Notesale: Turn your study into money

Document Preview

Extracts from the notes are below, to see the PDF you'll receive please use the links above

---

An ontological approach to the construction
of problem-solving models

ABSTRACT

Our ongoing work aims at defining an ontology-centered
approach for building expertise models for the CommonKADS methodology
...
In this article, our presentation of OntoKADS
will focus on the core ontology and, in particular, on roles the primitive situated at the interface between domain
knowledge and reasoning, and whose ontological status is
still much debated
...
We then show how this novel characterization of
the primitive allows definition of new rules for the construction of expertise models
...
۲٫٤ [Artificial Intelligence]: Knowledge Representation
Formalisms and Methods

Keywords

Knowledge Engineering and Modeling methodologies,
Problem-solving models, CommonKADS, Ontology Engineering, Foundational ontologies, DOLCE, Core problemsolving ontologies, OntoKADS, Ontologies of mental objects, I&DA, COM

INTRODUCTION

Since the late ۱۹۹۰s, the construction of explicit ontologies
has been considered as a promising way of improving the
knowledge engineering process: the elaboration of domain,
task and method ontologies early on in the design of problem-solving models was recommended [۲۲]
...
This component
(referred to as a Knowledge role in the CommonKADS
method [۲۱] and situated at the interface between domain

knowledge and reasoning) fulfils an important function: it
must allow problem-solving methods to be specified in
terms which are independent of particular application domains, thus facilitating re-use of these generic methods
...
In fact, methods which
recommend the use of ontologies all resort to a syntactic
ploy - transformation rules in PROTÉGÉ and bridges in
UPML - to link domain knowledge and reasoning [٦]
...
We show that
recent progress in the field of formal ontologies enables
one to account for this component in semantic terms,
within a coherent ontological framework
...
g
...
g
...
After having likened the latter to CommonKADS' Knowledge roles, we gave them the status of a metaproperty, i
...
along the same lines as Guarino's proposal
(making the role concept appear in an ontology of universals [۱۲])
...

Today, we are tackling this issue by using the OntoKADS
method [۳]
...

Hence, we now possess an ontological tool-box which is
both necessary and sufficient for making this type of proposal
...


۱

This article is an extended version of [۳]
...
The method comprises two main steps (see Figure 1)
...
To do this, the method
uses an ontology (also named OntoKADS) composed of
two main sub-ontologies:
•

•

A core problem-solving sub-ontology which enables
the engineer to define (by specialization) the application's concepts and specific reasonings
...

A meta-level sub-ontology coding the modeling primitives, which allows the engineer to label the concepts
from the previous ontology by using meta-properties
which represent modeling primitives
...


A software module (see Figure 1) then automatically translates this labeled ontology into three subcomponents of an
expertise model which resembles CommonKADS [21]: a
domain model, an inference model and a task model
...

In a second step, the knowledge engineer further specifies
the problem-solving methods linked to the tasks which
he/she has identified
...
In the remainder of this
article, our presentation of OntoKADS will concentrate on
describing the ontology's content
...


Task model

Labeled
problem-solving
application ontology

Step 2

Task
Method

Task

Inference Model
Domain Model

Figure 1
...


HOW OntoKADS EXTENDS DOLCE
DOLCE and the notion of knowledge
The OntoKADS ontology is defined as an extension of the
DOLCE ontology, which means that OntoKADS' concepts
and relations are defined by specialization of the abstract
concepts and relationships present in DOLCE
...
e
...
For the purposes of this article, we
consider only two of these here (see Figure 2):
•

The endurants (ED)
...
Within this
sub-domain, one can distinguish physical objects
(POB) and non-physical objects (NPOB), depending
on whether or not the objects have a direct physical location
...
These are entities which "occur
in time" but are only partially present at any time they
are present (events and states)
...
The latter (not being atomic) are considered to be accomplishments (ACC)
...
For example, the co-authors of this article (endurants) participated in the drafting of the text (a
perdurant)
...


3

Even though the goals are different a priori (building an expertise model
vs
...


