened the latter to Common-
struction of expertise models.
KADS' Knowledge roles, we gave them the status of a meta-
property, i.e. along the same lines as Guarino's proposal
Categories and Subject Descriptors (making the role concept appear in an ontology of univer-
I.۲٫٤ [Artificial Intelligence]: Knowledge Representation sals [۱۲]). However, in ۱۹۹۹, we were not in a position to
Formalisms and Methods provide a coherent ontological framework to account for
CommonKADS expertise models in their entirety.
Today, we are tackling this issue by using the OntoKADS
method [۳]. OntoKADS benefits from a broad range of
Keywords recent work and progress in i) clarifying the notion of role
Knowledge Engineering and Modeling methodologies, in ontological terms [۱۸][۱٦] ; ii) defining (via use of rich
Problem-solving models, CommonKADS, Ontology Engi- axiomatic) a top-level ontology such as DOLCE, the struc-
neering, Foundational ontologies, DOLCE, Core problem- turing principles of which are explicit [۱۷] ; iii) integrating
solving ontologies, OntoKADS, Ontologies of mental ob- mental representations (reified entities) like Propositions
jects, I&DA, COM [۸] / Descriptions [۱۱] into an ontology; and iv) defining an
ontology of meta-properties based on a set of clearly iden-
tified primitives (rigidity, dependence, identity) [۱۲].
INTRODUCTION Hence, we now possess an ontological tool-box which is
Since the late ۱۹۹۰s, the construction of explicit ontologies both necessary and sufficient for making this type of pro-
has been considered as a promising way of improving the posal.
knowledge engineering process: the elaboration of domain,
In the following sections of this article۱, we first present an
task and method ontologies early on in the design of prob-
overview of the OntoKADS method and then focus on its
lem-solving models was recommended [۲۲]. At the same
core problem-solving ontology, with particular emphasis
time, however, certain components of these problem-
on the part of the ontology that deals with roles.
solving models - notably roles - appeared to have been
excluded from ontological treatment [۲۳]. This component
(referred to as a Knowledge role in the CommonKADS
method [۲۱] and situated at the interface between domain ۱
This article is an extended version of [۳].
, OVERVIEW OF OntoKADS
Application knowledge
Expertise Model
Our proposal consists of a methodology called OntoKADS OntoKADS Task model
which, to a great extent, likens the construction of expertise
models to the construction of ontologies. The method com- Task
Task
Method
prises two main steps (see Figure 1). Step 1 Step 2
In a first step, the knowledge engineer develops a problem-
Labeled Inference Model
solving-driven application ontology whose concepts are
problem-solving
labeled by modeling primitives. To do this, the method application ontology
Domain Model
uses an ontology (also named OntoKADS) composed of
two main sub-ontologies:
• A core problem-solving sub-ontology which enables Figure 1. Main steps in the OntoKADS method.
the engineer to define (by specialization) the applica-
tion's concepts and specific reasonings. This sub- HOW OntoKADS EXTENDS DOLCE
ontology extends the high-level DOLCE ontology (De-
scriptive Ontology for Linguistic and Cognitive Engi- DOLCE and the notion of knowledge
neering) [17]2. The OntoKADS ontology is defined as an extension of the
• A meta-level sub-ontology coding the modeling primi- DOLCE ontology, which means that OntoKADS' concepts
tives, which allows the engineer to label the concepts and relations are defined by specialization of the abstract
from the previous ontology by using meta-properties concepts and relationships present in DOLCE. DOLCE's
which represent modeling primitives. This type of domain, i.e. the set of entities classified by the ontology's
practice is analogous to the labeling advocated in the concepts (referred to as a set of particulars, PT) is divided
OntoClean method [13]3. into four sub-domains. For the purposes of this article, we
consider only two of these here (see Figure 2):
A software module (see Figure 1) then automatically trans-
lates this labeled ontology into three subcomponents of an • The endurants (ED). These are entities which "are in
expertise model which resembles CommonKADS [21]: a time" and are wholly present whenever they are pre-
domain model, an inference model and a task model. This sent (objects, substances and also ideas). Within this
translation principally involves the extraction and reorgani- sub-domain, one can distinguish physical objects
zation of representations. (POB) and non-physical objects (NPOB), depending
on whether or not the objects have a direct physical lo-
In a second step, the knowledge engineer further specifies cation.
