=Paper=
{{Paper
|id=Vol-18/paper-1
|storemode=property
|title=Overview of Knowledge Sharing and Reuse Components: Ontologies and Problem-Solving Methods
|pdfUrl=https://ceur-ws.org/Vol-18/1-gomez.pdf
|volume=Vol-18
}}
==Overview of Knowledge Sharing and Reuse Components: Ontologies and Problem-Solving Methods==
https://ceur-ws.org/Vol-18/1-gomez.pdf
Overview of Knowledge Sharing and Reuse Components:
Ontologies and Problem-Solving Methods
Asunción Gómez Pérez V. Richard Benjamins
DIA SWI
Technical University of Madrid University of Amsterdam
Spain The Netherlands
asun@delicias.dia.fi.upm.es richard@swi.psy.uva.nl
systems. This approach would facilitate building bigger
and better systems cheaply...”
Abstract Since then considerable progress has been made in de-
veloping the conceptual bases needed for building technol-
Ontologies and problem-solving methods are ogy that allows knowledge-component reuse and sharing.
promising candidates for reuse in Knowledge En- However, we are still far from the ultimate objective. To
gineering. Ontologies define domain knowledge enable sharing and reuse of knowledge and reasoning be-
at a generic level, while problem-solving methods havior across domains and tasks, Ontologies and Problem-
specify generic reasoning knowledge. Both type Solving Methods (PSMs) have been developed. Ontologies
of components can be viewed as complementary are concerned with static domain knowledge and PSMs
entities that can be used to configure new knowl- with dynamic reasoning knowledge. The integration of on-
edge systems from existing, reusable components. tologies and PSMs is a possible solution to the “interaction
In this paper, we give an overview of approaches problem” [BC88], which hampered reuse in the eighties.
for ontologies and problem-solving methods. The interaction problem states that representing knowledge
for the purpose of solving some problem is strongly af-
1 Introduction fected by the nature of the problem and the inference strat-
egy to be applied to the problem. Through ontologies and
In 1991, the ARPA Knowledge Sharing Effort [NFF 91] PSMs this interaction can be made explicit in the notion
envisioned a new way in which intelligent systems could be of assumptions and taken into consideration. PSMs and
built. They proposed the following: “Building knowledge- ontologies can be seen as complementary reusable compo-
based systems today usually entails constructing new nents to construct knowledge systems from reusable com-
knowledge bases from scratch. It could be done by as- ponents. In order to build full applications of information
sembling reusable components. Systems developers would and knowledge systems from reusable components, both
then only need to worry about creating the specialized PSMs and ontologies are required in a tightly integrated
knowledge and reasoners new to the specific task of their way.
system. This new system would interoperate with exist-
Ontologies aim at capturing domain knowledge in a
ing systems, using them to perform some of its reasoning.
generic way and provide a commonly agreed understanding
In this way, declarative knowledge, problem-solving tech-
of a domain, which may be reused and shared across appli-
niques and reasoning services would all be shared among
cations and groups [CJB99]. Ontologies provide a com-
mon vocabulary of an area and define -with different levels
The copyright of this paper belongs to the papers authors. Permission to
copy without fee all or part of this material is granted provided that the of formality- the meaning of the terms and the relations
copies are not made or distributed for direct commercial advantage. between them. Ontologies are usually organized in tax-
Proceedings of the IJCAI-99 workshop on onomies and typically contain modeling primitives such as
Ontologies and Problem-Solving Methods (KRR5) classes, relations, functions, axioms and instances [Gru93].
Stockholm, Sweden, August 2, 1999 Popular applications of ontologies include knowledge man-
(V.R. Benjamins, B. Chandrasekaran, A. Gomez-Perez, N. Guarino, M. agement, natural language generation, enterprise modeling,
Uschold, eds.) knowledge-based systems, ontology-based brokers, and in-
http://sunsite.informatik.rwth-aachen.de/Publications/CEUR-WS/Vol-18/ teroperability between systems.
A. Gómez Pérez, V.R. Benjamins 1-1
Problem-solving methods (PSMs) describe the reason- saying that: “Ontologies are defined as a formal specifica-
ing process of a knowledge-based system (KBS) in an tion of a shared conceptualization”. These two definitions
implementation- and domain-independent manner. A PSM have been explained by Studer and Colleagues [SBF98] as
defines a way of how to achieve the goal of a task. It has in- follows: “ Conceptualization refers to an abstract model of
puts and outputs and may decompose a task into subtasks. some phenomenon in the world by having identified the rel-
In addition, a PSM specifies the data flow between its sub- evant concepts of that phenomenon. Explicit means that the
tasks. Control knowledge determines the execution order type of concepts used, and the constraints on their use are
and iterations of the subtasks of a PSM. explicitly defined. Formal refers to the fact that the ontol-
In the following sections, we will discuss several aspects ogy should be machine-readable. Shared reflects the notion
of both ontologies and problem-solving methods. At the that an ontology captures consensual knowledge, that is, it
end of the paper, we will mention some directions for future is not private to some individual, but accepted by a group.”
work in this area. Based on the definition of Gruber, many definitions of
what an ontology is have been proposed in the literature.
2 Ontologies In 1995, Guarino and Giaretta [GG95] collected seven def-
initions and provided corresponding syntactic and semantic
The aims of this section are to provide answers to the fol- interpretations. Other definitions are: “an ontology is a hi-
lowing questions: What is an ontology? What principles erarchically structured set of terms for describing a domain
should I follow to build an ontology? What are the com- that can be used as a skeletal foundation for a knowledge
ponents of an ontology? What types of ontologies exist? base” [SPKR97], and “An ontology provides the means
How are ontologies organized in libraries? What methods for describing explicitly the conceptualization behind the
should I use to build my own ontology? Which techniques knowledge represented in a knowledge base.” [BLC96].
are appropriate for each step? How do software tools sup- As a main conclusion to this section, we can say that the
port the process of building and using ontologies? What literature provides several definitions of the word ontology.
are the most well-known ontologies? What are the uses Different definitions provide different and complementary
of ontologies? Which principles should I use to select the points of view of the same reality.
best ontology for my application? To answer the above
questions, the section is organized as follows. First, the What principles should I follow to build ontologies?
theoretical foundations of the ontological engineering field
Here we summarize some design criteria and a set of prin-
will be presented. This will be followed by a presentation
ciples that have been proved useful in the development of
of some existing ontologies. The second part will address
ontologies.
methodologies for building ontologies. The third part will
present tools for building ontologies. Finally, the last will Clarity and Objectivity [Gru95], which means that the
be related to uses of ontologies in applications. ontology should provide the meaning of defined terms
by providing objective definitions and also natural lan-
2.1 Theoretical Foundations guage documentation.
What is an ontology? Completeness [Gru95], which means that a definition
expressed in terms of necessary and sufficient condi-
The word ontology has been taken from Philosophy, where tions is preferred over a partial definition (defined only
it means a systematic explanation of Existence. In the through necessary or sufficient condition).
Artificial Intelligent field, first Neches and colleagues
[NFF 91] defined an ontology as follows “An ontology Coherence [Gru95], to permit inferences that are con-
defines the basic terms and relations comprising the vo- sistent with the definitions.
cabulary of a topic area as well as the rules for combin-
Maximum monotonic extendibility [Gru95]. It means
ing terms and relations to define extensions to the vocabu-
that new general or specialized terms should be in-
lary”. We can say that this definition tells us how to proceed
cluded in the ontology in a such way that is does not
to build an ontology, giving us vague guidelines: identify
require the revision of existing definitions.
basic terms and relations between terms, identify rules to
combine them, provide definitions of such terms and rela- Minimal ontological commitments [Gru95], which
tions. Note that according to this definition, an ontology means to make as few claims as possible about the
includes not only the terms that are explicitly defined in world being modeled, giving the parties committed to
it, but also terms that can be inferred using rules. Later, in the ontology freedom to specialize and instantiate the
1993, Gruber’s definition [Gru93] becomes famous “an on- ontology as required.
tology is an explicit specification of a conceptualization”,
“Ontological commitments refer to agreement to use the shared vo-
being this definition the most referenced in the literature. cabulary in a coherent and consistent manner. They guarantee consistency,
In 1997, Borst [Bor97] slightly modify Gruber’s definition but not completeness of an ontology” [GO94].
A. Gómez Pérez, V.R. Benjamins 1-2
Ontological Distinction Principle [BGM96], which What types of ontologies already exist?
means that classes in an ontology should be disjoint.
