Formal Ontologies Meet Industry [1 ed.] 9781607504580, 9781607500476

Citation preview

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

FORMAL ONTOLOGIES MEET INDUSTRY

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

Frontiers in Artificial Intelligence and Applications FAIA covers all aspects of theoretical and applied artificial intelligence research in the form of monographs, doctoral dissertations, textbooks, handbooks and proceedings volumes. The FAIA series contains several sub-series, including “Information Modelling and Knowledge Bases” and “Knowledge-Based Intelligent Engineering Systems”. It also includes the biennial ECAI, the European Conference on Artificial Intelligence, proceedings volumes, and other ECCAI – the European Coordinating Committee on Artificial Intelligence – sponsored publications. An editorial panel of internationally well-known scholars is appointed to provide a high quality selection. Series Editors: J. Breuker, R. Dieng-Kuntz, N. Guarino, J.N. Kok, J. Liu, R. López de Mántaras, R. Mizoguchi, M. Musen, S.K. Pal and N. Zhong

Volume 198

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

Recently published in this series Vol. 197. R. Hoekstra, Ontology Representation – Design Patterns and Ontologies that Make Sense Vol. 196. F. Masulli et al. (Eds.), Computational Intelligence and Bioengineering – Essays in Memory of Antonina Starita Vol. 195. A. Boer, Legal Theory, Sources of Law and the Semantic Web Vol. 194. A. Petcu, A Class of Algorithms for Distributed Constraint Optimization Vol. 193. B. Apolloni, S. Bassis and M. Marinaro (Eds.), New Directions in Neural Networks – 18th Italian Workshop on Neural Networks: WIRN 2008 Vol. 192. M. Van Otterlo (Ed.), Uncertainty in First-Order and Relational Domains Vol. 191. J. Piskorski, B. Watson and A. Yli-Jyrä (Eds.), Finite-State Methods and Natural Language Processing – Post-proceedings of the 7th International Workshop FSMNLP 2008 Vol. 190. Y. Kiyoki et al. (Eds.), Information Modelling and Knowledge Bases XX Vol. 189. E. Francesconi et al. (Eds.), Legal Knowledge and Information Systems – JURIX 2008: The Twenty-First Annual Conference Vol. 188. J. Breuker et al. (Eds.), Law, Ontologies and the Semantic Web – Channelling the Legal Information Flood Vol. 187. H.-M. Haav and A. Kalja (Eds.), Databases and Information Systems V – Selected Papers from the Eighth International Baltic Conference, DB&IS 2008 Vol. 186. G. Lambert-Torres et al. (Eds.), Advances in Technological Applications of Logical and Intelligent Systems – Selected Papers from the Sixth Congress on Logic Applied to Technology Vol. 185. A. Biere et al. (Eds.), Handbook of Satisfiability

ISSN 0922-6389

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

Formal Ontologies Meet Industry

Edited by

Roberta Ferrario Laboratory for Applied Ontology, Institute for Cognitive Sciences and Technologies, National Research Council, Trento, Italy

and

Alessandro Oltramari

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

Laboratory for Applied Ontology, Institute for Cognitive Sciences and Technologies, National Research Council, Trento, Italy

Amsterdam • Berlin • Tokyo • Washington, DC

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

© 2009 The authors and IOS Press. All rights reserved. No part of this book may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, without prior written permission from the publisher. ISBN 978-1-60750-047-6 Library of Congress Control Number: 2009932691 Publisher IOS Press BV Nieuwe Hemweg 6B 1013 BG Amsterdam Netherlands fax: +31 20 687 0019 e-mail: [email protected]

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

Distributor in the UK and Ireland Gazelle Books Services Ltd. White Cross Mills Hightown Lancaster LA1 4XS United Kingdom fax: +44 1524 63232 e-mail: [email protected]

Distributor in the USA and Canada IOS Press, Inc. 4502 Rachael Manor Drive Fairfax, VA 22032 USA fax: +1 703 323 3668 e-mail: [email protected]

LEGAL NOTICE The publisher is not responsible for the use which might be made of the following information. PRINTED IN THE NETHERLANDS

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

Formal Ontologies Meet Industry R. Ferrario and A. Oltramari (Eds.) IOS Press, 2009 © 2009 The authors and IOS Press. All rights reserved.

v

Proceedings of FOMI 2009 Preface

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

R. FERRARIO and A. OLTRAMARI

From raw data of experience to linguistic forms, from cognitively structured conceptualizations to valuable social assets, the interest in the different nuances of the concept of ‘knowledge’ have characterized natural and social sciences since the dawn of human thought. Across Ages, several theorizations have been put forward and disputes contended around the theme of the multi shaped character of knowledge: in the last years, a new fervour has concerned the production and exchange of knowledge in the business domain. In this context “knowledge contents” have often been flattened to information encoded into computer systems, enabling activities like storage and retrieval through relational database structures, machine-tractability through dedicated computational languages and algorithm-based elaboration. This general phenomenon reveals an underlying conflation between “knowledge assets” possessed-by and transferred-through human resources of firms and “information contents” embedded into industrial information systems. In this picture, Knowledge Management (KM) methodologies and applications have been largely biased by Knowledge Engineering (KE) solutions, stressing the level of formal representation and computability but discarding the centrality of the cognitive construction of knowledge operated by human agents over information flows. Although the intersection between KE and KM has led to a general improvement of information systems in terms of massive data analysis and maintenance, decision-making strategies, information retrieval and exchange, etc., the overlap between these two fields have contributed to eclipse the interest on genuine knowledge processes (production, sharing and transfer of new knowledge) and on how to manage them. It is nowadays widely agreed that the semantic dimension of information plays an increasingly central role in a networked knowledge-centred economy: semanticbased applications aim to provide a framework for information sharing, reliable information exchange, enabling negotiation and coordination between distinct organizations or among members of the same organization. As testified by research and industrial projects, ‘Semantic Technologies’ are bound to a life-cycle constituted by acquisition, retrieval, modelling, reuse, publishing and maintenance of knowledge.1 These different constituent elements draw a new route in the information space along which hybrid agents (human and artificial) negotiate their “knowledge contents”. In particular, the ontological layer has been recognized as fundamental in communication: besides the protocol layer, where the syntax of the com1

In Computer Science, this particular notion of ‘knowledge’ would correspond to T-box (terminological) statements in a knowledge base (i.e. student is a subclass of person). The assertion component (A-box), namely factual knowledge associated to terminology (i.e. John is a person), is not so central in the present context. Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

vi

munication languages is specified,2 the ontological layer formally defines the concepts needed to exchange messages, in particular the semantics of those concepts. Historically framed in the philosophical tradition and “imported” into the KE framework in the late 90’s, “Formal Ontology” (as a discipline) has recently contributed to broaden the scope of inquiry into the notion of knowledge, exploring interdisciplinary areas like Cognitive Science, Social Sciences, Biology, etc. Nevertheless, the problem of managing such knowledge, especially in the business domain and industrial scenarios, has only been skimmed by Formal Ontology. In one sense, if formalization of the ontological categories/concepts is a task per se and, as such, it fosters knowledge sharing/transfer, it is also true that sharing doesn’t come “for free”, and efforts need to be made to 1) understand the role of formal ontologies in the production of new knowledge; 2) turn formal ontologies into effective tools for a new generation of KM software; 3) develop standard formal ontologies for different industrial domains. These remarks constitute the main reason why we chose to co-locate the fourth edition of the “Formal Ontologies Meet Industry” workshop (FOMI) 3 with the 10th European Conference of Knowledge Management (ECKM): with the explicit aim of bringing together KR and KM (sub)-communities, thus providing a common framework for discussion of the above-mentioned topics. The invited contribution by Guido Vetere (IBM Center of Advanced Studies), “From Data to Knowledge, the Role of Formal Ontology”, has to be seen, in this context, both as a general introduction to a FOMI area of interest and as a paradigmatic case where innovative research meets industrial needs. New studies and applications in the field of Ontology and Semantic Technology will be presented, gathering together international specialists from university, research centres, industry. In fact the contributions presented under the scope of FOMI show a variety from both the methodological as well as the stylistic viewpoint. Some of these are more theoretically oriented and are especially concerned with genuine ontological analysis. The paper by Hamdani and Gargouri, with the title “Towards an approach for evolving information systems’ ontologies” is focused on the evolution of ontologies in accordance with the evolution of the domain they are designed for. They build an approach based on operators of change that is aimed at keeping track of the evolution of the information system ontology design and at the same time keeping the consistency and the coherence with the domain through the whole life cycle of the system. In some other papers ontological analysis is still the main concern, but it is carried out within the study of a particular domain. “Parts, Compositions and Decompositions of Functions in Engineering Ontologies” by Vermaas analyzes the relations of functional composition and decomposition and compares them with the part-whole relation in mereology in order to understand whether these can be seen as a sort of part-whole relation specific for functions of technical artefacts. A clearer understanding of these relations is particularly important to improve engineering reasoning in systems that deal with functions, like CAD-CAM systems or engineering knowledge bases. A work focused on design and in particular on architectural design is the one by Hois, Bhatt and Kutz, “Modular Ontologies for Architectural Design”. Their claim is that, given the heterogeneity of the information of the specific architectural domain 2

KQML (Knowledge Query and Manipulation Language) is one of the most well known language protocols in the field, together with FIPA ACL (Foundation for Intelligent Physical Agents – Agent Communication Language). 3 Past editions: 2005 (Verona), 2006 (Trento), 2008 (Torino). Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

vii

(quantitative spatial constraints, qualitative relations, functionally-dependent conceptualizations), a modular ontological approach is the one best suited to the integration of the different but equally important perspectives given by this varied information. Their approach is based on the theory of ε-connections and is aimed at being applied to smart office environments. Another interesting domain is that of electromagnetics, which the paper by Esposito, Tarricone, Vallone and Zappatore, “Towards an Ontology Infrastructure for Electromagnetics” deals with. An interesting feature of the ontological framework they propose is that it is based on a publicly available top level ontology and the choice has been determined by the will of providing a modular and layered architecture that should enable knowledge sharing and reusability for the electromagnetics scientific community. “Do you still want to vote for your favourite politician? Ask Ontobella!”, the article by Garbacz, Lechniak, Kulicki and Trypuz presents an interesting approach based on the philosophy of Roman Ingarden to the ontology of beliefs, a domain that, despite its relevance for artificial intelligence, has not so far received the attention it deserves from scholars in the ontology community. The paper also contains an interesting preliminary formalization and an application to an example taken from political debates. In contrast, some works are concerned with the application of ontology-based methodologies and techniques to a particular knowledge management related issue. The contribution of Hadj Tayeb and Noureddine with the title “Ontological representation for Algerian enterprise modelling” has the purpose of representing – through an ontological framework – a variety of enterprise modelling techniques, distinguished by a list of criteria identified from a theoretical study and the analysis of real case studies relative to Algerian companies. The ontological framework is represented with Protégé. Borgo and Pozza’s paper, “Disentangling Knowledge Objects” is concerned with building a KM framework based on ontological techniques. In particular, a new notion is introduced in the paper, that of knowledge object, comprising three perspectives: material, informational and organizational, proposed as a key concept for enterprise modelling and KM in general. In addition, some articles provide a contribution in terms of the application of formal ontologies to specific industry domains. In “A First-Order Cutting Process Ontology for Sheet Metal Parts”, Grüninger and Delaval build an ontology in first order logic that extends the ontology of the ISO 18629 (Process Specification Language) to support the representation of cutting processes in 2D (two dimensions) to be applied to the domain of manufacturing sheet metal parts. The work by Grenon and De Francisco, “Ontology-strength Industry Standards” deals with the telecommunication domain under the scope of the European project SUPER, where standards, technology neutral architectures, best practices and guidelines are collected in a framework called NGOSS (New Generation of Operation Support Systems). The paper shows that, through an ontologization of standards, it is not only possible to represent concepts in a shared vocabulary like XML, but it also contributes to the enhancement of standards development, dissemination and operationalization. Finally, Corsar, Moss, Sleeman and Sim, in their article “Supporting the Development of Medical Ontologies” present an approach to biomedicine based on the construction of small domain ontologies, tailored to some very specific tasks, that are very efficient from the point of view of tractability and inferencing. These ontologies are

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

viii

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

then integrated with an alignment meta-ontology, enabling interoperability with other standard medical knowledge sources. As this overview suggests, the double focus on methodological and applicative issues represents the main feature of FOMI 2009 articles, confirming the leitmotiv of the past years’ editions: ontologies and ontology-driven methodologies are not simply considered as “closed” systems but as dynamic modules embedded in knowledge technologies. We think that this comprehensive perspective can advance progress towards new frontiers in information systems and knowledge management, where research and development in Formal Ontology plays a leading role.