the problem-solving methods linked to the tasks which
he/she has identified. • The perdurants (PD). These are entities which "occur
in time" but are only partially present at any time they
A software environment for running this method is cur-
are present (events and states). Within this sub-
rently being developed (in the Conclusion section, we ex-
domain, one can distinguish events (EV) and statives
plain our choice of software tools). In the remainder of this
(STV) according to a "cumulativity" principle: the
article, our presentation of OntoKADS will concentrate on
mereological sum of two instances of a stative (for ex-
describing the ontology's content. We shall first show how
ample, "being seated") is an instance of the same type,
the OntoKADS core ontology extends the DOLCE high-
which is not the case for the sum of instances of
level ontology.
events, for example the K-CAP 2003 and K-CAP 2005
conferences. The latter (not being atomic) are consid-
ered to be accomplishments (ACC).
The main relationship between endurants and perdurants is
that of participation, with PC(x,y,t) holding for: "x (neces-
sarily an endurant) participates in y (necessarily a per-
durant) at time t". For example, the co-authors of this arti-
cle (endurants) participated in the drafting of the text (a
perdurant).
2
This ontology was chosen mainly for the reasons outlined in the Intro- More particularly in terms of the domain of knowledge
duction but also because it integrates a class of mental objects which which directly involves OntoKADS, DOLCE's commit-
turn out to be very important for analyzing problem-solving knowledge.
3
Even though the goals are different a priori (building an expertise model
vs. verifying the logical coherence of subsumption links), we shall see
that the labeling meta-properties are of the same nature.
of problem-solving models
ABSTRACT knowledge and reasoning) fulfils an important function: it
Our ongoing work aims at defining an ontology-centered must allow problem-solving methods to be specified in
approach for building expertise models for the Common- terms which are independent of particular application do-
KADS methodology. This approach (which we have mains, thus facilitating re-use of these generic methods.
named "OntoKADS") is founded on a core problem- Even today, the extra-ontological status of this component
solving ontology which distinguishes between two concep- does not seem to have progressed. In fact, methods which
tualization levels: at an object level, a set of concepts en- recommend the use of ontologies all resort to a syntactic
able us to define classes of problem-solving situations, and ploy - transformation rules in PROTÉGÉ and bridges in
at a meta level, a set of meta-concepts represent modeling UPML - to link domain knowledge and reasoning [٦].
primitives. In this article, our presentation of OntoKADS Here, we re-examine this presupposition. We show that
will focus on the core ontology and, in particular, on roles - recent progress in the field of formal ontologies enables
the primitive situated at the interface between domain one to account for this component in semantic terms,
knowledge and reasoning, and whose ontological status is within a coherent ontological framework.
still much debated. We first propose a coherent, global,
In our previous work [۱٤], we suggested drawing a distinc-
ontological framework which enables us to account for this
tion between two types of role: roles played by objects (e.g.
primitive. We then show how this novel characterization of
Physician, Student) and those played by concepts (e.g. Hy-
the primitive allows definition of new rules for the con-
pothesis, Sign). After having likened the latter to Common-
struction of expertise models.
KADS' Knowledge roles, we gave them the status of a meta-
property, i.e. along the same lines as Guarino's proposal
Categories and Subject Descriptors (making the role concept appear in an ontology of univer-
I.۲٫٤ [Artificial Intelligence]: Knowledge Representation sals [۱۲]). However, in ۱۹۹۹, we were not in a position to
Formalisms and Methods provide a coherent ontological framework to account for
CommonKADS expertise models in their entirety.
Today, we are tackling this issue by using the OntoKADS
method [۳]. OntoKADS benefits from a broad range of
Keywords recent work and progress in i) clarifying the notion of role
Knowledge Engineering and Modeling methodologies, in ontological terms [۱۸][۱٦] ; ii) defining (via use of rich
Problem-solving models, CommonKADS, Ontology Engi- axiomatic) a top-level ontology such as DOLCE, the struc-
neering, Foundational ontologies, DOLCE, Core problem- turing principles of which are explicit [۱۷] ; iii) integrating
solving ontologies, OntoKADS, Ontologies of mental ob- mental representations (reified entities) like Propositions
jects, I&DA, COM [۸] / Descriptions [۱۱] into an ontology; and iv) defining an
ontology of meta-properties based on a set of clearly iden-
tified primitives (rigidity, dependence, identity) [۱۲].