Nowadays, it is easy to get information from organizations
that have ontologies on the WWW. Many ontologies like
Diversification of hierarchies to increase the
Ontolingua ontologies at the Ontology Server [FFR97]
power provided by multiple inheritance mechanisms
and WordNet [Mil90] at Princenton are freely available
[AGLP98].
over the Internet. Other ontologies, like Cyc ontologies
Modularity [BLC96] to minimize coupling between [LG90], are partially freely available on the web. However,
modules. the majority of ontologies have been developed by compa-
nies for their own use and are not available. The Ontology
Minimization of the semantic distance between sib- Page (also known as TOP) and (Onto)2Agent [AGLP98]
ling concepts [AGLP98] which means that similar (an ontology-based www broker that helps to select ontolo-
concepts are grouped and represented using the same gies) might help to select ontologies.
primitives. This section does not seek to give an exhaustive typol-
ogy of ontologies as presented in [vSW97, MVI95]. How-
Standardization of names whenever is possible ever, it presents the most commonly used types of ontolo-
[AGLP98]. gies.
Knowledge Representation ontologies [vSW97] cap-
What are the components of ontologies?
ture the representation primitives used to formalize
Knowledge in ontologies is formalized using five kinds of knowledge in knowledge representation paradigms.
components: classes, relations, functions, axioms and in- The most representative example is the Frame-
stances [Gru93]. Classes in the ontology are usually orga- Ontology [Gru93], which captures the representation
nized in taxonomies. Sometimes, the notion of ontology is primitives used in frame-based languages. It allows
diluted, in the sense that taxonomies are considered to be other ontologies to be specified using frame-based
full ontologies [SBF98]. conventions. It is implemented in KIF 3.0 [GF92].
General/Common ontologies [Gua98] include vocab-
Concepts are used in a broad sense. A concept can
ulary related to things, events, time, space, causality,
be anything about which something is said and, there-
behavior, function, etc.
fore, could also be the description of a task, function,
action, strategy, reasoning process, etc. The CYC ontology [LG90] is a common sense on-
tology that provides a vast amount of fundamental
Relations represent a type of interaction between con- human knowledge. The Cyc ontology is divided
cepts of the domain. They are formally defined as any into many micro-theories. Cyc Ontologies are imple-
subset of a product of n sets, that is: R: C1 x C2 x ... x mented in CycL language.
Cn. Examples of binary relations include: subclass-of
and connected-to. Top-Level Ontologies provide general notions un-
der which with all the terms in existing ontolo-
Functions are a special case of relations in which the gies are related. Examples of top level ontolo-
n-th element of the relationship is unique for the n-1 gies are: Sowa’s boolean lattice [Sow97], PAN-
preceding elements. Formally, functions are defined GLOSS [KL94], Penman Upper Level [BKMW90],
as: F: C1 x C2 x ... x Cn-1 Cn. Examples of Cyc [LG90], Mikrokosmos [Mah96] and Guarino’s
functions are Mother-of and Price-of-a-used-car that top level proposal [Gua98].
calculates the price of a second-hand car depending Meta-ontologies, also called Generic Ontologies or
on the car-model, manufacturing date and number of Core Ontologies [vSW97] are reusable across do-
kilometers. mains. The Mereology ontology [Bor97] could be the
most typical example. It defines the part-of relation
Axioms are used to model sentences that are always
and its properties. This relation allows to express that
true.
devices are assembled of components, each of which
Instances are used to represent elements. might -on its turn- be decomposed in subcomponents.
http://www-ksl.stanford.edu:5915 or the European mirror site at
Once the main components of ontologies have been rep- http://www-ksl-svc-lia.dia.fi.upm.es:5915
http://www.tio.darpa.mil/Summaries95/B370-Princenton.html
resented, the ontology can be implemented in a various lan- http://www.cyc.com/
guages: highly informal, semi-informal, semi-formal and http://www.medg.lcs.mit.edu/doyle/top
rigorously formal languages [Usc96]. http://delicias.dia.fi.upm.es/REFERENCE ONTOLOGY/
A. Gómez Pérez, V.R. Benjamins 1-3
Domain ontologies [MVI95, vSW97] are reusable in Domain-Task ontologies are task ontologies reusable
a given domain. They provide vocabularies about in a given domain, but not across domains.
the concepts within a domain and their relationships,
about the activities that take place in that domain, and Method ontologies provide definitions of the relevant
about the theories and elementary principles govern- concepts and relations used to specify a reasoning pro-
ing that domain. cess to achieve a particular task [CJB99].
In the domain of engineering ontologies, the Eng-
Math ontology [GO94] and PhysSys [Bor97] deserve Application ontologies [vSW97] contain the neces-
special mention. EngMath is an Ontolingua ontol- sary knowledge for modeling a particular application.
ogy developed for mathematical modeling in engi-
neering. PhysSys is an engineering ontology for mod- Meta-ontologies, domain ontologies and applications
eling, simulating and designing physical systems. ontologies capture static knowledge in a problem-solving
independent way, where as PSMs ontologies, task ontolo-
In the domain of enterprise modeling process, the En-
gies and domain-task ontologies are concerned with prob-
terprise Ontology [Usc96] is a collection of terms
lem solving knowledge. All these kind of ontologies can
and definitions relevant to business enterprises. On-
be combined to build a new ontology. The reusability-
tologies built at the TOVE [GF95] (Toronto Virtual
usability trade-off problem [KBD 91] applied to the on-
Enterprise) project are: Enterprise Design Ontology,
tology field states that the more reusable an ontology is,
Project Ontology, Scheduling Ontology, or Service
the less usable it is, and vice versa. The first thing to
Ontology.
do to model a new ontology using existing ontologies
An illustrative example of ontologies for Knowledge from the library is to decide which knowledge represen-
Management is the (KA) ontology [BFDGP97], to tation paradigm to use to formalize knowledge, which
be used by the Knowledge Annotation Initiative of the will then be committed to into a knowledge representa-
Knowledge Acquisition Community. This ontology tion ontology. Having selected the knowledge represen-
is being built jointly and distributively with people at tation ontology, the next step is to decide whether gen-
different locations. eral/common ontologies are needed in the new ontology.
If they are required, new ontologies are built and entered
The most illustrative linguistic ontologies are the Gen-
into the library or reused from the library. This is when
eralized Upper Model [BMF95], WordNet [Mil90]
knowledge-component modeling starts. Simultaneously,
and Sensus [SPKR97]. The Generalized Upper Model
domain knowledge and problem-solving knowledge can be
(GUM ) is a general task and domain-independent
modeled. So, when domain knowledge is modeled, first
linguistic ontology. To make it portable across dif-
generic ontologies, then domain ontologies, and finally
ferent languages (English, German, Spanish, Italian,
application domain ontologies are built. When problem-
etc.), the GUM ontology only includes the main lin-
solving knowledge is modeled, first Task and PSMs on-
guistic concepts and how they are organized across
tologies, then domain task ontologies and finally appli-
languages, and omits details that differentiate lan-
cation domain task ontologies are modeled. Method and
guages. WordNet is a lexical database for English
task ontologies allow the assumption-based interaction be-
based on psycholinguistic principles. Its information
tween problem-solving and domain ontologies to be explic-
is organized in units called “synsets”, which are sets
itly stated [BFS96, BPG96, FS98].
of synonyms that are interchangeable in a particu-
lar context and are used to represent different mean-
ings. SENSUS is a natural language based ontology 2.2 Methodologies for building ontologies
whose goal is to provide a broad conceptual struc-
The ontology building process is a craft rather than an en-
ture for work in machine translation. It was developed
gineering activity. Each development team usually follows
by merging and extracting information from existing
its own set of principles, design criteria and phases in the
electronic resources.
ontology development process. The absence of commonly
Task ontologies [MVI95] provide a systematic vocab- agreed on guidelines and methods hinders the development
ulary of the terms used to solve problems associated of shared and consensual ontologies within and between
with tasks that may or may not be from the same teams, the extension of a given ontology by others and
domain. They include generic names, generic verbs its reuse in other ontologies and final applications. If on-
generic adjectives and others in the scheduling tasks. tologies are built on a small scale, some activities can be
skipped. But, if you intend to build large-scale ontologies
http://www.aiai.ed.ac.uk/project/enterprise
http://www.ie.utoronto.ca/EIL
with some guarantees of correctness and completeness, it
http://www.aifb.uni-karlsruhe.de/WBS/broker/KA2.html is advisable to steer clear of anarchic constructions and to
http://www.darmstadt.gmd.de/publish/komet/gen-um/newUM.html follow a methodological approach.