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

ix

Contents Proceedings of FOMI 2009 – Preface R. Ferrario and A. Oltramari

v

From Data to Knowledge, the Role of Formal Ontology Guido Vetere

1

Towards an Approach for Evolving Information Systems’ Ontologies Hanêne Hamdani, Mohamed Mhiri and Faiez Gargouri

10

A First-Order Cutting Process Ontology for Sheet Metal Parts Michael Grüninger and Arnaud Delaval

22

Parts, Compositions and Decompositions of Functions in Engineering Ontologies Pieter E. Vermaas

34

Towards an Ontology Infrastructure for Electromagnetism 46 Alessandra Esposito, Luciano Tarricone, Laura Vallone and Marco Zappatore Ontological Representation for Algerian Enterprise Modeling Sabria Hadj Tayeb and Myriam Noureddine

58

Modular Ontologies for Architectural Design Joana Hois, Mehul Bhatt and Oliver Kutz

66

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

Ontology-Strength Industry Standards – The Case of the Telecommunication Domain Pierre Grenon and David de Francisco Disentangling Knowledge Objects Stefano Borgo and Giandomenico Pozza

78 90

Do You Still Want to Vote for Your Favorite Politician? Ask Ontobella! Pawel Garbacz, Marek Lechniak, Piotr Kulicki and Robert Trypuz

102

Supporting the Development of Medical Ontologies David Corsar, Laura Moss, Derek Sleeman and Malcolm Sim

114

Subject Index

127

Author Index

129

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

This page intentionally left blank

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

Formal Ontologies Meet Industry R. Ferrario and A. Oltramari (Eds.) IOS Press, 2009 © 2009 The authors and IOS Press. All rights reserved. doi:10.3233/978-1-60750-047-6-1

1

From Data to Knowledge, the Role of Formal Ontology Guido VETERE IBM Center for Advanced Studies of Rome, Via Sciangai 54, 00153 Rome, Italy [email protected] Abstract. The increasing availability of large amounts of data and the growing capability of accessing and processing them, gives us today unprecedented opportunities to advance in many fields, including science, commerce, social relations, government, and business, through information technologies. However, in order to have computing machines supporting this progress, data must be turned into processable knowledge. This “epistemic ascent” cannot be driven by data themselves, as some technologist suggests, but requires hypotheses, theories, and models. Formal ontology is part of the theoretical framework that technology needs in order to get computing systems working with networked data in a consistent way. Formal ontology, however, roots in still open philosophical hypotheses. Yet, only formal ontology has the key of notions such as parthood or dependence, that are relevant when discovering knowledge into data. This paper wants to argue that, regardless of philosophy, formal ontology can already provide computer science with many useful tools. On the other hand, without stepping into metaphysics, business communities are on the way of using commonsense notions with more formal consciousness. Making more clear the role of formal ontology can help this process and boost progress towards new frontiers in information management.

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

Introduction Digital information available over the Internet is growing at the impressive rate of thousands of petabyes per year. This information includes authored items publishes by millions of people, databases, digital records flowing from intelligent devices, sensors and a vast array of instrumented objects. The technology has moved forward on the way of capturing and processing huge data streams, in a sophisticated and efficient way. Entity Analytics systems can identify persons, groups, organizations, and their relationships. Business Intelligence systems can capture and analyse changes in markets, trends and consumer preferences better and faster than ever. Information Integration infrastructures allow accessing heterogeneous and distributed data through conceptual models of business entities. In sum, many kinds of systems based on algorithmic methods are showing the ability of detecting facts and predicting events based on the unprecedented availability of networked information, that is gathered and sorted more and more efficiently by an ever growing amount of computational resources. However, much of the data flowing into the Internet today consists in unstructured content, that is inherently difficult to be dealt with. Processing the relatively little amount structured data is not easy either, since it is produced on the basis of many different, idiosyncratic, and often very tangled schemas. Nevertheless, the impressive amount of data available, along with progresses in information management technologies, has given birth to a new kind of positivism, that tends to demote the role of theo-

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

2

G. Vetere / From Data to Knowledge, the Role of Formal Ontology

retical constructions in favour of the mere access to empirical evidence, through data. The famous editorial of Chris Anderson [1] on Wired last year has fully represented this point of view. Models, including ontologies, tend to be wrong, Anderson says, but the good new is that we don’t need them any longer, since all the truth is now over the Internet, within the reach of our search engines. Many of the reactions to this new brand of positivism have started from classic western epistemology, to observe that data cannot upgrade to knowledge just by virtue of their amount. Instead, dealing with digital information with the purpose of carving out useful knowledge, requires specific representation models, whose relevance increases at the same pace of the data volumes growth. At his time, Karl Popper objected to logical positivists [2] that there is no such a thing as the number of positive evidences that can make a scientific hypothesis true. A fortiori, there is not such a thing as the number of positive evidences that can guarantee the construction of scientific hypotheses. For a similar reason, there’s not the number of occurrences that can make an explanation of what data patterns mean: the kind of things we are in seek of must be known in advance, as well as the way data and entities may be related. To paraphrase Kant, information without models is empty, models without information are blind.1 Epistemology apart, it is easy to see how models play a key role in information technologies today, and how they will increase their role in the coming years, exactly for the purpose of coping with the huge and fast growing volume of networked information available. Current practices of large-scale information management for enterprises and business organizations revolve around developing and sharing conceptual artefacts such as business vocabularies, taxonomies, industry models, that altogether amount at a corpus of a priori knowledge about business notions, such as processes, events, abstractions, qualities, and individuals, including their kinds and their relationships. At the scale of the Web, that is a totally decentralised and almost totally free federation of independent information systems, the vision of a pervasive integration based on shared ontologies, although by far less easy than Berners-Lee had originally envisioned [3], is still concretely pursued. Countless initiatives for the standardisation of business vocabularies, XML schemas, and metadata are underway. The evolution of the Web is seen, among other ways, as the capability of representing and processing meaningful data structures [4]. The idea of linking data items across the boundaries of each single site, however, wouldn’t make any sense without the prospect of a global semantic convergence of the information growing on the Web every day. All these facts show that our Information Technology is firmly bound to an hypotheticodeductive framework, where, like in science, theories, in vest of models, play the crucial role of driving the process of organising the continuum of available information into an intelligible space. The investigation method at the basis of sciences consists in formulating conjectures about the real world, in terms of the existence of certain types of objects and relationships, the subsistence of certain states of affairs, and so on. Such hypotheses are intuitive propositions often based on abductive reasoning [5]. When a set of promising hypotheses reach a certain degree of formality, e.g. by means of mathematics, is internally coherent, and is compatible with what is known, it may become (part of) a theory. By reasoning on theories, one can explain observed facts and predict unobserved ones. However, no matter how many facts are explained or predicted, theories are neither ‘false’ nor ‘true’; instead, they must be falsifiable, and until they are not falsified they can be considered empirically adequate. 1

Thoughts without content are empty, intuitions without concepts are blind (Critique of Pure Reason, 1787). Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

G. Vetere / From Data to Knowledge, the Role of Formal Ontology

3

Figure 1. The scientific method.

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

Figure 2. Information systems.

The way we design information systems today has something in common with the framework described above. Realities (i.e. business entities) are captured by models, under specific (albeit mostly tacit) ontological assumptions. Then, with a certain degree of automatism, models are turned into computable resources, that allow automatic reasoning on both models and data to achieve knowledge and rational behaviours. Models can be regarded to as applied theories, whose adequacy is given by systems effectiveness. In information systems, like in natural sciences, the bridge that joins models and knowledge on one side, with reality and data on the other side, is build on the two pillars of theorisation and verification. For information science, theorisation consists in modelling, i.e. drawing classes, relationships, and constraints, by means of some formal or semi-formal framework. Verifying knowledge items with respect to data, on the other hand, depends on establishing how data should be deployed in order to make known propositions true, or, conversely, finding out which knowledge items are compatible with existing data patterns. According to the tradition of logics [6], providing this function is the specific role of semantics. In ontology-based data integration, for instance, the theory that models the application domain (global view) is mapped to source database schemas by means of correspondence rules, which is where semantics dwells [7]. Within the paradigm of information sciences, models are not only useful; indeed, they are needed and play a central role, like theories in natural sciences. The question is then whether models in information sciences need to be ontologies, or not.

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

4

G. Vetere / From Data to Knowledge, the Role of Formal Ontology

Figure 3. An ER model.

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

Models, Vocabularies, and Ontologies In information systems design, models are usually intended to capture domain entities and relationships, to drive software development and serve as run-time resources. Nevertheless, these models do not necessarily contain ontological commitments. Models, in themselves, are at the level of abstraction of logics. Like predicate symbols in a logic theory, classes and relations are meaningless, unless they are mapped with something outside logics, i.e. a “domain”. This usually goes implicitly, by using natural language as a “semantic glue”. Simply, classes and relationships are labelled with lexical items, whose linguistic meaning should intuitively give the user a grasp of the intended interpretation of the model with respect to the application domain. Nevertheless, nothing existing in a model, including its linguistic labels, implies that something really exists. Models can posit non existing things along with everyday life entities, like in video games, or mix different ontological layers, e.g. technical and business metadata, as in many hastily developed applications. Also, models can miss relevant notions just because they are not mentioned in application requirements. Conceptual modelling in information systems development consists in analyzing functional requirements to produce a representation of entities and relationships involved. Note that, in this context, the correspondence of analytical entities and relationships to classes of existing things is not a requirement. Actually, the ultimate goal of conceptual modelling is to drive the production of database schemas, application code, and other technical resources. In turn, the purpose of these technical resources is to make systems working: application code must run, database schemas must accommodate data in a suitable way, procedures must support workflows, and so on. Developers are focussed on applications, rather than the “real world”, whatever it is. Thus, it not surprising that conceptual models are very often concerned with applicative problems, rather than with theories about what there is in the world. On the other hand, even if it is generally true that conceptualization aim at modelling real situations, so that many conceptual models can be reasonably called “ontologies”, the adjective “conceptual” cannot set a model to be a model of the reality. In fact, unless a sort of naive realism is embraced, nothing, in the process of conceptualizing, ensures that the real world is reflected in the model. In this light, the classic definition of ontology (in information sci-