More particularly in terms of the domain of knowledge
which directly involves OntoKADS, DOLCE's commit-

ment (corresponding to a consensus viewpoint within the
AI and KE communities) can be summarized as follows4:
•

Knowledge is the ability of an entity to perform an
action, i
...
to produce changes in a world
...


•

This knowledge or ability is embodied by an entity
(the agentive5), which confers the latter with the potential to repeat actions in which it participates (in terms
of the PC relationship) as an agent
...
e
...
e
...

The action (AC) is defined in DOLCE-Lite+ as an "accomplishment exemplifying the intentionality of an agent"6
...


Figure 2
...
In the rest of this article and in
order notably to illustrate the first sub-ontology, we will
consider two examples of problem-solving situations: diagnosing a car breakdown and calibrating a simulation
4

More precisely, DOLCE's commitment appears to us to be coherent with
this point of view
...


6

DOLCE-Lite+ contains various DOLCE extensions under study
...
15]
...
The first example corresponds to the teaching example dealt with in the CommonKADS reference book [21]
...

The first sub-ontology extends DOLCE in order to enable
description of problem-solving activities
...
Analysis of the models shows that
two distinct, general categories of actions suffice for tackling problem-solving activities, regardless of the latter's
complexity: Reasonings and Communications8
...
g
...
Reasonings contrast in this respect with actions, which seek to transform the real world
...
These are carried out by a
human being or a (knowledge-based) system and may require the latter to interact with another human being or
system - if only to exchange information with the outside
world
...
Within Interactions, we consider Communications, classified as Interactions whose goal is to exchange
information9 (A2)
...
g
...
g
...
On the basis of distinct goals (transforming the reasoner's mental world vs
...

(A1)

Reasoning(x) → AC(x)

(D1)
Interaction(x) =def AC(x) ∧∃yzt(isAgentOf(y,x) ∧
z≠y ∧ (APO(z) ∨ ASO(z)) ∧ PC(z,x,t))
(A2)

Communication(x) → Interaction(x)

(A3)

Reasoning(x) → ¬Communication(x)

7

A simulation code implements a model of a class of systems
...


8

In the rest of the article and in order to distinguish OntoKADS categories
from those of DOLCE, the names of the former will be noted in a
JAVA-like notation (e
...
CalibrationData, isAuthorOf), whereas abbreviations for the latter (e
...
ED, AC) will be used in the axioms
...
In particular, we do not take account of collaborative, problemsolving activities and the coordination mechanisms with which these activities are associated
...

Technically, we consider that this ontology is situated at a
"meta" level with respect to the previous one, and thus corresponds to an ontology of meta-properties
...
This relationship allows us to specify (for example) that the Diagnosis concept
is modeled as a Task at a certain time t1 (A4), that the CalibrationData concept is considered to be a FormalKnowledgeRole (A5) or indeed that EmptyFuelTank is a DomainConcept (A6)
...
The classifications are additionally
accompanied by constraints which express the fact that a
modeling primitive can only classify certain types of concepts in the problem-solving ontology
...
Figure 3 provides a graphic summary of
several constraints (the modeling primitives are noted in
bold characters)
...
In the next
section, we tackle analysis of the endurants which participate in these perdurants
...


10

This relation was introduced in [18] so as to account for the temporal
classification of an instance by a concept
...


11

The relation subsume(x,y) signifies that all instances of the concept y
are necessarily instances of the concept x
...
Structure of the OntoKADS ontology

The OntoKADS kernel
In this section, we show how OntoKADS can answer the
following questions: what is the ontological nature of the
entities participating in Reasonings and Communications?
How do these entities participate in actions? How can we
characterize (in terms of meta-properties) the concepts representing such participation modes?

Theoretical and practical knowledge and its
objects
Let us return to the notion of knowledge
...
e
...
g
...