A. Gómez Pérez, V.R. Benjamins 1-4
Uschold’s methodology [UG96, Usc96] is based on the methodology proposes coding in a formal language and
experience of building the Enterprise Ontology, which in- METHONTOLOGY proposes expressing the idea as a set
cludes a set of ontologies for enterprise modeling, and pro- of intermediate representations (IR). Then the ontology is
poses the following steps: (1) identify the purpose and generated using translators. These IRs bridge the gap be-
scope of the ontology; (2) build the ontology by capturing tween, on the one hand, how people see a domain and,
knowledge, coding knowledge and integrating the knowl- on the other hand, the languages in which ontologies are
edge with existing ontologies; (3) evaluate the ontology; formalized. These intermediate representations provide a
(4) documentation; and (5) guidelines for each phase. user-friendly approach for both knowledge acquisition and
Grüninger and Fox’s methodology [GF95] is based on evaluation by computer scientists and domain experts who
the experience of building an enterprise modeling ontology are not knowledge engineers [ABB 98].
in the framework of the TOVE project. Essentially, it in- The need for ontology evaluation is also identified in
volves building a logical model of the knowledge that is the three above methodologies. Uschold’s methodology
to be specified in the ontology. This model is not built di- includes this activity but it does not state how it should
rectly. First, the specifications that are to be met by the on- be carried out. Grüninger and Fox propose identifying a
tology are described informally by identifying a set of com- set of competency questions. Once the ontology has been
petency questions, and this description is then formalized expressed formally, it is compared against this set of com-
in a language based on first-order predicate calculus. The petency questions. Finally, METHONTOLOGY proposes
competency questions are the basis for a rigorous charac- that evaluation activities be carried out throughout the en-
terization of the knowledge that the ontology has to cover, tire lifetime of the ontology development process. Most of
and they specify the problem and what constitutes a good the evaluation is done in the conceptualization phase.
solution to the problem. By a composition and decompo- As main conclusion at this point we can say that each
sition mechanism, competency questions and their answers group has and uses its own methodology and there does not
can be used to answer more complex competency questions yet exist a common methodology that everybody agrees on.
in other ontologies, allowing the integration of ontologies. Therefore, additional research has to be performed in this
The METHONTOLOGY framework [GP98, FGPPP99] direction.
enables the construction of ontologies at the knowledge
level. It includes: (a) the identification of the ontology de- 2.3 Languages and environments for building ontolo-
velopment process, which refers to which tasks (planning, gies
control, specification, knowledge acquisition, conceptual- Which are the most commonly used languages to build on-
ization, integration, implementation, evaluation, documen- tologies?
tation, configuration management, etc.) one should carry
out when building ontologies; (b) a life cycle based on Basically, several representation systems have been re-
evolving prototypes, which identifies the stages through ported for formalizing ontologies under a frame-based
which the ontology passes during its lifetime; and (c) the modeling approach, a logic-based approach or even both.
methodology itself, which specifies the steps to be taken The most representative languages are Ontolingua [Gru93],
to perform each activity, the techniques used, the prod- CycL [LG90], Loom [Mac91] and FLogic [KLW95].
ucts to be output and how they are to be evaluated. The Ontolingua is a language based on KIF and on the Frame
main phase is the conceptualization phase. During both Ontology, and is the ontology-building language used by
specification and conceptualization, a process of integra- the Ontology Server. The Ontolingua language allows on-
tion was completed using in-house and external ontolo- tologies to be built in any of the following three manners:
gies. This framework is partially supported by a software (1) using KIF expressions; (2) using exclusively the Frame
environment called Ontology Design Environment (ODE) Ontology vocabulary; (3) using both languages at the same
[BFGPGP98], [FGPPP99]. Several ontologies have been time, depending on ontology developer preferences. In any
developed using METHONTOLOGY and ODE: CHEM- case, the Ontolingua definition is composed of a heading,
ICALS [FGPPP99], Environmental pollutants ontologies an informal definition in natural language, and a formal
[GPR99], the Reference-Ontology [AGLP98] and the re- definition written in KIF or using the frame ontology vo-
structured version of the (KA) ontology [BFGPGP98]. cabulary. A GFP [CFF 97] application is required in order
This methodology has been proposed to build ontologies by to reason with Ontolingua Ontologies.
the Foundation for Intelligent Physical Agents (FIPA ). CycL is Cyc’s knowledge representation language.
All these methodologies have in common that they start CycL is a declarative and expressive language, similar to
from the identification of the purpose of the ontology and first-order predicate calculus with extensions. CycL uses
the need for domain knowledge acquisition. However, hav- a form of circumscription, includes the unique names as-
ing acquired a significant amount of knowledge, Uschold’s sumption, and can make use of the closed world assump-
tion where appropriate. CycL has an inference engine to
http://www.fipa.org perform several kinds of reasonings.
A. Gómez Pérez, V.R. Benjamins 1-5
LOOM is a high-level programming language based tional database). So, non-experts in the languages in which
on first-order logic which belongs to the KL-ONE fam- ontologies are implemented could specify and validate on-
ily. The LOOM language provides: an expressive and tologies using this environment.
explicit declarative model specification language, a pow- Tadzebao and WebOnto are complementary tools that
erful deductive support, several programming paradigms, are being developed by the Knowledge Media Institute at
and knowledge-base services. The Open University. Tadzebao enables knowledge en-
FLogic is an integration of frame-based languages and gineers to hold synchronous and asynchronous discussion
first-order predicate calculus. It includes objects (simple about ontologies and WebOnto supports the collaborative
and complex), inheritance, polymorphic types, query meth- browsing, creation and editing of ontologies.
ods and encapsulation. Its deductive system works with the
theory of predicate calculus and structural and behavioral 2.4 Applications that use ontologies
inheritance.
Although ontologies can be used to communicate be-
tween systems, people, and organizations, interoperate be-
How do software tools support the process of building and
tween systems, and support the design and development of
using ontologies?
knowledge-based and general software systems [Usc96],
The main tools for building ontologies are: The On- the number of applications built that use ontologies to
tology Server [FFR97], Ontosaurus [SPKR97], ODE model the application knowledge is small. That is, many
[BFGPGP98, FGPPP99] and Tadzebao and Webonto times such ontologies have been built just for a given appli-
[Dom98]. cation without special consideration for sharing and reuse.
The Ontology Server is the best known environment for Several problems make difficult the reuse of existing on-
building ontologies in the Ontolingua language. It is a set tologies in applications [AGLP98]: Ontologies are dis-
of tools and services that support the building of shared on- persed over several servers; the formalization differs de-
tologies between geographically distributed groups. It was pending on the server on which the ontology is stored;
developed in the context of the ARPA Knowledge Shar- ontologies on the same server are usually described with
ing Effort by the Knowledge Systems Laboratory at Stan- different levels of detail; and there is no common format
ford University. The ontology server architecture provides for presenting relevant information about the ontologies so
access to a library of ontologies, translators to languages users can decide which ontology best suits their purpose.
(Prolog, CORBA’s IDL, CLIPS, Loom, KI) and an editor These problems are probably the cause for the relatively
to create and browse ontologies. There are three modes small number of known applications until now. Several
of interaction: remote collaborators that are able to write applications that use ontologies can be found in the pro-
and inspect ontologies; remote applications that may query ceedings of the workshop on Applications of ontologies
and modify ontologies stored at the server over the Internet and PSMs held in conjunction with ECAI98.
using the generic frame protocol; and stand-alone applica- There exist several applications that use natural lan-
tions. guage ontologies. The GUM is being used in natural lan-
Ontosaurus is being developed by the Information Sci- guage generation applications in different languages: Pen-
ences Institute at the University of South California. It con- man [BKMW90], KOMET [Bat94], TechDoc [Ros94], Al-
sists of two parts: an ontology server that uses Loom as Fresco [SCC 93], OntoGeneration [ABB 98], and the
knowledge representation system and an ontology browser language of [FvdR98]. WordNet is used by Hermes
server that dynamically crates html pages (including im- [Hoe98] and OntoSeek [GMV99].
age and textual documentation) that displays the ontology In the domain of enterprise modeling, the Enterprise tool
hierarchy and it uses html forms to allow the user to edit set (see:
the ontology. Translators from loom to Ontolingua, KIF, http://www.aiai.ed.ac.uk/project/enterprise for more infor-
KRSS and C++ have also been developed. mation) is the most relevant environment built using the
ODE (Ontology Design Environment) is being devel- Enterprise ontology. The Enterprise Design Workbench
oped by the Computer Science School at Universidad and the Integrated Supply Chain Management Project use
Politécnica de Madrid. The main advantage of ODE is the TOVE Ontologies.
conceptualization module for building ontologies, which Recently, ontologies are being used by www brokers
allows the ontologist to develop the ontology at the knowl- in different domains. Ontobroker [FDES98] for knowl-
edge level using a set of intermediate representations (IRs) edge management in the context of the Knowledge Anno-
that are independent of the target language in which the tation Initiative of the Knowledge Acquisition Community,
ontology will be implemented. Once the conceptualiza- (Onto)2Agent [AGLP98] for selecting ontologies that sat-
tion is complete, the code is generated automatically us- isfy a given set of constraints and Chemical OntoAgent
ing ODE code generators (Ontolingua, FLogic and a rela- [AGLP98] for teaching chemistry.
http://indra.isi.edu:8000 http://www.aifb.uni-karlsruhe.de/WBS/broker/
A. Gómez Pérez, V.R. Benjamins 1-6
In the domain of information systems design, Comet
Comptence
[WMK95] supports the design of software systems, and
Cosmos [WMK95] supports engineering negotiation. Both is-realized-by
systems give design feedback to their users.
KACTUS [SWJ95] was an ESPRIT project on modeling Operational specification
knowledge of complex technical systems for multiple use (inf1;inf2)*
and the role of ontologies to support it.