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

G. Vetere / From Data to Knowledge, the Role of Formal Ontology

5

ences) as the “specification of a conceptualization” [8] has been criticized as too generic. Generally speaking, business vocabularies, or glossaries, are committed to real world entities through natural language. In fact, their main purpose is to elicit and organize business concepts, and to tie them to technical and informational resources. Since they start from business concepts, instead of application needs, vocabularies are usually deemed as carrier of ontological commitments, in the extent in which natural language is committed to reality. Thus, the ontological import of business vocabularies is bound to linguistic practices, whose relation with reality, however, is generally in question.2 Whether linguistic realism holds or not, relationships between words and entities are far from being linear; vagueness and polysemy are the foremost evidence of that. Take the noun ‘part’ for instance: it enfolds a number of very different relationships, from mereological to merely logical ones. Broadly speaking, the ontology of vocabulary entries lies in the background knowledge of business communities that use them, and emerges within specific situations and contexts. In the process of creating business vocabularies, linguistic habits are often mistaken for ontology, since the aim of information designers is to reach stability in human-computer interactions, rather than “deconstructing” linguistic practices. On the other hand, when writing vocabularies, information designers cannot dive into complex socio-linguistic analyses. Ontologies are not (and should not be regarded to as) another, more sophisticated and maybe esoteric way to draw conceptual models or business vocabularies. There’s a radical change of perspective in developing an ontology, rather than a conceptual schema, or a glossary. Ontological analysis focuses on what exists in the application domain, independently of application requirements and linguistic practices. Technicalities and linguistic renderings, if needed, should become part of subsequent modelling activities. Therefore, ontologies are not just conceptual models specified in some rich logic language; instead, they are models in which a definite correspondence between representational items and real entities is set. Once again, neither conceptual modelling in software engineering, nor linguistic analysis of business vocabularies, necessarily embrace this perspective. As Smith and Welty have pointed out [9], ontologies aim at modelling the essence of what there is in the world where applications run, the fundamental properties of real entities that every agent in the digital ecosystem is supposed to be acquainted to. This essence should be invariable with respect to application environments, practical requirements, linguistic practices, individual biases, and whatever. What really differentiate ontologies form other kind of models is therefore the specific perspective they adopt. Ontological properties, if consistently recognized by every agent, provide semantic foundations, thus allowing meaningful communication to take place. When set at the core of information systems, ontologies serve as a basis for interoperability, cooperation, information integration and exchange, and many other applications. Ontologies, and only ontologies, can give the process of acquiring knowledge from data a suitable semantic grounding. Through semantic mapping, data are put into relation with knowledge items, that can acquire an epistemic import only within a framework where specific hypotheses about reality are set. As long as knowledge is knowledge of the reality, and not of other things, ontology is therefore needed, and, conversely, any other modelling whose purpose is not that of explaining the world, would not be relevant for knowledge acquisition.

2

According to many linguists and philosophers, the language does not reflects the reality, but the other way around. Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

6

G. Vetere / From Data to Knowledge, the Role of Formal Ontology

Figure 4. Different ontologies for ‘husband’.

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

Formal Ontology Understanding the benefit of ontological analysis is straightforward, but ontological analysis is not. While conceptual modelling requires, for the most, some suitable language, ontological analysis, besides any formal language, requires a theory of what exists. In the history of western philosophy, this theory has been largely debated, and many issues are still discussed today. For instance, are there “social objects” such as rights, obligations, privileges, entitlements, and permissions? And do they have the same ontological status of mountains, roads, sunsets, and flowers? [10] So the question is: how information science could concretely use a framework that, to say the less, is still under construction? The answer may be sought in the world of modern web information systems, where the motto is: “beta first”. Today, developers are used to release early versions of their systems, and then refine them based on users’ reactions. Such a pragmatic approach may be someway applied to the use of ontologies, whose theoretical framework is still under development. After all, information science doesn’t need to solve all philosophical problems, before it can take advantage of what comes out from our tradition of thought, nor is interested in taking position with respect to foundational problems. Given an ontological individuation, information science is interested in its practical import, rather than its philosophical consequences. Take for instance “quaindividuals” [11]. Roughly, they are “parasitic” individuals that depend on some “actual” individual to convey sets of properties which, in turn, depend on certain specific relations. “Guido-qua-husband” would designate a sort of counterpart of Guido, who has the property of being married. Whether these kind of instances really exist or not (it would be nice if there were) may have a deep impact on the way ontologies are developed (Fig. 4). However, information systems designers can overlook the whole ontological question and adopt either one approach or another, in a relativistic way. Of course, rather than promoting semantic standardization, this “ontological relativism” would result in a proliferation of many different semantics, which is, by the way, what is happening within business communities at the time being. In this scenario, mapping different ontologies the one another has the form of a “radical translation” [12] in which arbitrary assumptions about correspondences among different ontological models are taken. In any case, any coherent ontology provides information engineers with a consistent theory of what is supposed to exists, be it spatio-temporal entities, abstractions, or social objects. What really matters to information system developers is the availability of models committed to something whose existence, by hypotheses, is supposed to be acknowledged by the whole community of users. The benefit of sharing an ontology, whatever it means and how effective it is, primarily consists in sharing a commitment to reality, not a commitment whatsoever. The quality of these commitments has to be seen in terms of empirical adequacy, that is, in terms of their effective usability. Guarino has pointed out that ontologies serve to “re-

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

G. Vetere / From Data to Knowledge, the Role of Formal Ontology

7

strict the interpretation of a logical theory on the basis of [...] a priori distinctions among the entities of the world ” [13]. Restricting the interpretation of predicates of a descriptive theory amounts at introducing “meaning axioms”, i.e. binding these predicates with basic categories whose interpretation can be taken for granted. We know that, in principle, interpretation is free and, even if they were sharing the very same ontology, there would not be any guarantee that, given two agents in a digital ecosystem, they would always interpret predicate symbols in the same way. This uncertainty is exactly where formal ontology unveils its value. The more compelling meaning axioms are, the more is likely that agents will eventually converge on the very same interpretation of predicate symbols. Of course, to make meaning axioms compelling, one needs compelling ontological categories. There are many sources from where ontological categories can be drawn, from metaphysics to cognitive sciences and linguistics. Analyzing these sources would require a detailed discussion, which is out of the scope of this paper. But it is important to remark that, wherever they come from, ontological categories are mere hypotheses, whose only strength is in their pragmatic import [14]. If ontology (in the sense in which this term is used in information science) can be ultimately regarded to as a “scientific image of man-in-the-world” [15], then its epistemic adequacy is a matter of concrete usefulness. Similarly, to industry, formal ontology is relevant insofar as it increases the effectiveness of conceptual models used in information extraction, integration, and exchange.

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

Knowledge and Ontology Mining knowledge out of the huge mass of data available on information systems today starts from interpreting data patterns as if they were propositions regarding entities in the world. Then, since we are interested in knowing generalities, rather than bunches of singular unrelated facts, we base propositions on ontologies that provide us with categories and relations. If we look at this process, we can see that it has much in common with the method of scientific investigations. In fact, ontologies in information management play the same role as theories in natural sciences, with the difference that the main objective of scientific investigation is extending, refining or even rebuilding theories to accommodate the observation of facts, while information management aims at collecting and classifying observations of facts, based on available ontological theories. Without ontologies, knowledge would be an heap of unrelated propositions, documents would be meaningless sequences of statements, databases would be stand-alone information silos, services wouldn’t be interoperable if not at a merely syntactic level. Whenever our systems succeed in understanding documents, integrating databases, or orchestrate services, there is a reality bound reasoning process behind, based on ontologies. What makes the process of sharing meaningful knowledge possible, is the ontological commitment encoded in software models and technical resources. Information Technology industry is increasingly aware of the need of keeping these commitments in some specific place of the architecture, rather than everywhere in each single system, and powerful ontology representation languages, along with a vast array of tools, are available nowadays to serve this purpose. With this technical and architectural setting in place, one of the most relevant challenges we can expect to face in the next future is related to the quality of ontological resources. It is now important for industry to raise the quality of available ontologies, which implies understanding what “quality of ontologies” means.

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

8

G. Vetere / From Data to Knowledge, the Role of Formal Ontology

A pragmatic approach to the enhancement of ontological resources could be quite agnostic with respect to philosophical issues, but this wouldn’t be a good reason for throwing away useful notions that we can borrow from philosophy, in particular from contemporary analytical thought. Some of these notions has been elaborated to provide meta-level formal quality criteria for ontologies [16]. As for categories of particulars, the philosophical tradition provides substantial distinctions based on primitive intuitions such as space-time dimensions and part-whole relationships [17]. Whether these intuitions can be arranged in a consistent theory to produce concrete results such as a single reference top-level ontology and a widespread ontology development methodology, is something that we will see in the coming years. As a matter of facts, there is an increasing demand of standardized ontological resources, and people currently engaged in developing models of the real world would greatly benefit of any preliminary version of any working ontological machinery. Currently, global standardization initiatives, such as those related to the Ontology Web Language, are limited to logics and syntaxes, while content-level standardization is scattered among a vast array of many different industry bodies. Time has come to foster more coordination in the global ongoing process of theorizing about business realities. It is true that none of the foundational ontologies developed from the time of early Artificial Intelligence up to now has never gained the status of reference standard [18]. However, this cannot be considered a proof that the program of identifying basic ontological categories to enhance business applications is unfeasible and must be abandoned. On the contrary, the inherent difficulty of the task should drive more effort in learning all the lessons we have took so far and defining the agenda of further research.

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

Conclusion The ultimate goal of information management is to help people and enterprises to acquire, manage, and exploit valuable knowledge. Progresses in computational and networking capabilities make available today a terrific amount of accessible data. For information systems to benefit from this “data deluge”, rather than being just flooded, along with technologies, new paradigms and methods are needed. A key aspect of this progress consists in the capability, for information systems, of extracting knowledge from data. As long as knowledge is knowledge of the real world, this “epistemic ascent” requires hypotheses about “what there is”. In other words, ontology is a bridge between data and knowledge of the real world, and this bridge is needed to reach the ultimate goal of any information system. Ontology must be regarded to as a theoretical framework that enables computing systems to work with networked data in a consistent and interoperable way. Formal notions such as “parthood” or “dependence” are of key relevance when discovering knowledge into data or sharing information in a meaningful way. Even though formal ontology roots philosophical hypotheses that are still debated, this discipline can already provide computer science with many useful tools. Taken pragmatically, different metaphysical options can be evaluated with respect to their concrete adequacy. At the same time, without stepping into metaphysics, business communities are on the way of being more conscious about the need of eliciting their knowledge in a consistent way. Making more definite the role of formal ontology, and more clear the need of adopting theories of real world entities, rather than uncommitted conceptual models, can help this process and boost progress towards new frontiers in information management.