•

theoretical knowledge (i
...
know-how “to think”)
deals with theoretical objects (mental objects) and enables action in the mental world (e
...
calculating, deciding)
...
The goal certainly
exists (as a mental object), whereas if the action fails, the
desired world state may not be reached and thus may not
exist
...
What is the situation for Reasonings and Communications?
By defining Reasoning as an action seeking to transform
the agent's mental world, we have forced its result - if one
exists - to be a mental object
...
Two other participants enable us to define Reasonings: the data and the
solving method
...

As for Communications, and by complementing the participation of mental objects, the exchange of information
between agents requires use of physical objects which
"convey" information – “documents”, in other words
...
To achieve this, OntoKADS
calls on the I&DA ontology, defined as an extension of
DOLCE and designed to enable description of documents
according to their content [8]
...
g
...

Finally, I&DA introduces a signifier-signified pairing
which specializes in the communication of information
between agents: the Discourse (likened to a statement) and
its content, the Message
...


Population of mental objects in I&DA
In order to account for documents and their content, I&DA
distinguishes between three types of entity:
•

The Document
...


•

The Expression
...
An example of an Expression is Text - a system
of signifying units (words and phrases) structured according to language-defined rules
...
A Content corresponds to the Expression's signified
...
e
...
Let us add (and this a point that turns
out to be important for OntoKADS) that a Proposition
"uses" or "has for its subject" Concepts (the hasForSubject relation) (A9)(D2)
...
This enables us to consider that any given
Content can be expressed (using different communication
codes) by different Expressions and, equally, that any
given Expression can be performed by different material
12

This type of hypothesis may appear to be surprising, a priori
...
The point here is that the eventualities depend on the methods implemented in given calculation situations
...
In contrast, we consider the data and the solving method to be essential participants
...


Figure 4
...
This identity appears
to be both pertinent and sufficiently general to account for
the meaning of the following expressions (which appear as
task data in CommonKADS expertise models): EmptyFuelTank, EmptyFuelTankHypothesis, LowBatteryLevelComplaint and CarModel
...
Hence, the expression "empty fuel
tank" can be likened to the Concept of a class of states or to
an Assertion (a Proposition considered to be true for an
agent) whose Subject is the said Concept
...

It is also noteworthy that the Concept (either alone or playing the role of a Subject) can vary and may correspond to a
Concept of state, process or behavior
...


mode (e
...
visible, invisible, observed, observable, etc
...
g
...

The above remark clarifies the generality of the modeling
framework
...
The expressions "empty fuel tank
hypothesis" and "low battery level complaint" can thus be
likened to Propositions resulting respectively from hypothetical reasoning and a discursive act consisting in "complaining about something"
...
This category
covers knowledge models exploited by Reasonings as well
as mathematical models used to simulate system behavior
...
This type of expression - useful for naming knowledge roles (in the CommonKADS
sense) in task inputs and outputs [4] refers, in fact, to ways
in which Contents participate in Reasonings, for example
as data or results
...

These roles (also referred to as "casual roles" or "thematic
roles" in the literature) are defined in OntoKADS as particularizing the endurant concept
...
Hence, the participation roles or specialized participants are defined by introducing relationships which
particularize the PC participation equation
...
It is noteworthy
that we have forced the Data i) to be a Content participating in an Action (A12) and ii) to participate from the start
of the perdurant onwards (A13) (in contrast, the Result
participates at the end of the perdurant)
...

(A10)

isAffectedBy(x,y) → ∃t(PC(x,y,t))

(D3)

Patient(x) =def ∃y(isAffectedBy(x,y))

(T1)

Patient(x) → ED(x)

15

According to DOLCE axioms: Ad33 (PC(x,y,t) → ED(x) ∧ PD(y) ∧
T(t)) and Ad35 (ED(x) → ∃y,t(PC(x,y,t))), only the endurants participate in the perdurants and, incidentally, all endurants participate necessarily in a perdurant
...
In the CommonKADS method, this distinction reflects the difference between two modeling primitives, the
"domain concept" primitive and the "knowledge role"
primitive
...
We end by defining
novel modeling primitives for the OntoKADS method
...
We
also adopt their definition of three meta-properties involving a notion of "role": role, formal role and material role
...
Its anti-rigidity, (i
...
the property of being non-essential for all its instances) translates into
dynamic behavior over time: an instance only plays a
role by accident
...