Plinius [vdVSM94] is a semi-automatic knowledge ac- role1 inf1 role2 inf2 role3
quisition system from natural language text in the domain
of ceramic materials, their properties and their production uses
processes. Requirements/
Assumptions
3 Problem-Solving Methods
Figure 1: The architecture of a PSM.
Problem-Solving Methods (PSMs) are nowadays recog-
Before discussing different approaches to PSMs, we
nized as valuable components for constructing knowledge-
will briefly present a general architecture of PSMs (taken
based systems (KBSs). This is manifested by the
from[BFS96]).
fact that the notion of PSM is present in lead-
ing knowledge engineering frameworks such as Task
Structures [CJS92], Role-Limiting Methods [Mar88a], 3.1 Architecture of PSMs
CommonKADS [SWdH 94], Protégé [Mus93], MIKE Most approaches agree that a PSM consists of three related
[AFS98], Components of Expertise [Ste90], EXPECT parts, describing what a PSM can achieve, how it achieves
[SG95], GDM [TvHWS93] and VITAL [DMW93]. PSMs it and what it needs to achieve it, respectively referred to as
describe the reasoning process of a knowledge-based sys- the PSM’s competence, operational specification and re-
tem (KBS) in an implementation- and domain-independent quirements/assumptions (see Figure 1).
manner.
Work on PSMs covers different areas such as the iden-
Competence The competence of a PSM is a declarative
tification of task-specific PSMs (for diagnosis, planning,
description of the input-output behavior and describes
assessment, etc.), how to store and index PSMs in libraries,
what can be achieved by the PSM.
how to formalize PSMs, etc. The issues involved in reusing
PSMs include finding the right PSM (that does -part of- the Operational specification The operational specification
job), checking whether it is applicable in the situation at of a PSM describes the reasoning process which de-
hand, and modifying it to fit the domain. In order to reuse livers the specified competence if the required knowl-
PSMs successfully in a real-life application, one has to un- edge is provided. It consists of inference steps and the
derstand these processes. A PSM may be characterized as knowledge and control-flow between them. The in-
follows: ference steps specify the reasoning steps that together
accomplish the competence of the method. They are
A PSM specifies which inference steps have to be car-
described by their input/output relation and can be
ried out for achieving the goal of a task.
achieved by either a method (which means that a PSM
A PSM defines one or more control structures over can be hierarchically decomposed) or a primitive in-
these steps. ference (an atomic reasoning step which is not fur-
ther decomposed). The knowledge flow takes place
Knowledge roles specify the role that domain knowl- through dynamic roles, which are stores that act as in-
edge plays in each inference step. These knowledge put and output of inferences. Finally, the control of a
roles define a domain-independent generic terminol- PSM describes the order of execution of the inference
ogy. There are two types of roles: static roles describe steps. Control knowledge can be specified in advance,
the domain knowledge needed by the PSM; dynamic if known, or can be opportunistically determined at
roles form the input and output of inference steps. run time depending on the dynamic problem-solving
situation [Ben95]. Problem-solving methods can be
PSMs play an important role in knowledge engineering used to efficiently achieve goals of tasks through the
and knowledge acquisition. They can for instance be used application of domain knowledge [FS98]. They can
to efficiently achieve goals of tasks through the application play several roles in the knowledge engineering pro-
of domain knowledge [FS98], they can guide the acquisi- cess, such as guiding the acquisition process of do-
tion process of domain knowledge, and they can facilitate main knowledge and facilitating KBS development
KBS development through their reuse. through their reuse.
A. Gómez Pérez, V.R. Benjamins 1-7
Domain KBS specification (reasoning): a PSM can describe an
PSM task
knowledge efficient reasoning process that achieves the goal of a
task. In this sense, a PSM concerns the product of the
Figure 2: The two possible gaps that may prevent a PSM creation process, and is related to the design model of
for being applied to its context. a KBS.
Requirements/Assumptions Requirements/assumptions
Cognitive modeling: a PSM can describe a cogni-
of a PSM describe the domain knowledge needed by
tive model of human problem-solving. An interesting
the PSM to achieve its competence. Examples of such
question is to what extent PSMs can be used to gener-
requirements in a parametric design task include the
ate cognitively adequate explanations of the reasoning
availability of heuristics that link violated constraints
process of a knowledge-based system.
to possible repair actions (fixes), and the fact that a
preference relation must describe a complete ordering.
The requirements describe what a PSM expects in re- PSM development
turn for the competence it provides. Work in this area is concerned with how PSMs are con-
structed in the first place. One way to do this, is by ana-
The internal relationship between the competence and lyzing human problem-solving behavior and representing
operational descriptions of the method is that it has to be this behavior computationally. This has been tradition-
ensured that, assuming that the knowledge requirements ally the focus of Cognitive Psychology. Another way to
are satisfied, the operational description describes a way do this, is to perform reverse engineering of existing ex-
to achieve the competence [FS97]. pert systems, as has been performed by Clancey [Cla85]
when he “discovered” Heuristic Classification. These two
A PSM in context ways of developing PSMs essentially involve a creative
PSMs can be used to realize tasks by applying domain activity, for which no methodological support exists. In
knowledge. Thus, the external context of a PSM is formed the last decade, several methodologies have been devel-
by two parties: a task to be realized and domain knowl- oped to support knowledge modeling and the develop-
edge to be applied. When we want to use a PSM to build a ment of knowledge-based systems, such as CommonKADS
knowledge-based system, we have thus to connect the PSM [SAA 99, SWdH 94], Protégé [Mus93], MIKE [AFS98],
with both the task and the domain knowledge. Since PSMs Components of Expertise [Ste90], GDM [TvHWS93] and
are generic, reusable components, they may not always fit VITAL [DMW93].
perfectly in the context, or, in other words, there may be Other approaches propose principled or even semi-
gaps (see Figure 2). automatic approaches to PSM development. One can for
These gaps can exist for several reasons. In both direc- example start with specifying the global required compe-
tions (i.e. towards the domain knowledge and the task) the tence of the problem-solving method and then step-by-step
PSM may use different terminology than that of the domain refine this competence description into an operational prob-
knowledge and task, in which case a renaming process can lem solver [WAS98]. Another approach views the con-
bridge the gap. In the direction of the task, it may happen struction process of PSMs as a specific type of a config-
that the PSM’s competence is not strong enough to realize uration problem [tTvHSW98] and applies a well-known
what is specified by the task. In this case, to bridge the problem-solving method to solve this problem: propose-
gap, the task may be weakened by making simplifying as- critique-modify. Coming up with PSMs is one thing, but
sumptions. Towards the domain knowledge, the knowledge coming up with correct PSMs is another (a PSMs is correct
required by the PSM may not be fully given by the domain if it actually provides what is specified in its competence).
knowledge, in which case additional knowledge needs to Formal methods are applied to develop such correct PSMs
be acquired or can be assumed to exist. [PG98, FS97].
3.2 Issues in PSM research Reuse and libraries of PSMs
Problem-solving methods play an important role in knowl- When PSMs have been successfully developed for a par-
edge acquisition and knowledge engineering where they ticular application, it is worthwhile to formulate the PSMs
have several purposes: at a generic level. That is, the reusable parts of the PSMs
are identified and stored in a repository or a library. When
KBS construction (knowledge engineering): a PSM building a new application, this library can then be con-
can be helpful to describe the process of creating a sulted, preventing the system engineer from developing
problem solver that achieves the goal of a particu- a complete new system from scratch. Generally, reuse
lar task. Often this implies a task decomposition ap- of PSMs includes the following questions: which generic
proach. PSMs exist and how should a library of these methods be
A. Gómez Pérez, V.R. Benjamins 1-8
organized? How can PSMs be indexed in a way to support informal libraries provide structured textual represen-
their selection for a given application? How can we sup- tations of PSMs. Note that within the informal ap-
port the process of adapting a generic PSM to the specific proaches, PSM descriptions can vary from just tex-
circumstances of a given application? How can individual tual descriptions [Cha90], to highly structured de-
PSMs from a library be configured into a coherent problem scriptions using diagrams [Ben93].
solver? PSM libraries are of central importance if our aim
is to reuse as much as possible in a correct way. The granularity dimension distinguishes between li-
Current work in the braries with complex components, in the sense that the
PSM area focuses on method-description languages such PSMs realize a complete task [MZ98], and libraries
as UPML [FBMW99, GGM98]. Problem-solving meth- with fine-grained PSMs that realize a small part of the
ods that reside in libraries can be annotated with such lan- task. Several libraries contain both large and small
guages, so that they become more accessible to others (peo- building-blocks where the former are built up from the
ple and software agents). latter [Ben93, Cha90, BVB96].