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

G. Vetere / From Data to Knowledge, the Role of Formal Ontology

9

References

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

[1] C. Anderson, The End of Theory: The Data Deluge Makes the Scientific Method Obsolete, Wired 23/06/98. [2] K. Popper, The Logic of Scientific Discovery (Routledge, 1959). [3] T. Berners-Lee, Semantic Web Road map, 1998. http://www.w3.org/DesignIssues/Semantic.html. [4] Linked Data – Connect Distributed Data across the Web. http://linkeddata.org/. [5] C.S. Peirce, Collected Papers of Charles Sanders Peirce, 8 vols., Charles Hartshorne and Paul Weiss, eds. (Cambridge, MA: Harvard University Press, 1958). [6] A. Tarski, 1944. The Semantic Conception of Truth and the Foundations of Semantics. Philosophy and Phenomenological Research 4. [7] M. Lenzerini, Data Integration: A Theoretical Perspective, proceedings of PODS 2002: 233-246. [8] Gruber, A Translation Approach to Portable Ontology Specifications, Academic Press 1993. [9] B. Smith, C.Welty, Ontology: Towards a New Synthesis, proceedings of FOIS’01, 2001. [10] John R. Searle, The Construction of Social Reality, New YorkFree Press, 1995. [11] C. Masolo et al., Relational Roles and Qua Individuals, proceedings of the AAAI Fall Symposium on Roles, an interdisciplinary perspective, 2005. [12] W.v.O. Quine, Ontological Relativity and Other Essays. Columbia Univ. Press, 1969. [13] N. Guarino, The Ontological Level, In R. Casati, B. Smith and G. White (eds.), Philosophy and the Cognitive Science. Holder-Pivhler-Tempsky, 1994. [14] C.S. Peirce, Issues of Pragmaticism, The Monist, vol. 15 1905, pp. 481-499. [15] W. Sellars, Philosophy and the Scientific Image of Man, in Frontiers of Science and Philosophy, University of Pittsburgh Press, 1962: 35-78. [16] N. Guarino and C. Welty. Evaluating Ontological Decisions with OntoClean. Communications of the ACM. 45(2):61-65. New York:ACM Press. 2002. [17] A. Gangemi et al., Sweetening Ontologies with DOLCE, In A. Gómez-Pérez, V.R. Benjamins (eds.) Knowledge Engineering and Knowledge Management. Ontologies and the Semantic Web, 13th International Conference, EKAW 2002, Siguenza, Spain, October 1-4, 2002, Springer Verlag. [18] B. Smith, Ontology, in L. Floridi (ed.) The Blackwell Guide to the Philosophy of Computing and Information, Blackwell Publishing, 2003.

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

10

Formal Ontologies Meet Industry R. Ferrario and A. Oltramari (Eds.) IOS Press, 2009 © 2009 The authors and IOS Press. All rights reserved. doi:10.3233/978-1-60750-047-6-10

Towards an approach for evolving information systems’ontologies Hanêne HAMDANIa , Mohamed MHIRIa and Faiez GARGOURIa MIRACL Laboratory, ISIM Institute, BP 1030-3018,Sfax. Tunisia { hanen.hamdani, med.mhiri, faiez.gargouri}@gmail.com

a

Abstract. Information system’s ontology is not frozen: it must evolve in the course of its life cycle to keep its consistency and its coherence in dynamic and multidisciplinary domains. However, any change to be taken care at the level of Information System (IS) design ontology is a wonderful source of new structural and semantic disconnectedness, which must be perfected to keep the ontology operational and consistent. In this paper, we define an approach for the recognition of the IS design ontology evolution. Change operators are defined for this goal. Besides, strategies of evolution are defined to manage the consequences of these changes. Key words. IS design ontology, elementary change, complex change, operator of change, consistency and coherence.

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

Introduction A lot of research works were accomplished for the ontologies’ building. The very quick multiplication of the building methodologies is parallelized with the multiplication of their application domains. However, the dynamic character of the ontologies’ domains requires the evolution’s management of the built ontologies and the definition of methodological points for this purpose. In our case, the IS design ontology makes a very strong coupling with the IS design [5]. It is based on a group of conceptual diagrams modeling a well determined domain. The building step gave birth to the definition of a semantic relations’ set (let us note that by semantic relation, we imply relations others than those offered by UML: Equivalence, Synonymy, etc) [3]. However, the IS ontology’s domain is dynamic; it necessitates the shedding light on the ontology’s evolution. This step of the ontology’s life cycle should not be overstepped to guarantee the IS ontology coherence with new use contexts in the domain. Currently, the approaches treating the ontologies’ evolution are not consensual to take care of all the IS design ontology’s specificities. It is for it, we define a method that allows to remedy the insufficiency of the existent approaches. This article is structured as follows: In section 1, we introduce a synthesis of the approach which we define for the IS design ontology evolution. In the sections that follow, we introduce this approach in detail: Section 2 describes the change need detection step. Section 3 is interested in the operators’ choice step. The change operators’ execution step is the objective of section 4. Section 5 introduces the

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

H. Hamdani et al. / Towards an Approach for Evolving Information Systems’ Ontologies

11

changes’ validation. Finally, we synthesize the different points of the defined approach and we introduce our future works.

1. Our approach for evolving an IS design ontology

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

Our approach draws upon some ideas from the developed approaches of data base schema evolution [1] and ontologies’s evolution [7], [8]. However, we take care of the IS design ontology’s specificities to know the management of the changes’ consequences, the semantic level during the change process and the complex changes’ effects. Our IS design ontology’s evolution method consists in evolving the ontology at its conceptual level; we are not interested in its instances. This method is composed of four steps such as: (1) change need detection, (2) change operators’ choice, (3) change operators’ execution and (4) changes’ validation. These four stages described in Figure 1 do not form a linear process due to the fact that the three steps of the change operators’ choice, the operators’ execution and the changes’ validation are applied in a cyclic way. A change validation can then cause the detection of a new change need. Also, a change operator execution can call several other operators to manage its consequences.

Figure 1. Evolution cycle for an IS system design ontology

2. Detection of a change need The approach for an IS design ontology building was conceived in previous works in four stages [4]: building the initial ontology, feeding the ontology, representing the ontology with UMLONTO [6] and updating the ontology. The IS ontology’s building was proposed due to an urgent need for the designer to correct syntactic, structural and semantic errors that can occur when accomplishing the IS design task [3]. The accomplishment of an IS ontology evolution begins with the changes’ needs which confront an already existent and operational ontology. Given that the ontology must remain operational during its life cycle, modifications are sometimes necessary to adapt it to the new functionalities existing in its domain. On the basis of the definition offered in [2] describing ontology as «an explicit specification of a domain’s conceptualization», changes to be brought in ontology can

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

12

H. Hamdani et al. / Towards an Approach for Evolving Information Systems’ Ontologies

then be due to changes in the domain, changes in the conceptualization and changes in the specification. In reply to these change needs, several evolution actions can be followed. That’s why, we offer the following classification for the IS design ontology change needs: The functional needs: A functional need can be a corrective need for the identification and correction of both conception and realization errors, or an adaptive need allowing its adaptation to all domain changes. The ontology refinement needs: These needs appear when utilizing a group of changes to improve the ontology structure. The operations of Merge, splitting and generalization which we will describe in the following sections can be applied to fulfill this refinement. The internal change needs: These needs appear internally in the ontology, thus resulting in a change auto-detection. In reply to these change needs, the IS ontology’s user will be confronted with its adaptation problem to every new functionality. The following section introduces the first step of our contribution. It consists in the definition of a changes’ group devoted to an IS design ontology and the study of these changes’ formalization.

3. Choice of the change operators The main objective to be attained at the level of this step is to solve the change needs launched before. The following sub-sections describe tasks to be fulfilled in the course of this step.

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

3.1. The identification of changes An ontological change is any change in the definition of the items of the ontology’s vocabulary such as the concepts, their properties and the relations between them. The relevance at the level of the changes’ definition, the adequacy of the defined changes and the flexibility of the changes’ execution are the criteria which must be taken when identifying changes for an IS design ontology. We define two categories of changes for an IS design ontology: elementary changes and complex changes. 3.1.1. Elementary changes An elementary change cannot be expressed using other changes. It cannot be divided into several other changes having a finer level of granularity. The elementary ontological changes are classified according to their effects on the ontology structure in three categories such as: Increase changes (Add-concept, Add-relation, Addproperty), Reduction changes (Remove-Concept, Remove-relation, Remove-property) and modification changes (Rename-concept, Rename-relation, Remame-property). The increase, reduction and modification changes are applied on the concepts, their properties (attributes and operations) and the relations between concepts. Only

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

H. Hamdani et al. / Towards an Approach for Evolving Information Systems’ Ontologies

13

simple conceptual relations are targeted by the renaming operation. This consists in the change of the name of a vocabulary’s element without modifying its semantics. Besides, semantic relations are not explicitly targeted by the above named changes. These relations are managed automatically on the basis of a group of rules [4]. Taking care of the semantic aspect assures the auto-evolution of the directed ontology. 3.1.2. Complex changes limiting the IS ontology’s changes in the changes which we have already stated does not guarantee the representation of changes discerned in the most convivial way for the user. Indeed, some change intentions can be more complex in comparison with the simple changes. We propose to extend the simple changes by defining other changes having a higher level of granularity to be able to answer complex change intentions. These changes, which are expressed according to a succession of elementary changes, are said complex changes. The complex changes which we define for an IS design ontology evolution are generalization, splitting and Merge/merging.

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

3.1.2.1. Change of generalization This change consists in grouping the common properties of two, or more, concepts in a single concept that represents their super-concept. Let us consider the example of Figure 2 that represents an extract of an IS design ontology targeting the e-commerce domain. This figure describes certain paying modes such as payment by money, payment by Credit-Card and payment by Debit-Card. Let us assume now that the user wants to generalize both concepts Credit-Card and DebitCard in a more generic concept Card. This latter represents the common direct superconcept of the former two concepts; it will include their common properties. Let us assume, also, that we are going to perform this change on the basis of the elementary changes defined below. The following ordered steps are to fulfill: (1) The creation of a new concept called Card, (2) the definition of the concept CreditCard as being a sub-concept of the concept Card, (3) the deletion of the Is-a relation linking up the concept Credit-Card with the concept Payment, (4) the definition of the concept Debit-Card as being a sub-concept of the concept Card, (5) the deletion of the Is-a relation linking up the concept Debit-Card with the concept Payment and (6) the definition of the concept Card, newly created concept, as being one sub-concept of the concept Payment.

Figure 2. Concepts’ generalization (CC: Class Concept)

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

14

H. Hamdani et al. / Towards an Approach for Evolving Information Systems’ Ontologies

The accomplishment result of all these steps gives the structure of Figure 2. Based only on the simple changes, the concepts’ generalization change is accomplishable. However, we identify several disadvantages during the application of this changes’ sequence. Among these disadvantages, we name (that): • a semantic distance between a change intention and the way of accomplishing it can occur, • a difficulty at the level of the changes’ order definition with possibility of loosing data and leaving redundancies (at the level of properties as well as at the level of relations). For example, in the context where the ontology users make the generalization change in Figure 2, they can forget to delete both IS-a relations linking up concepts Card-Credit and Debit-Card in the concept Payment. Consequently, there will be a redundancy at the level of the properties’ definition of the concept Payment in these two concepts, • the possibility of fulfilling irrelevant and undesirable changes in comparison with what we intend to do, and losing some relevant information existing in the ontology domain and • the disconnectedness of the ontology structure, which can be retrieved from the management of all consequences of elementary changes in an independent way. For example, during the deletion of the Is-a relations between both subconcepts to be generalized and the concept Payment, the users can choose to copy the properties of the concept Payment in Card-Credit and Debit-Card. Besides, if they make the choice of the recreation of an Is-a relation between the concepts Card and Payment, the redundancy at the level of the properties already copied at the level of steps 3 and 5 is then possible. These disadvantages are much more reinforced when we generalize more than two concepts as in picture 2. To remedy all these disadvantages, we offer to regroup all necessary elementary changes to fulfill concepts’ generalization in a unique complex change. Their accomplishment order as well as the management of their consequences is automatically made. This change will have a higher level of granularity and a clear intention. All is about the change of concepts’ generalization. A general strategy to treat such change must then be defined. General strategy for the concepts’ generalization change: Assume that we have n concepts cg1, cg2,…, cgn situated in the middle of a given concepts’ hierarchy, our objective then is to create a concept representing their common super-concept. By operating on the structural level and generalizing the n-concepts, we must consider the following points: Site of the generalization-concept in the ontology hierarchy : Assuming that Cg is the new concept of generalization, the location of Cg in the concepts’ hierarchy is as follows: It directly subsumes source concepts, while common super-concepts of the source concepts (if they exist) subsume it. Building the concept Cg : Consider the following steps: • Creation of a new concept. • Adding the common properties in the concepts cg1, cg2, …, cgn to the created concept (step1), thus undertaking a renaming process of some properties is possible.