•

A formal role is a role which does not carry an identity
criterion
...

The Agent and Patient concepts (which we qualified as
"participation roles") are examples of formal roles
...


•

A material role is a role carrying an identity criterion
...
Examples of material roles
are the Student and Employee concepts defined as Per-

16

For reasons of space, we are not able to give the notions’ formal definitions here
...


sons (a type) playing a formal role vis-à-vis a healthcare establishment or an employer
...
e
...
Figure 5
provides a graphical illustration of the concepts labeling
using these meta-properties
...


•

a FormalKnowledgeRole is a KnowledgeRole which
does not carry an identity criterion
...
The roles
can be played by Concepts or Propositions which are
Contents carrying incompatible identity criteria
...
The ModelToCalibrate and
DiagnosisHypothesis concepts are examples of this
...


Finally, the resulting modeling framework for entities participating in Reasonings can be summarized as follows:
•

The KnowledgeRole, FormalKnowledgeRole and MaterialKnowledgeRole modeling primitives are (like the
other primitives) meta-properties, i
...
properties which
classify other properties temporally
...
The Inputs and the Outputs (primitives
particularizing the KnowledgeRole primitive) classify
Data and Results, respectively
...


•

These latter Concepts classify the domain objects, their
components and the states and processes in which
these objects intervene
...
Labeling of participants using KnowledgeRoles

DISCUSSION
The OntoKADS core problem-solving ontology prompts a
knowledge modeling process which differs significantly
from that proposed by CommonKADS
...
The
difference is most notable for Reasonings and their participants
...
The categories that are usually selected for
knowledge roles in CommonKADS models correspond to
categories of Concepts and Propositions in OntoKADS
...

OntoKADS replaces the CommonKADS knowledge roles
(which are poorly defined in semantic terms) by MaterialKnowledgeRoles, whose definition is based on a rigorous
framework (in particular, these concepts must verify certain
meta-properties)
...
g
...
g
...
Our current work involves comparing these models by using examples of generic tasks and methods from
the literature
...
Other ontological works have also tackled this
field by pursuing a range of objectives:
•

modeling document content for the SUMO ontology's
"practical semiotics" [20] or, more specifically, the
origin of information contained in web pages for Fox
and Huang's ontology of propositions [9]
...
g
...
) for the D&S ontology (Descriptions and Situations) [11]
...
g
...
) and their links to mental objects for the
COM ontology (Computational Ontology of Mind) [7]
...
For example, entities referred to as "propositions"
are defined by both ontologies as "abstract entities corresponding to document content"
...
The two other ontologies (D&S
and COM) are, in contrast, defined as extensions of
DOLCE, which facilitates comparisons
...
Its domain is composed of two disjoint subdomains: i) mental states, which one can consider as being
parts (as understood in the P "is part of" relationship in
DOLCE) of Reasonings
...
It appears clear that the
Concepts and Propositions in OntoKADS can be defined
as computed objects
...
On this basis, a merger between COM and OntoKADS seems to be possible
...
15]) appear to cover the same domain as I&DA if
one performs the following alignment: the reified theories
or s-descriptions in D&S correspond to Propositions in
I&DA, the reified concepts or c-descriptions correspond to
Concepts and the information objects correspond to Expressions
...
However, differences do exist, as shown by
the recent application of D&S to construction of a core
biomedical ontology [10]
...
For example, according to D&S, the fever and critical systolic

blood pressure parameters "select" a body temperature
sub-region and a blood pressure sub-region, respectively
...
These concepts can then play the role of Signs by enabling the evocation of other concepts [14]
...
The D&S c-description concept is thus quite
different from the Concept in OntoKADS
...
Hence, at present, the domain is largely an open field
for research
...