The size dimension. The most comprehensive gen-
3.3 Libraries of PSMs eral library is the CommonKADS library [BvdV94]
PSMs represent a kind of best practice in KBS construc- which contains PSMs for diagnosis, prediction of
tion (cf. design patterns in object-oriented approaches behavior, assessment, design, planning, assignment
[GHJV95]). Instead of that knowledge engineers have to and scheduling and engineering modeling. The most
construct problem solvers from scratch, they can benefit extensive library for diagnosis [Ben93] contains 38
from previous successful experiences of other developers. PSMs for realizing 14 tasks related to diagnosis. The
The use of best-practice components has as benefits that library for parametric design [MZ98] consists of five
they reflect years of experience, enabling thorough valida- PSMs, several of them being variations of Propose &
tion and verification of the components, which enhances Revise [Mar88b]. The design library of [Cha90] men-
the quality of the software. Once we have a collection tions about 15 PSMs.
of such reasoning patterns, interesting issues arise such as
how to structure and organize the collection and how to in- The type of a library is determined by its characteri-
dex the components. zation in terms of the above dimensions. Each type has
a specific role in the knowledge engineering process and
3.3.1 Types of PSM libraries has strong and weak points. The more general (i.e. task-
neutral) PSMs in a library are, the more reusable they are,
Currently, there exist several libraries with PSMs. They all because they do not make any commitment to particular
aim at facilitating the knowledge-engineering process, yet tasks. However, at the same time, applying such a PSM in a
they differ in various ways. In particular, libraries differ particular application requires considerable refinement and
along dimensions such as generality, formality, granularity adaptation. This phenomenon is known as the reusability–
and size. usability trade-off [KBD 91]. Recently, research has
The generality dimension describes whether PSMs in been conducted to overcome this dichotomy by introduc-
a library are developed for a particular task. Task- ing adapters that gradually adapt task-neutral PSMs to task-
specific libraries contain PSMs that are specialized in specific ones [FG97] and by semi-automatically construct-
solving (parts of) a specific task such as diagnosis or ing the mappings between task-neutral PSMs and domain
design. Their “task-specificness” resides mainly in the knowledge [BBvH96].
terminology in which the PSMs are formulated. Ex- Libraries with informal PSMs provide above all support
amples include libraries for design [Cha90, MZ98], for the conceptual specification phase of the KBS, that is,
assessment [VL93], diagnosis [Ben93] and planning they help significantly in constructing the reasoning part
[BVB96, BHB97]. The CommonKADS library can of the expertise model of a KBS [SWB93]. Because such
be viewed as an extensive collection of task-specific PSMs are informal, they are relatively easy to understand
PSMs [BvdV94]. Task-independent libraries provide and malleable to fit a particular application. The disadvan-
problem-solving methods that are not formulated in tage is – not surprisingly – that still much work has to be
task-specific terminology [Abe93]. done before arriving at an implemented system. Libraries
with formal PSMs are particularly important if the PSMs
The formality dimension divides the libraries in in- need to have some guaranteed properties, e.g. for use in
formal, formal and implemented ones. Implemented safety-critical systems such as nuclear power plants. Their
libraries provide operational specifications of PSMs, disadvantage is that they are hard to understand for hu-
which are directly executable [PETM92, GTRM94]. mans [BH95] and limit the expressiveness of the knowl-
Formal libraries allow for formal verification of prop- edge engineer. Apart from the possibility to prove prop-
erties of PSMs [Abe93, Abe95, BA97, tT97]. Finally, erties, formal PSMs have the additional advantage of be-
A. Gómez Pérez, V.R. Benjamins 1-9
ing a step closer to an implemented system. Libraries with work is currently being performed to shed more light on the
implemented PSMs allow the construction of fully opera- role of assumptions in libraries for knowledge engineering
tional systems. The other side of the coin is, however, that [BFS96, FB98, FG97].
the probability that operational PSMs exactly match the re- A last proposal to organize libraries of PSMs is based
quirements of the knowledge engineer, is lower. on a suite of so-called problem types (or tasks, for the pur-
Developing a KBS using libraries with coarse-grained pose of this article tasks and problem types are treated as
PSMs, amounts to selecting the most suitable PSM and synonyms) [Bre94a, Bre94b]. The suite describes problem
then adapt it to the particular needs of the application types according to the way that problems depend on each
[MZ98]. The advantage is that this process is quite sim- other. The solution to one problem forms the input to an-
ple as it involves only one component. The disadvantage other problem. For example, the output of a prediction task
is, however, that it is unlikely that such a library will have is a certain state, which can form the input to a monitoring
broad coverage, since each application might need a dif- task that tries to detect problems, which on their turn can be
ferent (coarse-grained) PSM. The alternative approach is the input to a diagnosis task. It turns out that these problem
to have a library with fine-grained PSMs, which are then dependencies recur in many different tasks. According to
combined together (i.e. configured) into a reasoner, either this principle, PSMs are stored under the problem type they
manually [PETM92] or automatically [Ben95, BHB97]. can solve. Selection of PSMs in such a library would first
identify the problem type involved (or task), and then look
3.3.2 Organization of libraries at the respective PSMs for this task.
There are several alternatives for organizing a library and 3.4 Industrial applications
each of them has consequences for indexing PSMs and for
Building KBSs from reusable components in an academic
their selection. Finding the “best” organization principle
setting is one thing. Doing the same for real industrial
for such libraries is still an issue of debate. In the following,
applications is another. So far, several industrial applica-
we will present some organization principles.
tions have been built, but only a few have been reported
Several researchers propose to organize libraries as a
in the literature. Unilever reports on the successful use
task–method decomposition structure [CJS92, PETM92,
of a library with diagnostic problem-solving methods for
Ste93], and some available libraries are organized in this
building a knowledge-based system for diagnosing chem-
way [Ben93, Bre94a, BVB96]. According to this organiza-
ical production processes [SA97, SA98]. A road traffic
tion structure, a task can be realized by several PSMs, each
management knowledge-based system [MHC98] is opera-
consisting of primitive and/or composite subtasks. Com-
tional in the cities of Madrid and Barcelona in Spain. IBM,
posite subtasks can again be realized by alternative meth-
Japan reports a knowledge system for job scheduling of
ods, etc. Principles for library design according to this prin-
production processes [HY98]. The system has been built
ciple are discussed in [Ors96b, Ors96a]. In a library or-
by using a domain-oriented library of scheduling problem-
ganized according to the task-method principle, PSMs are
solving methods. Metrics show that a significant percent-
indexed, based on two factors: (1) on the competence of
age of existing code has been reused in the new application.
the PSMs – which specifies what a PSM can achieve, and
Knowledge-based systems for plant classification, service
(2) on their assumptions – which specify the assumptions
support for printing machines, and rheumatology have been
under which the PSM can be applied correctly, such as its
developed from reusable methods, as reported in [Pup98].
requirements on domain knowledge [HY98]. Selection of
PSMs from such libraries first considers the competence
of PSMs (selecting those whose competences match the 4 Conclusions and future work
task at hand), and then the assumptions of PSMs (select- In this paper, we reviewed recent work in the area of ontolo-
ing those whose assumptions are satisfied). gies [ST99, UT98, vSW97, GP95] and problem-solving
Libraries can also be organized, based on the functional- methods [BF98]. The current state-of-the-art is that there
ity of PSMs, in which case PSMs with similar functionality is now a fairly good understanding of what ontologies are
are stored together. In addition, the functionality of PSMs and what they do. Current work takes this body of existing
can be configured from pre-established parameters and val- work and starts from that in new directions.
ues [tT97]. In the ontology world, emphasis is now put on integra-
Another criterion to structure libraries of PSMs is based tion of heterogeneous ontologies and on characterizing and
on assumptions, which specify under what conditions brokering ontologies on the WWW. Also, efforts are made
PSMs can be applied. Assumptions can refer to domain to connect to the object-oriented world and to databases.
knowledge (e.g. a certain PSM needs a causal domain Ontologies are clearly becoming more and more important
model) or to task knowledge (a certain PSM generates lo- in a large variety of areas.
cally optimal solutions). To our knowledge, there does Also, work in the problem-solving method world
not exist a library organized following this principle, but moves on. Several libraries of methods exist and
A. Gómez Pérez, V.R. Benjamins 1-10
efforts are made to make these libraries accessible Sharable and Reusable Problem-Solving Methods,
and interoperable. The European IBROW project KAW’96,
(http://www.swi.psy.uva.nl/projects/IBROW3/home.html) http://ksi.cpsc.ucalgary.ca/KAW/KAW96/KAW96Proc.html
aims at building a brokering service that can configure
knowledge system out of reusable PSMs that reside in li- References
braries on the Internet.
The integration of ontologies and PSMs is also a [ABB 98] G. Aguado, J. Bateman, A. Bañon, S. Bernar-
promising new direction, especially when ontologies and dos, M. Fernändez, A. Gómez-Pérez, E. Nieto,
A. Olalla, R. Plaza, and A. Sanchez. ONTOGEN-
databases are integrated. The PSMs can then provide rea-
ERATION: Reusing domain and linguistic ontolo-
soning services on top of these databases, which can lead
gies for spanish text generation. In A. Gomez-
to dynamically configured active databases rather than pas- Perez and V. R. Benjamins, editors, Proceedings
sive repositories of static knowledge waiting to be queried. of the Workshop on Applications of Ontologies
and Problem-Solving Methods, held in conjunc-
5 Relevant links tion with ECAI-98, pages 1–10, Brighton, UK, Au-
gust 1998. ECAI.