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

H. Hamdani et al. / Towards an Approach for Evolving Information Systems’ Ontologies

15

•

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

Removal of the properties immigrated towards super-concept from the source concepts. • Addition of properties (if necessary) in the generalization-concept Cg. • Removal of the relations Is-a linking up the source concepts and their common direct super-concept. • Connection of the concept Cg to the common direct super-concept of the nsource concepts, if available, by an Is-a relation. • Re-connection of the source concepts cg1, cg2, …, cgn to the concept Cg by an Is-a relation. The above mentioned steps must be taken during the execution of the generalization operation. They will only focus on the structure of the concept to be created. Indeed, the concepts’ relations factorization is made according to the use needs. 3.1.2.2. Change of splitting This change consists in subdividing a concept into two, or more, concepts having an equivalent semantic value. Let us assume, in the example of Figure 3, that the users need to subdivide the concept Customer into two other concepts Customer and Address. This subdivision has as objective to make the aggregation of all the properties describing the address in a separate concept. Indeed, this aggregation is very advantageous especially when removing the concept Customer from the ontology. For example, we need to keep the semantic trace of the Address because it could be used for the definition of other concepts such as Supplier, Factory, etc… The newly generated concept Address can have other relations with the ontology concepts. But, obviously it will have a conceptual relation; its type and its name are to be determined by the designer, with the concept Customer. In this example, the necessary steps to fulfill splitting are: (1)The creation of a new concept called Address, (2) the migration of the properties defining the address towards the concept Address. In other words, migration will be made in two sub-steps: (a) Addition of these properties in the concept Address and (b) removal of these properties from the concept Customer, (3) the renaming, if necessary, of the properties of the concept Address and (4) the addition of the necessary relation type that exists between the concepts Customer and Address. In addition to the disadvantages which we have just named in the case of the concepts’ generalization, we point out that the user can: •

forget to remove some properties aggregated in the newly created concept from the initial one and

•

have an ambiguity in the relation type which must be added (step 4). For that, the user must be informed on the relations types that can be used.

General strategy for a concept splitting : To split a concept into other concepts we must resolve the following three problems:

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

16

H. Hamdani et al. / Towards an Approach for Evolving Information Systems’ Ontologies

Figure 3. Splitting the concept « Customer »

When is it possible to apply splitting : At the level of the IS design ontology’s evolution, splitting can be served in several situations such as : Splitting with the intention of a concept’s specialization and this according to the value of an attribute or splitting a concept with the intention of a proprieties set aggregation in a new concept as shown in Figure 3. What are relation types that can exist between the concepts result from splitting: The splitting operation’s execution leads to the creation of new concepts among which relations must be envisaged. Generally, concepts of splitting can be linked up by one of the following relations: (1) A simple conceptual relation whose name is given by the user, (2) a composition conceptual relation, (3) an aggregation conceptual relation (4) and an IS-a conceptual relation (used especially when splitting is made according to the value of an attribute).

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

How to fulfill splitting : The steps to be followed to make a concept splitting are: (1) the creation of the new splitting concepts : cs1, cs2,…., csn , (2) the migration of the properties to be aggregated towards these concepts, to undertake renaming (if necessary), (3) the removal of these properties from the initial concept, (4) the addition of other properties (if necessary) and (5) the creation of the relations in which the new concepts have to participate. 3.1.2.3. Change of merging The merge operation is the opposite operation of the splitting one. It fulfils the semantic unification of n-concepts in the same concept that has an equivalent semantic value. General strategy for the concepts’ merging : The following two problems have to be resolved: Application’s conditions of the merge operation: This operation requires the semantic coherence between the merge concepts. This semantic interrelation concerns the relationship type between the concepts under scope. This type can be of: (1) a simple conceptual relation, (2) a composition conceptual relation, (3) an aggregation conceptual relation, (4) an Is-a conceptual relation. Building the merge-concept Cm : Assuming that cm1, cm2, …, cmn are concepts to be merged, and Cm is the merge concept, two scenarios can come: • If the merge is done in one of the directed concepts, in this case we have: Cm = cm1 ⊕ cm2 ⊕ … ⊕ cmn Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

(1)

17

H. Hamdani et al. / Towards an Approach for Evolving Information Systems’ Ontologies

And so tasks to be fulfilled are: (1) The migration of the properties of cm1, cm2, …, cmn towards the concept Cm, (2) the accomplishment of the renaming process(if necessary) and (3) the removal of the concepts cm1, cm2, …, cmn and their relations, especially those linking them up to Cm. •

If the merge is made in a new concept, tasks to be fulfilled are those described above, but before being accomplished, a new concept Cm have to be created.

In brief, to make the IS design ontology’s changes operational, we propose to formalize them in the next sub-section. 3.2. Definition of the change operators To finalize the previous step, a definition that is easily translatable towards a language comprehensible by the machine must be made. In order to do that, a definite representation of these changes is necessary to define the syntax of every change. For that, we associate to every defined change a change operator. Table 1. List of the IS ontology evolution operators Add-concept (Otcsi, c : C, [ci: C], [cj : C], [Name-relation]) the three last parameters are to define for an association class concept(CCA) [3] Add-property (Otcsi, ci : C, Type- property, Name- property) ; type-property ∈ {attribute, operation} Add-relation (Otcsi, ci : C, cj : C, Type-relation, Name-relation) ; Type-relation

∈ {conceptual, semantic}

Remove-concept (Otcsi, ci : C) Remove-Property (Otcsi, ci : C, Type-property, Name-property)) Remove-relation (Otcsi, ci: C, cj : C, Type-relation, Name-relation) Rename-concept (Otcsi, ci : C, New-Name-Concept) Rename-property(Otcsi, ci : C, Type-property, Name-property, New-Name-Property) Rename-relation (Otcsi, ci:C, cj:C, Type-relation, Name-relation, New-name-relation)

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

Merge-concept (Otcsi, cm1 : C, cm2 : C,…,cmn : C, cm:C) Split-concept (Otcsi, ci : C; cs1 : C, cs2 : C,…,csn : C ) Generalize-concept (Otcsi, cg1 : C, cg2 : C,… cgn : C, cg:C)

With the intention of the formalization of the changes defined in the previous subsection, we offer to define a group of change operators. An operator named Nameoperator, applied to an IS design ontology Otcsi is formalized by the Eq (2). The elements of the ontology are linked to the arguments (Args) of the operator. Name-operator Where Ontology Otisd: C is the concepts set, A is the attributes set of the C elements, O is the operations set of the C elements and R is the ontology’s relations set

(2)

The operators which we offer to bring changes in an IS ontology as well as their parameters are described in Table 1.

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

18

H. Hamdani et al. / Towards an Approach for Evolving Information Systems’ Ontologies

In the case of Table 1 we are especially interested in the ontology’s consistency. Indeed, a simple definition of an operators’ set remains deficient to evolve the ontology and to guarantee its consistency. It is therefore necessary to envisage resolution strategies for the execution of every operator. These strategies consist in fixing rules allowing to assure the structural and semantic consistency of the IS ontology. These two aspects are to detail in the following section.

4. Execution of the change operators When the user is limited to the requested-change application, it is possible to obtain an inconsistent and incoherent ontology. For example, if we take the operator AddConcept (Otcsi, ci: C), the concept ci to be added could have semantic ontological relations with the already existent concepts. The semantic relations cannot be completely managed by the user especially when the ontology size is very big. In this case, we can envisage evolution strategies allowing the resolution, according to the user’s needs, of every change brought in the ontology. Strategies to be defined have to manage the user’s needs, the ontology’s coherence in comparison with the domain and the ontology’s structural and semantic consistency. These aspects require the management of an iterative evolution cycle. Then, we propose to extend the definition of the change operators’ semantics. This extension consists in the prediction and the management of every the change effects on both structural and semantic levels. 4.1. Propagation of ontological changes at the structural level of the IS design ontology Addition and renaming changes do not have structural effects. Their derived changes concern the semantic relations defined for the IS design ontology (synonymy and equivalence).

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

4.1.1. The removal change’s propagation In this section, we are interested in the presentation of changes diverted from the change operators Remove-concept() and Remove-relation(). Changes diverted from the removal of a property do not have effects to be taken care at this level because their removal affects semantic relations. The existence of these relations is conditioned by the check on a group of rules which must be validated on the basis of the concepts’ properties [2]. We itemize the effects of the removal of properties on the semantic aspect in what follows Removal changes allow the retirement of ontological elements from the ontology structure. For example, a concept removal procreates the removal of all its semantics from the ontology. However, a concept’s existence can be necessary for the existence of other concepts. The ontology’s structural change following a concept removal is made according to the relations linking up this concept with its neighborhoods. The relations which must be taken during their removal are the IS-a relation, the composition relation, the equivalence relation and the synonymy relation. The possible effects of a concept removal as well as its relations are described in table 2.

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

H. Hamdani et al. / Towards an Approach for Evolving Information Systems’ Ontologies

19

Table 2. General evolution strategies for the removal of the IS design ontology’s relations

Relation

Problem

Evolution strategies(S)

Management of the orphans concepts

sH1. Removal of the orphans concepts. sH2. Re-connection of the orphans concepts in the direct super-concept of the removed concept sH3. Removal of the properties of the concept to be removed. sH4. Propagation of the properties inherited towards the direct sub-concepts of the concept to be removed. sH5. Removal of super-concept’s relations. sH6. Propagation of the super-concept’s relations towards its direct sub-concepts. SC7. Removal of the properties of the component concept. SC8. Propagation of the properties of the component concept towards the compound one. sS9. Removal of all relations. sS10. Propagation of these relations towards synonymous concepts. sE1. Removal of all relations. sE2. Propagation of these relations towards equivalent concepts.

Management of inheritance properties Is-a (H)

Composition (C)

Management of the conceptual relations of the super-concept Management of the component concept properties

Synonymy (S)

Management of conceptual relations

the

Equivalence (E)

Management of conceptual relations

the

4.1.2. Propagation of the complex changes We define new strategies of evolution for complex changes allowing to take care of the conceptual and semantic relations. These strategies are defined at the level of Table 3. We extend the operator of generalization in the case of the IS design ontology, by factorizing properties added to conceptual relations linking up sub-concepts.

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

Table 3. Evolution strategies to solve complex changes

Changment

Relations

Evolution strategies (S)

Generalization of concepts (G) Split a concept (S)

Conceptual relations

Merge of concepts (M)

Conceptual relations

SG1. Leaving all relations of the concepts to be generalized undamaged. SG2. Migration of all relations towards the new super-concept. SS1. Removal of the conceptual relations of the splitted-concept. SS2. Copying the conceptual relations for all concepts of splitting. SS3. Conservation of the conceptual relations in the concept directed by the splitting. SM1. Removal of the conceptual relations of the concepts to be merged. SM2. Migration of all the merged concepts’ relations towards the merge- concept (a selective migration according to the users’ needs can be performed).

Conceptual relations

4.2. Propagation of ontological changes at the semantic level of the IS design ontology Any modification to be brought on the IS design ontology’s conceptual level can be accompanied by auto-evolution actions focusing on semantic relations. Table 4 describes the change operators’ semantic effects defined to evolve the IS ontology. The operators which are not invoked there (Rename-Property(), Add-Relation(), RenameRelation()and Remove-Relation()) do not have effects on the semantic relations.