CONCLUSION
In the present article, we have set the foundations of a radically ontology-centered approach to the construction of
expertise models
...
We defend the following thesis: recent progress in the
field of formal ontologies (and notably work in the area of
reasoning objects) means that such an approach is now
within our reach
...
As we saw in the previous section, the ontology itself
is being expanded
...
This environment is defined as an extension of the TERMINAE ontology construction platform [1] and uses the OntoSpec
semi-informal ontology specification language [15]
...
Aussenac-Gilles, B
...
Szulman
...
In R
...
Corby (eds
...
172-188
...
Bachimont
...
Authorization to Supervise Research,

Université de Technologie de Compiègne, January
2004
...
Bruaux and G
...
OntoKADS: a core ontology
for developing expertise models for the CommonKADS methodology
...
64-75
...
org/Vol-118/>
...
Bruaux, G
...
Morel
...
Proceedings of the Conférence en Ingénierie des Connaissances IC'2003, Laval (France),
July 2003, p
...

[5] C
...
Modelling social action for AI
agents
...

[6] M
...
Musen
...
In S
...
Studer (eds
...

321-341
...
Ferrario and A
...
Towards a Computational Ontology of Mind
...
C
...
Vieu
(eds
...
287-297
...
-Y
...
Kassel
...
In
Proceedings of the ECAI'2004 Workshop on Knowledge Management and Organizational Memories, August 2004, Valencia (Spain), p
...

[9] M
...
Fox and J
...
Knowledge provenance: An
approach to modeling and maintaining the evolution
and validity of knowledge
...

of
Toronto,
May
2003
...
ei
...
ca/km/papers/fox-kpl
...

[10] A
...
Catenacci and M
...
Inflammation Ontology Design Pattern: an exercise in building a
core biomedical ontology with Descriptions and Situations
...
M
...
), Ontologies in Medicine,
IOS Press, Amsterdam, Berlin, 2004, p
...

[11] A
...
Mika
...
In R
...
(eds), Proceedings of the International Conference on Ontologies, Databases and Applications of Semantics (ODBASE 2003), Catania (Italy), November 2003
...
Guarino and C
...
A formal Ontology of Properties
...
Dieng and O
...
), Proceedings of
the 12th International Conference on Knowledge Engineering and Knowledge Management: EKAW2000,

View publication stats

Lecture Notes on Computer Science, Springer Verlag,
2000, p
...

[13] N
...
Welty
...

In S
...
Studer (eds
...
151-171
...
Kassel
...
In Proceedings of the IJCAI’99 Workshop on Ontologies and
Problem-Solving Methods: Lessons Learned and Future Trends, Stockholm (Sweden), August 1999, p
...
Available at ...

[15] G
...
OntoSpec : une méthode de spécification
semi-informelle d'ontologies
...
75-87
...
Loebe
...
Master’s Thesis of the University of
Leipzig (Germany), August 2003
...
Masolo, S
...
Gangemi, N
...

Oltramari and L
...
The WonderWeb Library
of Foundational Ontologies and the DOLCE ontology
...
1
...

[18] C
...
Vieu, E
...
Catenacci, R
...
Gangemi and N
...
Social Roles
and their Descriptions
...
Dubois, C
...
A
...
), Proceedings of the Ninth International Conference on the Principles of Knowledge
Representation and Reasoning, Whistler, BC, Canada,
June 2004, p
...

[19] A
...
Unified Theories of Cognition
...

[20] A
...
Niles
...
In Proceedings of the International Conference
on Information and Knowledge Engineering (IKE’02),
Las Vegas, Nevada, June 2002
...
Schreiber, H
...
Anjewierden, R
...
Shadbolt, W
...
J
...
MIT Press, Cambridge,
Massachusetts, London, England, 2000
...
Van Heijst, A
...
Schreiber and B
...
Wielinga
...
International Journal of Human-Computer Studies, 46:183292, 1997
...
Van Heijst, A
...
Schreiber and B
...
Wieling
...
International Journal of Human-Computer Studies, 46:311318, 1997

---

Title: An ontological approach to the construction of problem-solving models
Description: An ontological approach to the construction of problem-solving models

Buy These NotesPreview