For an extensive collection of (alphabetically ordered) links
to work on ontologies and problem-solving methods, in- [Abe93] M. Aben. Formally specifying re-usable knowl-
cluding proceedings and events, see: edge model components. Knowledge Acquisition,
5:119–141, 1993.
http://www.cs.utexas.edu/users/mfkb/related.html. The
homepage of the PSM mailing list can be accessed at: [Abe95] M. Aben. Formal Methods in Knowledge Engi-
http://www.swi.psy.uva.nl/mailing-lists/kaw- neering. PhD thesis, University of Amsterdam,
psm/home.html. Amsterdam, 1995.
A list of relevant workshops that are accessible on the [AFS98] J. Angele, D. Fensel, and R. Studer. Developing
WWW is included below: knowledge-based systems with mike. Journal of
Automated Software Engineering, to appear, 1998.
Applications of Ontologies and Problem-Solving [AGLP98] J. Arpirez, A. Gómez-Pérez, A. Lozano, and
Methods, ECAI’98 (European Conference on AI), S. Pinto. (onto)2agent: An ontology-based www
http://delicias.dia.fi.upm.es/WORKSHOP/ECAI98/index.html broker to select ontologies. In A. Gomez-Perez
and V. R. Benjamins, editors, Proceedings of
Building, Maintaining, and Using Organizational the Workshop on Applications of Ontologies and
Memories, ECAI’98, Problem-Solving Methods, held in conjunction
http://www.aifb.uni-karlsruhe.de/WBS/ECAI98OM/ with ECAI-98, pages 16–24, Brighton, UK, Au-
gust 1998. ECAI.
Formal Ontologies in Information Systems
(FOIS’98), [BA97] V. R. Benjamins and M. Aben. Structure-
http://krr.irst.itc.it:1024/fois98/program.html preserving KBS development through reusable li-
braries: a case-study in diagnosis. International
Intelligent Information Integration, ECAI’98, Journal of Human-Computer Studies, 47:259–288,
http://www.tzi.de/grp/i3/ws-ecai98/ 1997.
[Bat94] J. A. Bateman. KPML: The KOMET-Penman
Sharable and Reusable Components for Knowledge (multilingual) development environment. Tech-
Systems, KAW’98 (Workshop on Knowledge Acqui- nical report, GMD/IPSI, Darmstadt (Germany),
sition, Modeling, and Management), 1994.
http://ksi.cpsc.ucalgary.ca/KAW/KAW98/KAW98Proc.html
[BBvH96] P. Beys, V. R. Benjamins, and G. van Hei-
Ontological Engineering, AAAI Spring Symp. Series, jst. Remedying the reusability-usability tradeoff
for problem-solving methods. In B. R. Gaines
Stanford, Calif., 1997,
and M. A. Musen, editors, Proceedings of the
http://www.aaai.org/Symposia/Spring/1997/sss-97.html
10th Banff Knowledge Acquisition for Knowledge-
Problem-Solving Methods, IJCAI’97 (Int’l Joint Based Systems Workshop, pages 2.1–2.20, Alberta,
Canada, 1996. SRDG Publications, University of
Conf. AI),
Calgary.
http://www.aifb.uni-karlsruhe.de/WBS/dfe/PSM/main.html
[BC88] T. Bylander and B. Chandrasekaran. Generic
Ontological Engineering, ECAI’96, tasks in knowledge-based reasoning: The right
http://wwwis.cs.utwente.nl:8080/kbs/EcaiWorkshop/homepage.html level of abstraction for knowledge acquisition. In
B. Gaines and J. Boose, editors, Knowledge Ac-
Sharable and Reusable Ontologies, KAW’96, quisition for Knowledge Based Systems, volume 1,
http://ksi.cpsc.ucalgary.ca/KAW/KAW96/KAW96Proc.html pages 65–77. Academic Press, London, 1988.
A. Gómez Pérez, V.R. Benjamins 1-11
[Ben93] V. R. Benjamins. Problem Solving Methods for [BMF95] J. A. Bateman, B. Magnini, and G. Fabris. The
Diagnosis. PhD thesis, University of Amsterdam, generalized upper model knowledge bare: Organi-
Amsterdam, The Netherlands, 1993. zation and use. In N. J. I. Mars, editor, Towards
Very Large Knowledge Bases: Knowledge Build-
[Ben95] V. R. Benjamins. Problem-solving methods for di-
ing & Knowledge Sharing., pages 60–72. IOS
agnosis and their role in knowledge acquisition.
Press, Amsterdam, NL, 1995.
International Journal of Expert Systems: Research
and Applications, 8(2):93–120, 1995. [Bor97] W. N. Borst. Construction of Engineering Ontolo-
gies. PhD thesis, University of Twente, Enschede,
[BF98] V. R. Benjamins and D. Fensel. Editorial: 1997.
Problem-solving methods. International Journal
[BPG96] V. R. Benjamins and C. Pierret-Golbreich. As-
of Human-Computer Studies, 49(4):305–313, Oc-
sumptions of problem-solving methods. In
tober 1998. Special issue on Problem-Solving
N. Shadbolt, K. O’Hara, and G. Schreiber, edi-
Methods.
tors, Lecture Notes in Artificial Intelligence, 1076,
[BFDGP97] V. R. Benjamins, D. Fensel, S. Decker, and 9th European Knowledge Acquisition Workshop,
A. Gomez-Perez. (KA) : Building ontologies for EKAW-96, pages 1–16, Berlin, 1996. Springer-
the internet: a mid term report. International Jour- Verlag.
nal of Human-Computer Studies, page in press, [Bre94a] J. Breuker. Components of problem solving and
1997. types of problems. In L. Steels, A. T. Schreiber,
[BFGPGP98] M. Blázquez, M. Fernández, J. M. García-Pinar, and W. van de Velde, editors, Lecture Notes in Ar-
and A. Gómez-Pérez. Building ontologies at the tificial Intelligence, 867, 8th European Knowledge
knowledge level using the ontology design envi- Acquisition Workshop, EKAW-94, pages 118–136,
ronment. In Proceedings of the Eleventh Workshop Berlin, Germany, 1994. Springer-Verlag.
on Knowledge Acquisition, Modeling and Man- [Bre94b] J. Breuker. A suite of problem types. In J. Breuker
agement, KAW’98, Banff, Canada, 1998. and W. van de Velde, editors, CommonKADSLi-
brary for Expertise Modeling, pages 57–87. IOS
[BFS96] V. R. Benjamins, D. Fensel, and R. Straatman. As-
Press, Amsterdam, The Netherlands, 1994.
sumptions of problem-solving methods and their
role in knowledge engineering. In W. Wahlster, [BVB96] Barros, L. Nunes de, A. Valente, and V. R. Ben-
editor, Proc. ECAI–96, pages 408–412. J. Wiley & jamins. Modeling planning tasks. In Third In-
Sons, Ltd., 1996. ternational Conference on Artificial Intelligence
Planning Systems, AIPS-96, pages 11–18. Amer-
[BGM96] S. Borgo, N. Guarino, and C. Masolo. Stratified ican Association of Artificial Intelligence (AAAI),
ontologies: the case of physical objects. In Pro- 1996.
ceedings of the Workshop on Ontological Engi-
[BvdV94] J. Breuker and W. van de Velde, editors. Com-
neering, held in conjunction with ECAI-96, pages
monKADS Library for Expertise Modeling. IOS
5–15, Budapest, August 1996. ECAI.
Press, Amsterdam, The Netherlands, 1994.
[BH95] J. P. Bowen and M. G. Hinchey. Ten commands [CFF 97] V. K. Chaudhri, A. Farquhar, R. Fikes, P. Karp, and
of formal methods. IEEE Computer, 28(4):56–63, J. Rice. The generic frame protocol 2.0. Technical
1995. report, Stanford, 1997.
[BHB97] Barros, L. Nunes de, J. Hendler, and V. R. Ben- [Cha90] B. Chandrasekaran. Design problem solving: A
jamins. Par-KAP: a knowledge acquisition tool for task analysis. AI Magazine, 11:59–71, 1990.
building practical planning systems. In M. E. Pol- [CJB99] B. Chandrasekaran, J.R. Josephson, and V.R. Ben-
lack, editor, Proc. of the 15th IJCAI, pages 1246– jamins. Ontologies: What are they? why do we
1251, Japan, 1997. International Joint Conference need them? IEEE Intelligent Systems and Their
on Artificial Intelligence, Morgan Kaufmann Pub- Applications, 14(1):20–26, 1999. Special Issue on
lishers, Inc. Also published in Proceedings of the Ontologies.