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

20

H. Hamdani et al. / Towards an Approach for Evolving Information Systems’ Ontologies

Table 4. The IS design ontology operators’ effects on the semantic relations

X

X X X X

X X X

Removeantonymy ()

X X X X

Addantonymy()

X

X X X X X X

X X

Removehomonymy()

X X X X X X

Addhomonymy

X X X X X X

Removeequivalence ()

X

Addequivalence ()

Removesynonymy ()

Addsynonymy ()

Add-concept Rename-concept Remove-concept Add-property Remove-Property Merge-Concept Generalize-concept Split-concept

X X

X X X

X X X X

We notice that it is not any more a question of defining explicit operators to manipulate semantic relations. But we envisage functions to be automatically called during the change operators’ application. Functions which we define to take care of the ontology auto-evolution are: Add-synonymy (), Remove-synonymy (), Add-equivalence (), Remove-equivalence (), Add-antonymy (), Remove-antonymy (), Add- Homonymy() and Remove- Homonymy().

5. Validation of changes The validation is made according to the ontology coherence in comparison with its domain, as well as to its structural and semantic consistency. In the course of this step, the users evaluate the evolution result and start again the evolution process, if necessary. Validation can be partial or complete.

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

5.1. Partial validation Partial validation is the one which is made after any change’s accomplishment. This validation can be applied in the following cases: The experimentation aspect: It happens that the ontology is evolved with the intention of its functionality’s experimentation. The users can fail to choose the derived changes: The ontology users can fail to understand the possible effects of a performed change. As a result, they has to choose the most relevant effects The sharing out aspect at the use level: If the ontology is shared or used in a distributed context, different users can have different ideas at the level of the changes’ resolution. 5.2. Complete validation The main objective of the complete validation is to define some rules allowing to control the evolved ontological structure. For example, the execution of a conceptual

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

H. Hamdani et al. / Towards an Approach for Evolving Information Systems’ Ontologies

21

relation removal can procreate the presence of an isolated concept. In that case the user must be informed about the effect of this change. Among the rules which we offer to fulfill this validation, we name: • Superfluous concepts must be removed. • Empty concepts are candidates to be removed. • A concept having an only sub-concept is candidate to be merged with it. • Isolated concepts are candidates to be removed: The user must be informed about the isolated concepts. Two decisions can be made to solve the isolated concepts: (1) the removal of this concept or (2) the creation of a new conceptual relation allowing to link it to the ontology structure. • Superfluous conceptual relations must be removed: For example two simple equivalent conceptual relations can exist between the same concepts. Besides, if the direct super-concept of a concept can be found in an indirect way, the direct Is-a relation (if exists) linking up this concept with its direct superconcept must be removed. • The same property which exists in all sub-concepts is candidate to be immigrated towards their direct common super-concept (if exists).

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

6. Conclusion In brief, in this paper, we have put the emphasis on the fact that the IS design ontology’s evolution is necessary to keep its survival and spread. The IS design ontological evolution approach offered allows to maintain minimizing the human intervention when solving ontological changes. Indeed, assistance is fulfilled at the evolution strategies’ definition level accompanying the defined operators. The change operators’ application is not limited to the ontology’s evolution domain. Indeed, other research topics interested in the ontology’s reuse such as ontologies’ integration can make appeals to these operators. As perspectives, we mention the extension of this approach to support the ontology’s instance level, the definition of a model for the IS design ontology’s management systems and the definition of an algebra for these systems (if possible). References [1] Amed-nacer, M. Un modèle de gestion et d’évolution de schéma pour les bases de données de génie logiciel, thèse présentée dans l’institut national polytechnique de grenoble, 1994. [2] Gruber, T.R. Toward Principles for the Design of Ontologies Used for knowledge Sharing, Stanford Knowledge Systems Laboratory, (1993). [3] M.mhiri,A.mtibaa,F.gargouri,Towards an approach for building information systems’ ontologies, first Workshop on Formal Ontologies Meet Industry, 2005. [4] M'hiri, M. Méthodologie de construction des ontologies pour la résolution de conflits de Systèmes d’Information, premières Journées Francophones sur les Ontologies : JFO, Sousse-Tunisie, Octobre 2007. [5] M'hiri, M. Gargouri, F. Benslimane, D. Détermination automatique des relations sémantiques entre les concepts d’une ontologie », INFORSID’2006, Hammamet, 31 Mai au 3 juin 2006. [6] M'hiri, M, Mtibaa, A. F.Gargouri. UMLonto : towards an approach for the specification of information systems' ontologies, the seventeenth international conference on software engineering and knowledge engineering : (SEKE) Taipei, Taiwan-chine, 2005. [7] M. Klein, Supporting evolving ontologies on the internet, paper presented at the Proceedings of the EDBT 2002 PhD Workshop, Prague, Czech Republic, 2002. [8] Stojanovic, L. Methods and Tools for Ontology Evolution. Mémoire de thèse, Université of Karlsruhe, (2004).

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

22

Formal Ontologies Meet Industry R. Ferrario and A. Oltramari (Eds.) IOS Press, 2009 © 2009 The authors and IOS Press. All rights reserved. doi:10.3233/978-1-60750-047-6-22

A First-Order Cutting Process Ontology for Sheet Metal Parts Michael GRÜNINGER a , Arnaud DELAVAL b of Mechanical and Industrial Engineering, University of Toronto, Toronto, Ontario, Canada M5S 3G8 b IFMA Les Cezaux BP 265 63175 Clermont Ferrand, France

a Department

Abstract. The semantic integration of manufacturing systems has been impeded by the lack of rigorous ontologies for specific domains of manufacturing processes and resources. In this paper we present a cutting process ontology for 2D shapes such as sheet metal parts, axiomatized in first-order logic. This ontology is an extension of the ontology of ISO 18629 (Process Specification Language) and an earlier shape ontology first used to support object recognition. The full ontology consists of an axiomatization of all possible ways to change a surface as the result of a cutting process and a taxonomy of cutting processes. All component ontologies are verified using representation theorems. Keywords. manufacturing, ontologies, first-order logic, Process Specification Language

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

1. Motivation Although 3D machining receives a great deal of attention, manufacturing of twodimensional parts is also widespread. The cutting of sheet metal has applications in car bodies, airplane wings, medical tables, and construction materials. Similar processes are used not only in metal machining but also in the cutting of materials ranging from paper to fabric. Earlier work in the application of knowledge representation to sheet metal manufacturing processes has been restricted to approaches that are not based on a formal logic. The work in [6] uses a rule-based approach to capture the relationship between particular features of a sheet metal part and the possible operations that can be performed to produce the desired features. Approaches such as [7] specify grammar rules to define the slitting operations that can be done on any given sheet metal plate. One drawback of this earlier work is the lack of sharability and reusability for the specifications of the manufacturing processes such as cutting or punching. Rather than a set of rules, we need a logical framework that can support both consistency-checking (to verify plans composed of cutting processes) as well as automated inference to reason about the consequences of particular cutting processes (for example, can one perform a different cutting process within a plan and still achieve a particular feature within the final product). Another desirable application that is not supported by earlier work is the retrieval of cutting operations and partial plans from process repositories. Such repositories presume the existence of a classification of cutting processes. Moreover, this cannot be an ad hoc

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

M. Grüninger and A. Delaval / A First-Order Cutting Process Ontology for Sheet Metal Parts

23

classification; it should be one that is provably correct and complete with respect to the underlying definitions of cutting processes. To address the above shortcomings and additional requirements, we propose an ontology for cutting processes in which the class definitions and other constraints are axiomatized in first-order logic. If we consider the class of manufacturing processes defined as activities that change shape, then cutting processes are the subclass of such shapechanging processes in which at least two new edges are created as a result of an occurrence of the process. Consequently, the cutting process ontology presented in this paper is based on two existing first-order ontologies – the ontology of 2D shapes introduced in the CardWorld Ontology [4], and the ontology of ISO 18629 (Process Specification Language [3], [2]). The Cutting Process Ontology consists of three sets of first-order axioms: • Shape Ontology • Shape Cutting Ontology • Cutting Process Taxonomy In the remainder of this paper, we will consider each of the first-order ontologies within the Cutting Process Ontology. We not only present the axioms within each ontology, but we also demonstrate the verification of the axioms with respect to their intended models.

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

2. PSL Ontology The purpose of PSL-Core ([3], [2]) is to axiomatize a set of intuitive semantic primitives that is adequate for describing the fundamental concepts of manufacturing processes. Consequently, this characterization of basic processes makes few assumptions about their nature beyond what is needed for describing those processes, and the Core is therefore rather weak in terms of logical expressiveness. Within PSL-Core 1 , there are four kinds of entities required for reasoning about processes – activities, activity occurrences, timepoints, and objects. Activities may have multiple occurrences, or there may exist activities which do not occur at all. Timepoints are linearly ordered, forwards into the future, and backwards into the past. Finally, activity occurrences and objects are associated with unique timepoints that mark the begin and end of the occurrence or object. Within the PSL Ontology, the theory Tocctr ee extends the theory of T pslcor e 2 . An occurrence tree is a partially ordered set of activity occurrences, such that for a given set of activities, all discrete sequences of their occurrences are branches of the tree. An occurrence tree contains all occurrences of all activities; it is not simply the set of occurrences of a particular (possibly complex) activity. Because the tree is discrete, each activity occurrence in the tree has a unique successor occurrence of each activity. Every sequence of activity occurrences has an initial occurrence (which is the root of an occurrence tree). 1 The axiomatization of PSL-Core in CLIF (Common Logic Interchange Formt) can be found at

http://www.mel.nist.gov/psl/psl-ontology/psl_core.html 2 The axioms of T occtr ee in CLIF can be found at http://www.mel.nist.gov/psl/psl-ontology/ part12/occtree.th.html Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

24

M. Grüninger and A. Delaval / A First-Order Cutting Process Ontology for Sheet Metal Parts

Most applications of process ontologies are used to represent dynamic behaviour in the world so that intelligent agents may make predictions about the future and explanations about the past. In particular, these predictions and explanations are often concerned with the state of the world and how that state changes. The PSL core theory Tdisc_state is intended to capture the basic intuitions about states and their relationship to activities3 . Within the PSL Ontology, state is changed by the occurrence of activities. Intuitively, a change in state is captured by a state that is either achieved or falsified by an activity occurrence. We therefore use the prior relation to specify the properties (known as fluents) that are intuitively true prior to an activity occurrence and also the holds relation that specifies the fluents that are intuitively true after an activity occurrence. Furthermore, state can only be changed by the occurrence of activities. Thus, if some state holds after an activity occurrence, but after an activity occurrence later along the branch it is false, then an activity must occur at some point between that changes the state. This also leads to the requirement that the state holding after an activity occurrence will be the same state holding prior to any immediately succeeding occurrence, since there cannot be an activity occurring between them.

3. Shape Ontology The Shape Ontology (Tshape ) is based on the CardWorld Ontology [4], which is a firstorder ontology for 2D-object recognition in scenes with occlusion and images with noise. In the axiomatization of the Shape Ontology (see Figures 1 and 2), we focus on the mereotopological relations (i.e. parthood and connection) rather than geometric relations (such as relative alignment and length of segments, or the notions of curvature or surface area).