INTRODUCTION Hence, we now possess an ontological tool-box which is
Since the late ۱۹۹۰s, the construction of explicit ontologies both necessary and sufficient for making this type of pro-
has been considered as a promising way of improving the posal.
knowledge engineering process: the elaboration of domain,
In the following sections of this article۱, we first present an
task and method ontologies early on in the design of prob-
overview of the OntoKADS method and then focus on its
lem-solving models was recommended [۲۲]. At the same
core problem-solving ontology, with particular emphasis
time, however, certain components of these problem-
on the part of the ontology that deals with roles.
solving models - notably roles - appeared to have been
excluded from ontological treatment [۲۳]. This component
(referred to as a Knowledge role in the CommonKADS
method [۲۱] and situated at the interface between domain ۱
This article is an extended version of [۳].
, OVERVIEW OF OntoKADS
Application knowledge
Expertise Model
Our proposal consists of a methodology called OntoKADS OntoKADS Task model
which, to a great extent, likens the construction of expertise
models to the construction of ontologies. The method com- Task
Task
Method
prises two main steps (see Figure 1). Step 1 Step 2
In a first step, the knowledge engineer develops a problem-
Labeled Inference Model
solving-driven application ontology whose concepts are
problem-solving
labeled by modeling primitives. To do this, the method application ontology
Domain Model
uses an ontology (also named OntoKADS) composed of
two main sub-ontologies:
• A core problem-solving sub-ontology which enables Figure 1. Main steps in the OntoKADS method.
the engineer to define (by specialization) the applica-
tion's concepts and specific reasonings. This sub- HOW OntoKADS EXTENDS DOLCE
ontology extends the high-level DOLCE ontology (De-
scriptive Ontology for Linguistic and Cognitive Engi- DOLCE and the notion of knowledge
neering) [17]2. The OntoKADS ontology is defined as an extension of the
• A meta-level sub-ontology coding the modeling primi- DOLCE ontology, which means that OntoKADS' concepts
tives, which allows the engineer to label the concepts and relations are defined by specialization of the abstract
from the previous ontology by using meta-properties concepts and relationships present in DOLCE. DOLCE's
which represent modeling primitives. This type of domain, i.e. the set of entities classified by the ontology's
practice is analogous to the labeling advocated in the concepts (referred to as a set of particulars, PT) is divided
OntoClean method [13]3. into four sub-domains. For the purposes of this article, we
consider only two of these here (see Figure 2):
A software module (see Figure 1) then automatically trans-
lates this labeled ontology into three subcomponents of an • The endurants (ED). These are entities which "are in
expertise model which resembles CommonKADS [21]: a time" and are wholly present whenever they are pre-
domain model, an inference model and a task model. This sent (objects, substances and also ideas). Within this
translation principally involves the extraction and reorgani- sub-domain, one can distinguish physical objects
zation of representations. (POB) and non-physical objects (NPOB), depending
on whether or not the objects have a direct physical lo-
In a second step, the knowledge engineer further specifies cation.
the problem-solving methods linked to the tasks which
he/she has identified. • The perdurants (PD). These are entities which "occur
in time" but are only partially present at any time they
A software environment for running this method is cur-
are present (events and states). Within this sub-
rently being developed (in the Conclusion section, we ex-
domain, one can distinguish events (EV) and statives
plain our choice of software tools). In the remainder of this
(STV) according to a "cumulativity" principle: the
article, our presentation of OntoKADS will concentrate on
mereological sum of two instances of a stative (for ex-
describing the ontology's content. We shall first show how
ample, "being seated") is an instance of the same type,
the OntoKADS core ontology extends the DOLCE high-
which is not the case for the sum of instances of
level ontology.
events, for example the K-CAP 2003 and K-CAP 2005
conferences. The latter (not being atomic) are consid-
ered to be accomplishments (ACC).
The main relationship between endurants and perdurants is
that of participation, with PC(x,y,t) holding for: "x (neces-
sarily an endurant) participates in y (necessarily a per-
durant) at time t". For example, the co-authors of this arti-
cle (endurants) participated in the drafting of the text (a
perdurant).
2
This ontology was chosen mainly for the reasons outlined in the Intro- More particularly in terms of the domain of knowledge
duction but also because it integrates a class of mental objects which which directly involves OntoKADS, DOLCE's commit-
turn out to be very important for analyzing problem-solving knowledge.
3
Even though the goals are different a priori (building an expertise model
vs. verifying the logical coherence of subsumption links), we shall see
that the labeling meta-properties are of the same nature.