Ninth Dutch Conference on Artificial Intelligence,
[CJS92] B. Chandrasekaran, T. R. Johnson, and J. W.
NAIC’97, K. van Marcke, W. Daelemans (eds),
Smith. Task-structure analysis for knowl-
University of Antwerp, Belgium, pages 137–148.
edge modeling. Communications of the ACM,
[BKMW90] J. A. Bateman, R. T. Kasper, J. D. Moore, and 35(9):124–137, 1992.
R. A. Whitney. A general organization of knowl- [Cla85] W. J. Clancey. Heuristic classification. Artificial
edge for natural language processing: the penman Intelligence, 27:289–350, 1985.
upper model. Technical report, USC/ISI, Marina
[DMW93] J. Domingue, E. Motta, and S. Watt. The emerg-
del Rey, CA (USA), 1990.
ing VITAL workbench. In Aussenac et al., editor,
[BLC96] A. Bernaras, I. Laresgoiti, and J. Corera. Building EKAW’93 Knowledge Acquisition for Knowledge-
and reusing ontologies for electrical network ap- Based Systems. Lecture Notes in Artificial In-
plications. In Proceedings of the 12th ECAI, pages telligence, LNCS 723, Berlin, Germany, 1993.
298–302, 1996. Springer-Verlag.
A. Gómez Pérez, V.R. Benjamins 1-12
[Dom98] J. Domingue. Tadzebao and webonto: Dis- [GG95] N. Guarino and P. Giaretta. Ontologies and knowl-
cussing, browsing, and editing ontologies on the edge bases: Towards a terminological clarification.
web. In Proceedings of the Eleventh Workshop In N. J. I. Mars, editor, Towards Very Large Knowl-
on Knowledge Acquisition, Modeling and Man- edge Bases: Knowledge Building & Knowledge
agement, KAW’98, Banff, Canada, 1998. Sharing, pages 25–32. IOS Press, Amsterdam, NL,
1995.
[FB98] D. Fensel and V. R. Benjamins. The role of
assumptions in knowledge engineering. Inter- [GGM98] J. H. Gennari, W. Grosso, and M. Musen. A
national Journal on Intelligent Systems (IJIS), method-description language: An initial ontol-
13(7):715–748, 1998. ogy with examples. In B. R. Gaines and M. A.
Musen, editors, Proceedings of the 11th Banff
[FBMW99] D. Fensel, V. R. Benjamins, E. Motta, and
Workshop on Knowledge Acquisition, Modeling
B. Wielinga. UPML: A framework for knowledge and Management (KAW’98), pages SHARE.11.–
system reuse. In Proceedings of the 16th Interna- SHARE.11.18, Alberta, Canada, 1998. SRDG
tional Joint Conference on AI (IJCAI-99), page to
Publications, University of Calgary.
appear, Sweden, 1999.
[GHJV95] E. Gamma, R. Helm, R. Johnson, and J. Vlissides.
[FDES98] D. Fensel, S. Decker, M. Erdmann, and R. Studer. Design Patterns: Elements of Reusable Object-
Ontobroker: The very high idea. In Proceed- Oriented Software. Addison Wesley, 1995.
ings of the 11th International Flairs Conference
(FLAIRS-98), Sanibal Island, Florida, 1998. [GMV99] N. Guarino, C. Masolo, and G. Vetere. Ontoseek:
Using large linguistic ontologies for accessing on-
[FFR97] A. Farquhar, R. Fikes, and J. Rice. The ontolingua line yellow pages and product catalogs. IEEE In-
server: a tool for collaborative ontology construc- telligent Systems and Their Applications, page to
tion. International Journal of Human-Computer appear, 1999.
Studies, 46(6):707–728, June 1997.
[GO94] T. Gruber and R. Olsen. An ontology for engi-
[FG97] D. Fensel and R. Groenboom. Specifying neering mathematics. Technical report, Knowl-
knowledge-based systems with reusable compo- edge Systems Laboratory, Stanford University,
nents. In Proceedings 9th Int. Conference on CA, 1994.
Software Engineering and Knowledge Engineer- [GP95] N. Guarino and R. Poli. The role of ontology in
ing SEKE’97, pages 349–357, Madrid, 1997. the information technology. International Jour-
[FGPPP99] M. Fernandez, A. Gomez-Perez, J. Pazos, and Ale- nal of Human-Computer Studies, 43(5/6):623–
jandro Pazos. Ontology of tasks and methods. 965, 1995. Special issue on ontology.
IEEE Intelligent Systems and Their Applications, [GP98] A. Gómez-Pérez. Knowledge sharing and reuse.
14(1):37–46, January/February 1999. In J. Liebowitz, editor, The Handbook of Applied
[FS97] D. Fensel and A. Schönegge. Using KIV to spec- Expert Systems. CRC, 1998.
ify and verify architecturs of knowlege-based sys- [GPR99] A. Gomez-Perez and M. D. Rojas. Ontologi-
tems. In Proc. of 12th IEEE International Confer- cal reengineering for reuse. In D. Fensel and
ence on Automated Software Engineering (ASEC- R. Studer, editors, Proceedings of the EKAW-99,
97). IEEE, 1997. page to appear. Springer-Verlag, 1999.
[FS98] D. Fensel and R. Straatman. The essence of [Gru93] T. R. Gruber. A translation approach to portable
problem-solving methods: making assumptions to ontology specifications. Knowledge Acquisition,
gain efficiency. IJHCS, 48:181–215, 1998. 5:199–220, 1993.
[FvdR98] M. Frohlich and R. P. van de Riet. Using multi- [Gru95] T. R. Gruber. Towards principles for the design
ple ontologies in a framework for natural language of ontologies used for knowledge sharing. In-
generation. In A. Gomez-Perez and V. R. Ben- ternational Journal of Human-Computer Studies,
jamins, editors, Proceedings of the Workshop on 43:907–928, 1995.
Applications of Ontologies and Problem-Solving [GTRM94] J. H. Gennari, S. W. Tu, T. E. Rotenfluh, and
Methods, held in conjunction with ECAI-98, pages M. A. Musen. Mapping domains to methods in
67–77, Brighton, UK, August 1998. ECAI. support of reuse. International Journal of Human-
[GF92] M. R. Genesereth and R. E. Fikes. Knowledge in- Computer Studies, 41:399–424, 1994.
terchange format, version 3.0, reference manual. [Gua98] N. Guarino. Some ontological principles for de-
Technical report, Logic-92-1, Computer Science signing upper level lexical resources. In Pro-
Dept., Stanford University, 1992. ceedings of the First International Conference on
[GF95] M. Grüninger and M. Fox. Methodology for the Language Resources and Evaluation, pages –,
design and evaluation of ontologies. In Proceed- Granada, 1998.
ings of the Workshop on Basic Ontological Issues [Hoe98] E. Hoenkamp. Spotting ontological lacunae
in Knowledge Sharing held in conjunction with through spectrum analysis of retrieved documents.
IJCAI-95, Montreal, Canada, 1995. In A. Gomez-Perez and V. R. Benjamins, editors,
A. Gómez Pérez, V.R. Benjamins 1-13
Proceedings of the Workshop on Applications of the search paradigm with task and method ontolo-
Ontologies and Problem-Solving Methods, held in gies. International Journal of Human-Computer
conjunction with ECAI-98, pages 73–77, Brighton, Studies, 49(4):437–470, 1998. Special issue on
UK, August 1998. ECAI. Problem-Solving Methods.
[HY98] M. Hori and T. Yoshida. domain-oriented library [NFF 91] R. Neches, R. E. Fikes, T. Finin, T. R. Gru-
of scheduling methods: design principal and real- ber, T. Senator, and W. R. Swartout. Enabling
life application. International Journal of Human- technology for knowledge sharing. AI Magazine,
Computer Studies, 49(4):601–626, 1998. Special 12(3):36–56, 1991.
issue on Problem-Solving Methods. [Ors96a] K. Orsvärn. Knowledge Modelling with Libraries
[KBD 91] G. Klinker, C. Bhola, G. Dallemagne, D. Marques, of Task Decomposition Methods. PhD thesis,
and J. McDermott. Usable and reusable program- Swedish Institute of Computer Science, 1996.
ming constructs. Knowledge Acquisition, 3:117– [Ors96b] K. Orsvärn. Principles for libraries of task decom-
136, 1991. position methods – conclusions from a case-study.