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

3.1. Axiomatization There are three sorts of objects in the domain – surfaces, edges, and points. Every point is part of some edge and every edge is part of a unique surface. Every surface contains at least two edges and every edge contains at least two points. A vertex is a point that is part of two edges, and only two edges may meet at a point. In the original CardWorld Ontology, the objects such as surfaces, edges, and points were represented as classes, and the relationships between these objects were represented by relations. On the other hand, we need to be able to represent how properties of surfaces, edges, and points can change as the result of activity occurrences. Moreover, we need to be able to capture the creation of new surfaces, edges, and points. Within the PSL Ontology, properties of the world that change as the result of activity occurrences are represented as fluents. Consequently, all of the relations in the original CardWorld Ontology become fluents in the Shape Ontology. Thus, the three classes of objects become the unary fluent functions sur f ace(x), edge(x), and point (x), since new surfaces, edges, and points may be created by an activity. The binary fluent function par t (x, y) specifies the containment relationships between the elements of surfaces. The ternary fluent function meet (e1 , e2 , v) specifies the relationship between two edges that 3 The axioms of T disc_state in CLIF can be found at http://www.mel.nist.gov/psl/ psl-ontology/part12/disc_state.th.html

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

M. Grüninger and A. Delaval / A First-Order Cutting Process Ontology for Sheet Metal Parts

25

meet at a vertex (common point). The unary fluent function outer (e) distinguishes edges that are part of a hole within a surface from those edges that are part of the outer boundary of the surface. Finally, the binary fluent function connected(e1 , e2 ) captures the relationship between edges that are part of the same boundary within a surface, whether this is the outer boundary or the boundary of a hole. Within the Cutting Process Ontology, the axioms of the Shape Ontology are state constraints, that is, sentences that must be true prior to any activity occurrence. By the axioms of the PSL Ontology, these sentences also hold after any activity occurrence in the occurrence tree, since they cannot be falsified by any activity occurrence. 3.2. Verification of the Shape Ontology The ontology is verified by providing a complete characterization of all models of the axioms up to isomorphism. One approach to this problem is to use representation theorems – we evaluate the adequacy of the ontology with respect to some well-understood class of mathematical structures (such as partial orderings, graph theory, and geometry) that capture the intended interpretations of the ontology’s terms. Given the definition of some class of structures M, we prove that the class exists and is nonempty, which also provides a characterization of the structures in the class up to isomorphism. We prove that every structure in the class is a model of the ontology and that every countable model of the ontology is isomorphic to some structure in the class. To formally capture these intuitions for the Shape Ontology, we first define a few classes of combinatorial structures [1] which will be the building blocks of models for the ontology.

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

Definition 1 A tripartite incidence structure is a tuple G = (1 , 2 , 3 , I ), where 1 , 2 , 3 are pairwise disjoint sets such that I ⊆ (1 × 2 ) ∪ (1 × 3 ) ∪ (2 × 3 ). Two elements of G that are related by I are called incident. A flag of G is a set of elements of 1 ∪ 2 ∪ 3 that are mutually incident. Using these definitions, our intuitions tell us that scene elements should form tripartite incidence structures in which all maximal flags have three elements, since all objects should be part of some surface, i.e., there should not exist any isolated edges or points. Definition 2 A shape structure is a tripartite incidence structure S = S, E, P, part such that all elements of E are elements of two maximal flags in S which contain a unique element in S. The existence of shape structures is established by the following theorem, which also provides a characterization up to isomorphism: Theorem 1 A tripartite incidence structure O = S, E, V, part is a shape structure iff it is isomorphic to the incidence structure P(G), G, V, ∈ in which G is a set of cyclic graphs with vertices V and P(G) is a partitioning of G. Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

26

M. Grüninger and A. Delaval / A First-Order Cutting Process Ontology for Sheet Metal Parts

∀x, o ¬( prior ( point (x), o) ∧ prior (edge(x), o))

(1)

∀x, o ¬( prior ( point (x), o) ∧ prior (sur f ace(x), o))

(2)

∀x, o ¬( prior (edge(x), o) ∧ prior (sur f ace(x), o))

(3)

(∀x, s, o) prior ( par t (x, s), o) ∧ prior (sur f ace(s), o) ⊃ ¬ prior (sur f ace(x), o)

(4)

(∀x, e, o) prior ( par t (x, e), o) ∧ prior (edge(e), o) ⊃ ¬ prior (sur f ace(x), o) ∧ ¬ prior (edge(x), o)

(5)

(∀x, p, o) prior ( par t (x, p), o) ∧ prior ( point ( p), o) ⊃ ¬ prior (sur f ace(x), o) ∧ ¬ prior (edge(x), o) ∧ ¬ prior ( point (x), o)

(6)

(∀x, s, o) prior ( par t (s, x), o) ∧ prior (sur f ace(s), o) ⊃ ¬ prior (sur f ace(x), o) ∧ ¬ prior (edge(x), o) ∧ ¬ prior ( point (x), o)

(7)

(∀x, e, o) prior ( par t (e, x), o) ∧ prior (edge(e), o) ⊃ ¬ prior (edge(x), o) ∧ ¬ prior ( point (x), o)

(8)

(∀x, p) par t ( p, x) ∧ point ( p) ⊃ ¬ point (x)

(9)

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

(∀e, s, v, o) prior (edge(e), o) ∧ prior (sur f ace(s), o) ∧ prior ( par t (e, s), o) ∧ prior ( par t (v, e), o) ⊃ prior ( par t (v, s), o)

(10)

(∀x, o) prior (edge(x), o) ⊃ (∃s) prior (sur f ace(s), o) ∧ prior ( par t (x, s), o)

(11)

(∀x, o) prior ( point (x), o) ⊃ (∃e) prior (edge(e), s) ∧ prior ( par t (x, e), o)

(12)

(∀v, s1 , s2 , o) prior ( par t (v, s1 ), o) ∧ prior ( par t (v, s2 ), o) ∧ prior ( point (v), o) ∧ prior (sur f ace(s1 ), o) ∧ prior (sur f ace(s2 ), o) ⊃ (s1 = s2 )

(13)

(∀e, s1 , s2 , o) prior ( par t (e, s1 ), o) ∧ prior ( par t (e, s2 ), o) ∧ prior (edge(e), o) ∧ prior (sur f ace(s1 ), o) ∧ prior (sur f ace(s2 ), o) ⊃ (s1 = s2 )

(14)

(∀s, o) prior (sur f ace(s), o) ⊃ (∃e1 , e2 , e3 ) prior (edge(e1 ), o) ∧ prior (edge(e2 ), o) ∧ prior (edge(e3 ), o) ∧ (e1 = e2 ) ∧ (e1 = e3 ) ∧ (e2 = e3 ) ∧ prior ( par t (e1 , s), o) ∧ prior ( par t (e2 , s), o) ∧ prior ( par t (e3 , s), o)

(15)

(∀e, o) prior (edge(e), o) ⊃ (∃p1 , p2 ) prior ( point ( p1 ), o) ∧ prior ( point ( p2 ), o) ∧( p1 = p2 ) ∧ prior ( par t ( p1 , e), o) ∧ prior ( par t ( p2 , e), o)

Figure 1. Tshape : Shape axioms. Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

(16)

27

M. Grüninger and A. Delaval / A First-Order Cutting Process Ontology for Sheet Metal Parts

Two edges meet at a vertex iff they are distinct and the vertex is part of both edges. (∀e1 , e2 , v, o) prior (meet (e1 , e2 , v), o) ≡ ( prior (edge(e1 ), o) ∧ prior (edge(e2 ), o) ∧ prior ( point (v), o) ∧ prior ( par t (v, e1 ), o) ∧ prior ( par t (v, e2 ), o) ∧ (e1 = e2 )

(17)

Every edge meets another distinct edge. (∀e1 , o) prior (edge(e1 ), o) ⊃ (∃e2 ) prior (meet (e1 , e2 , v), o)

(18)

Exactly two edges meet at a vertex. (∀e1 , e2 , e3 , v, o) prior (meet (e1 , e2 , v), o)∧ prior (meet (e1 , e3 , v), o) ⊃ (e2 = e3 )

(19)

All outer edges are connected. (∀e1 , e2 , o) prior (outer (e1 ), o) ⊃ ( prior (outer (e2 ), o) ≡ prior (connected(e1 , e2 ), o)) (20)

Figure 2. Tshape : Shape axioms.

We next define the class of structures that are isomorphic to the intended models of Tshape . Definition 3 Let Mshape be the class of structures such that M ∈ Mshape iff

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

1. there exists a model N of Tdisc_state ∪ Tocctr ee ∪ T pslcor e such that N ⊂ M; 2. each element of the occurrence tree in N is associated with a shape structure. The following representation theorem constitutes the verification of the Shape Ontology; as a consequence, it also demonstrates the consistency of the ontology. Theorem 2 M ∈ Mshape iff it is isomorphic to a countable model of Tshape ∪ Tdisc_state ∪ Tocctr ee ∪ T pslcor e .

4. Shape Cutting Ontology Each model of the original shape axioms from the CardWorld Ontology corresponds to a different state within an occurrence tree in a model of Tshape ∪ Tdisc_state ∪ Tocctr ee ∪ T pslcor e . On the one hand, changing a shape is equivalent to changing state within the occurrence tree; on the other hand, it is equivalent to a mapping between different models of the original shape axioms. Since we have already shown how the models of the Shape Ontology are isomorphic to a class of tripartite incidence structures, we can characterize all possible mappings between structures in this class. With respect to state, any property of a model that is not preserved by a mapping corresponds to a fluent that is either

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

28

M. Grüninger and A. Delaval / A First-Order Cutting Process Ontology for Sheet Metal Parts

achieved or falsified 4 . This correspondence forms the basis for the specification of the models of the axioms in the Shape Cutting Ontology. 4.1. Examples of Shape Cutting Activities Suppose we begin with the surface depicted in Figure 3(a). With the cutting activity depicted in Figure 3(b), three new edges e5 , e6 , e7 are created; of these three, e7 is a modified edge, since all of its points were formerly parts of the edge e1 . In Figure 3(c), two new edges e5 , e6 are created, neither of which is a modified edge. In both of these cases, no new surface is created. We next consider cases in which a new surface is created. In Figure 3(d), a new surface is created which contains two new modified edges e5 , e6 as well as an existing edge e2 and a new edge e7 . In addition, the original surface contains a new edge. In Figure 3(e), a new surface is created with two new edges, one of which is modified. In Figure 3(f), a new surface is created and each surface contains a new edge, but none of the new edges are modified. In all of the above examples, the outer fluent was unchanged. In Figure 3(g), two new edges are created which form a hole in the surface. Suppose we begin with the surface in Figure 3(h); the surface in Figure 3(i) depicts the effect of an occurrence of the cutting activity that falsifies the outer fluent for all of the edges that were formerly part of the whole. One new outer edge and one new non-outer edge are modified. The activity whose occurrence is depicted in Figure 3(j) is a variant in which the outer fluent is falsified but only two edges are created, neither of which is modified.

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

4.2. Axiomatization The Shape Cutting Ontology (Tcutshape in Figures 4 and 5) begins with the definition of a cutting process to be any activity that creates at least two new edges in some surface. The remaining axioms explicate the different ways that a cutting process can possibly change the properties of a surface and its components. In particular, the axioms specify the conditions under which the fluents that specify the properties of surfaces ( par t, meet, and outer ) can be achieved or falsified by occurrences of cutting processes. For example, the par t fluent changes as the result of edges or surfaces being created. New points are never created without an edge being created; if the edge already exists, some points simply become the vertices where other new edges meet. 4.3. Verification of the Shape Cutting Ontology Models of Tcutshape are based on the notion of partial automorphisms of a shape structure. A mapping ϕ : S → S is a partial automorphism iff it is an isomorphism between substructures of S. We use the notation dom(ϕ) to denote the set of elements in the substructure of S that is the domain of the mapping. The set of partial automorphisms of a shape structure forms an inverse semigroup, denoted by P Aut (S). 4 The following definitions are used from the PSL Ontology: (∀o, f ) achieves(o, f ) ≡ (¬ prior ( f, o) ∧ holds( f, o)) (∀o, f ) f alsi f ies(o, f ) ≡ ( prior ( f, o) ∧ ¬holds( f, o)) (∀o, f ) changes(o, f ) ≡ (achieves(o, f ) ∨ f alsi f ies(o, f )).