[KL94] K. Knight and S. Luk. Building a large knowledge In N. Shadbolt, K. O’Hara, and G. Schreiber, edi-
base for machine translation. In AAAI–94, 1994. tors, Lecture Notes in Artificial Intelligence, 1076,
9th European Knowledge Acquisition Workshop,
[KLW95] M. Kifer, G. Lausen, and J. Wu. Logical foun-
EKAW-96, pages 48–65. Springer-Verlag, 1996.
dations of object-oriented and frame-based lan-
guages. Journal of the ACM, 1995. [PETM92] A.R. Puerta, J. Egar, S. Tu, and M. Musen. A
multiple-method shell for the automatic generation
[LG90] D. B. Lenat and R. V. Guha. Building large
of knowledge acquisition tools. Knowledge Acqui-
knowledge-based systems. Representation and in-
sition, 4:171–196, 1992.
ference in the Cyc project. Addison-Wesley, Read-
ing, Massachusetts, 1990. [PG98] C. Pierret-Golbreich. Supporting organization and
use of problem-solving methods libraries by for-
[Mac91] R. MacGregor. Inside the LOOM classifier.
mal methods. International Journal of Human-
SIGART Bulletin, 2(3):70–76, June 1991.
Computer Studies, 49(4):471–495, 1998. Special
[Mah96] K. Mahesh. Ontology development for machine issue on Problem-Solving Methods.
translation: Ideology and methodology. Technical
[Pup98] F. Puppe. Knowledge reuse among diagnostic
report, Computer Research Laboratory, New Mex-
problem-solving methods in the shell-kit D3. In-
ico State University, 1996.
ternational Journal of Human-Computer Studies,
[Mar88a] S. Marcus, editor. Automating knowledge acquisi- 49(4):627–649, 1998. Special issue on Problem-
tion for expert systems. Kluwer, Boston, 1988. Solving Methods.
[Mar88b] S. Marcus. SALT: a knowledge-acquisition tool [Ros94] D. Rosner. Generating multilingual documents
for propose-and-revise systems. In S. Marcus, edi- from a knowledge base: The techdoc project.
tor, Automating Knowledge Acquisition for Expert Technical report, FAW Ulm, Ulm (Germany),
Systems, pages 81–123. Kluwer, Boston, 1988. 1994.
[MHC98] M. Molina, J. Hernández, and J. Cuena. A struc- [SA97] P-H. Speel and M. Aben. Applying a library of
ture of problem-solving methods for real-time de- problem solving methods on a real-life task. In-
cision support in traffic control. International ternational Journal of Human-Computer Studies,
Journal of Human-Computer Studies, 49(4):577– 46(May):627–652, 1997.
600, 1998. Special issue on Problem-Solving
[SA98] P-H. Speel and M. Aben. Preserving concep-
Methods.
tual structures in design and implementation of in-
[Mil90] G. A. Miller. WORDNET: an online lexical dustrial KBSs. International Journal of Human-
database. International Journal of Lexicography, Computer Studies, 49(4):547–575, 1998. Special
3(4):235–312, 1990. issue on Problem-Solving Methods.
[Mus93] M. Musen. An overview of knowledge acquisition. [SAA 99] A. Th. Schreiber, J. M. Akkermans, A. A. An-
In J. M. David, J. P. Krivine, and R. Simmons, edi- jewierden, R. de Hoog, N. R. Shadbolt, W. Van
tors, Second Generation Expert Systems. Springer de Velde, and B. J. Wielinga. Engineering
Verlag, 1993. and Managing Knowledge, The CommonKADS
[MVI95] R. Mizoguchi, J. Vanwelkenhuysen, and M. Ikeda. methodology. MIT Press, 1999.
Task ontology for reuse of problem solving knowl- [SBF98] R. Studer, V. R. Benjamins, and D. Fensel. Knowl-
edge. In N. J. I. Mars, editor, Towards Very Large edge engineering, principles and methods. Data
Knowledge Bases: Knowledge Building & Knowl- and Knowledge Engineering, 25(1-2):161–197,
edge Sharing., pages 46–57. IOS Press, Amster- 1998.
dam, NL, 1995. [SCC 93] O. Stock, G. Carenini, F. Cecconi, E. Franconi,
[MZ98] E. Motta and Z. Zdrahal. A library of problem- A. Lavelli, B. Magnini, F. Pianesi, M. Ponzi,
solving components based on the integration of V. Samek-Lodovici, and C. Strapparava. Alfresco:
A. Gómez Pérez, V.R. Benjamins 1-14
Enjoying the combination of natural language pro- pages 428–455. Springer-Verlag, Berlin Heidel-
cessing and hypermedia for information explo- berg, Germany, 1993.
ration. In Mark T. Maybury, editor, Intelligent [UG96] M. Uschold and M. Gruninger. Ontologies: prin-
Multimedia Interfaces, chapter 9, pages 197–224. ciples, methods, and applications. Knowledge En-
The MIT Press, 1993. gineering Review, 11(2):93–155, 1996.
[SG95] W. Swartout and Y. Gil. Expect: Explicit represen- [Usc96] M. Uschold. Building ontologies: Towards a uni-
tations for flexible acquisition. In Proceedings of fied methodology. In Expert Systems 96, 1996.
the Ninth Knowledge Acquisition for Knowledge-
Based Systems Workshop, 1995. [UT98] M. Uschold and A. Tate. Special issue on ontolo-
gies. The Knowledge Engineering Review, 13(1),
[Sow97] J. F. Sowa. Knowledge Representation: Logi-
1998.
cal, Philosophical, and Computational Founda-
tions. Book draft, 1997. [vdVSM94] P. E. van de Vet, P.-H. Speel, and N. J. I. Mars. The
plinius ontology of ceramic materials. In N. J. I.
[SPKR97] W. Swartout, R. Patil, K. Knight, and T. Russ.
Mars, editor, Working papers European Confer-
Toward distributed use of large-scale ontologies.
ence on Artificial Intelligence ECAI’94 Workshop
In Spring Symposium Series on Ontological En-
on Implemented Ontologies, pages 187–206, Am-
gineering, pages 33–40, Stanford, 1997. AAAI
sterdam, 1994. ECCAI.
Press.
[VL93] A. Valente and C. Löckenhoff. Organization as
[ST99] B. Swartout and A. Tate. Coming to terms with on-
guidance: A library of assessment models. In Pro-
tologies. IEEE Intelligent Systems and Their Ap-
ceedings of the Seventh European Knowledge Ac-
plications, 14(1):18–19, 1999. Special Issue on
quisition Workshop (EKAW’93), Lecture Notes in
Ontologies.
Artificial Intelligence, LNCS 723, pages 243–262,
[Ste90] L. Steels. Components of expertise. AI Magazine, 1993.
11(2):28–49, Summer 1990.
[vSW97] G. van Heijst, A. T. Schreiber, and B. J. Wielinga.
[Ste93] L. Steels. The componential framework and its Using explicit ontologies in KBS development. In-
role in reusability. In Jean-Marc David, Jean-Paul ternational Journal of Human-Computer Studies,
Krivine, and Reid Simmons, editors, Second Gen- 46(2/3):183–292, 1997.
eration Expert Systems, pages 273–298. Springer-
[WAS98] B. J. Wielinga, J. M. Akkermans, and A. Th.
Verlag, Berlin Heidelberg, Germany, 1993.
Schreiber. A competence theory approach to prob-
[SWB93] A. Th. Schreiber, B. J. Wielinga, and J. A. lem solving method construction. International
Breuker, editors. KADS: A Principled Approach Journal of Human-Computer Studies, 49(4):315–
to Knowledge-Based System Development, vol- 338, 1998. Special issue on Problem-Solving
ume 11 of Knowledge-Based Systems Book Series. Methods.
Academic Press, London, 1993.
[WMK95] J. Dukes-Schlossberg W. Mark and R. Kerber. On-
[SWdH 94] A. Th. Schreiber, B. J. Wielinga, R. de Hoog, tological commitments and domain-specific archi-
J. M. Akkermans, and W. Van de Velde. Com- tectures: Experience with comet and cosmos. In
monKADS: A comprehensive methodology for N. J. I. Mars, editor, Towards Very Large Knowl-
KBS development. IEEE Expert, 9(6):28–37, De- edge Bases: Knowledge Building & Knowledge
cember 1994. Sharing., pages 33–44. IOS Press, Amsterdam,
[SWJ95] A. Th. Schreiber, B. J. Wielinga, and W. H. J. Jan- NL, 1995.
sweijer. The KACTUS view on the ’O’ word. In
IJCAI Workshop on Basic Ontological Issues in
Knowledge Sharing, 1995.
[tT97] A. ten Teije. Automated Configuration of Problem
Solving Methods in Diagnosis. PhD thesis, Univer-
sity of Amsterdam, Amsterdam, The Netherlands,
1997.
[tTvHSW98] A. ten Teije, F. van Harmelen, A. Th. Schreiber,
and B. Wielinga. Construction of problem-solving
methods as parametric design. International Jour-
nal of Human-Computer Studies, 49(4):363–389,
1998. Special issue on Problem-Solving Methods.
[TvHWS93] P. Terpstra, G. van Heijst, B. Wielinga, and
N. Shadtbolt. Knowledge acquisition support
through generalised directive models. In Jean-
Marc David, Jean-Paul Krivine, and Reid Sim-
mons, editors, Second Generation Expert Systems,
A. Gómez Pérez, V.R. Benjamins 1-15