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

M. Grüninger and A. Delaval / A First-Order Cutting Process Ontology for Sheet Metal Parts

e1

e1 e5

e4

e2

e6 e2

e7

e3

e2

(d) e1 e5

e6

e4

e5

e4

e2

(e)

e3

(f)

e1

(g) e11

e1 e12

e5 e6

e2 e4

e5 e8

e1

e10 e9 e7

e2

e6

e3

e5

e3

e2

e7 e6

(c)

e6

e7

e8

e4

e1

e4

e5

e2

e3

(b)

e1

e8

e6

e4

e3

(a)

e1 e5

e4

e3

e4

e1

e7

29

e6

e10

e5

e9

e2 e4

e8

e6

e2

e7

e3 e3

(h)

e3

(i)

(j)

Figure 3. Examples of shape cutting processes.

Definition 4 Let Mcutshape be the class of structures such that M ∈ Mcutshape iff

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

1. there exists N ∈ Mshape such that N ⊂ M; 2. each element of the occurrence tree in N is associated with a unique ϕ ∈ P Aut (S) such that f ∈ dom(ϕ) iff o, f ∈ changes; 3. o, edge(e1 ), o, edge(e2 ) ∈ achieves Intuitively, the partial automorphisms capture the substructure of a shape structure that is not changed as the result of an occurrence of a cutting process. Theorem 3 M ∈ Mcutshape iff it is isomorphic to a countable model of Tcutshape ∪ Tshape ∪ Tdisc_state ∪ Tocctr ee ∪ T pslcor e . 5. Cutting Process Taxonomy The Cutting Process Taxonomy is a classification of cutting processes, that is, activities which satisfy Axiom 21. This is equivalent to characterizing all possible ways to change a surface that satisfies the axioms of Tshape in such a way that at least two edges are created. Since the axiomatization in Tcutshape provides such a characterization, the classification of cutting processes corresponds to the classification of the models of Tcutshape . 5.1. Classes of Cutting Processes Within the PSL Ontology, the taxonomy of activities arises from invariants that are used to classify the models of the core theories [5]. Invariants are properties of models that Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

.

30

M. Grüninger and A. Delaval / A First-Order Cutting Process Ontology for Sheet Metal Parts

A cutting process creates two new edges in some surface. (∀a) cutting(a) ≡ ((∀o) occurr ence_o f (o, a)

(21)

⊃ (∃e1 , e2 , s) holds(edge(e1 ), o) ∧ holds(edge(e2 ), o) ∧ (e1 = e2 ) ∧ prior (sur f ace(s), o) ∧ achieves(o, par t (e1 , s)) ∧ achieves(o, par t (e2 , s)) Surfaces, edges, and points are never destroyed. (∀x, o, a) cutting(a) ∧ occurr ence_o f (o, a) ⊃ ¬( f alsi f ies(o, sur f ace(x))∨ f alsi f ies(o, edge(x))∨ f alsi f ies(o, point (x))) (22) New surfaces contain existing edges. (∀s, o)achieves(o, sur f ace(s)) ⊃ (∃e) prior (edge(e), o)∧achieves(o, par t (e, s)) (23) Existing edges never contain new points. (∀e, v, o, a) cutting(a) ∧ occurr ence_o f (o, a) ∧ prior (edge(e), o) ⊃ ¬achieves(o, par t (v, e))

(24)

We can never achieve meet for existing edges. (∀e1 , e2 , s, v, o, a) cutting(a) ∧ occurr ence_o f (o, a) ∧ prior ( par t (e1 , s), o)∧ prior ( par t (e2 , s), o) ⊃ ¬achieves(o, meet (e1 , e2 , v))

(25)

Outer edges are always preserved.

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

(∀s, o, a) cutting(a) ∧ occurr ence_o f (o, a) ⊃ ¬ f alsi f ies(o, outer (s))

(26)

Every new outer edge contains an existing point. (∀e, o, a) cutting(a) ∧ occurr ence_o f (o, a) ∧ achieves(o, edge(e)) ∧achieves(o, outer (e)) ⊃ (∃p) holds( par t ( p, e), o) ∧ prior ( point ( p), o)

(27)

Figure 4. Axioms of Tcutshape : Shape Cutting Ontology.

are preserved by isomorphism. For some classes of structures, invariants can be used to classify the structures up to isomorphism; for example, vector spaces can be classified up to isomorphism by their dimension. For other classes of structures, such as graphs, it is not possible to formulate a complete set of invariants. Nevertheless, even without a complete set, invariants can still be used to provide a classification of the models of a theory. We use the following invariants to classify the models of Tcutshape :

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

M. Grüninger and A. Delaval / A First-Order Cutting Process Ontology for Sheet Metal Parts

31

A modified edge is a new edge that does not contain any new points. (∀e, o) modi f ied_edge(e, o) ≡

(28)

(achieves(o, edge(e)) ∧ ((∀ p) holds( par t ( p, e), o) ⊃ prior ( point ( p), o)) A modified edge corresponds to a subset of points for an existing edge. (∀e1 , o, a) cutting(a) ∧ occurr ence_o f (o, a) ∧ modi f ied_edge(e1 , o) ⊃ (∃e2 )holds(edge(e2 ), o)∧((∀ p)holds( par t ( p, e1 ), o) ⊃ prior ( par t ( p, e2 ), o)) (29) Every new modified edge must meet an existing edge. (∀e1 , o, a) cutting(a) ∧ occurr ence_o(o, a) ∧ modi f ied_edge(e1 , o) ⊃ (∃e2 , v) achieves(o, meet (e1 , e2 , v)) ∧ prior (edge(e2 ), o)

(30)

Figure 5. Axioms of Tcutshape : Shape Cutting Ontology.

1. 2. 3. 4.

number of surfaces that are created (which is either zero or one); number of holes that are destroyed (which is either zero or one); number of edges that are created (which is either two, three, or four); number of pairs of existing edges that are changed (which is either zero, one, or two).

The axioms of Tcut pr ocess in Figure 6 and Figure 7 explicitly define the classes of primitive activities that correspond to each of the possible values for the above invariants.

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

5.2. Verification of the Cutting Process Ontology The correctness of Tcut pr ocess is established by the following theorem, which has also been automatically derived using the Prover9 resolution theorem prover. Theorem 4 Let Tcpo = Tcut pr ocess ∪ Tcutshape ∪ Tshape ∪ Tdisc_state ∪ Tocctr ee ∪ T pslcor e . The classes activities related to each invariant are disjoint: Tcpo | ¬(∃a) (cr eate_sur f ace(a) ∧ pr eser ve_sur f ace(a)) Tcpo | ¬(∃a) (destr oy_hole(a) ∧ pr eser ve_hole(a)) Tcpo | ¬(∃a) ( pr eser ve_meet (a) ∧ change_one_meet (a) ∧ change_two_meet (a)) Tcpo | ¬(∃a)(cr eate_two_edge(a)∧cr eate_thr ee_edge(a)∧cr eate_ f our _edge(a))

6. Summary In this paper we have introduced an ontology for cutting processes in the manufacturing domain of sheet metal parts. We have presented a first-order axiomatization of classes of intended models and verified the ontology by proving representation theorems with respect to these intended models.

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

32

M. Grüninger and A. Delaval / A First-Order Cutting Process Ontology for Sheet Metal Parts

Occurrences of activities in this class create a new surface. (∀a) cr eate_sur f ace(a) ≡

(31)

(∀o) occurr ence_o f (o, a) ⊃ (∃s) achieves(o, sur f ace(s)) Occurrences of activities in this class do not create a new surface. (∀a) pr eser ve_sur f ace(a) ≡

(32)

(∀o) occurr ence_o f (o, a) ⊃ ¬(∃s) achieves(o, sur f ace(s)) Occurrences of activities in this class destroy a hole by changing all non-outer edges to outer edges in the same surface. (∀a) destr oy_hole(a) ≡

(33)

(∀o) occurr ence_o f (o, a) ⊃ (∃e) prior (edge(e), o) ∧ achieves(o, outer (e)) Occurrences of activities in this class do not change any existing edges in a surface to be outer edges. (∀a) pr eser ve_hole(a) ≡

(34)

(∀o, e) occurr ence_o f (o, a) ∧ prior (edge(e), o) ⊃ ¬achieves(o, outer (e)) Occurrences of activities in this class do not change the set of existing edges that meet other existing edges in the surface.

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

(∀a) pr eser ve_meet (a) ≡ ((∀o) occurr ence_o f (o, a) ⊃

(35)

¬(∃e1 , e2 , v) (achieves(o, meet (e1 , e2 , v)) ∨ f alsi f ies(o, meet (e1 , e2 , v))) Two existing edges no longer meet after occurrences of activities in this class. (∀a) change_one_meet (a) ≡

(36)

((∀o) occurr ence_o f (o, a) ⊃ (∃e1 , e2 , v) f alsi f ies(o, meet (e1 , e2 , v))) Two pairs of existing edges no longer meet after occurrences of activities in this class. (∀a) change_two_meet (a) ≡ ((∀o) occurr ence_o f (o, a) ⊃

(37)

(∃e1 , e2 , e3 , e4 , v 1 , v 2 ) f alsi f ies(o, meet (e1 , e2 , v 1 )) ∧ f alsi f ies(o, meet (e3 , e4 , v 2 )))

Figure 6. Axioms of Tcut pr ocess : Cutting Process Ontology.

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

M. Grüninger and A. Delaval / A First-Order Cutting Process Ontology for Sheet Metal Parts

33

Exactly two new edges are created by occurrences of activities in this class. (∀a) cr eate_two_edge(a) ≡ ((∀o) occurr ence_o f (o, a) ⊃

(38)

(∃e1 , e2 ) achieves(o, edge(e1 )) ∧ achieves(o, edge(e2 )) ∧((∀e) achieves(o, edge(e)) ⊃ ((e = e1 ) ∨ (e = e2 ))) Exactly three new edges are created by occurrences of activities in this class. (∀a) cr eate_thr ee_edge(a) ≡ ((∀o) occurr ence_o f (o, a) ⊃

(39)

(∃e1 , e2 , e3 ) achieves(o, edge(e1 )) ∧ achieves(o, edge(e2 )) ∧ achieves(o, edge(e3 )) ∧((∀e) achieves(o, edge(e)) ⊃ ((e = e1 ) ∨ (e = e2 ) ∨ (e = e3 ))) Exactly four new edges are created by occurrences of activities in this class. (∀a) cr eate_ f our _edge(a) ≡ ((∀o) occurr ence_o f (o, a) ⊃

(40)

(∃e1 , e2 , e3 , e4 ) achieves(o, edge(e1 )) ∧ achieves(o, edge(e2 )) ∧achieves(o, edge(e3 )) ∧ ∧achieves(o, edge(e4 )) ∧((∀e) achieves(o, edge(e)) ⊃ ((e = e1 ) ∨ (e = e2 ) ∨ (e = e3 ) ∨ (e = e4 )))

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

Figure 7. Axioms of Tcut pr ocess : Cutting Process Ontology.

Several lines for future work are evident. First is the extension of the Shape Ontology to incorporate geometric fluents, such as relative alignment and the length of edges. The second is the extension of the Cutting Process Ontology to complex activities, which will support process planning with cutting processes.

References [1] Buekenhout, F. (1995) Handbook of Incidence Geometry. North-Holland. [2] Bock, C. and Gruninger, M. (2005) PSL: A semantic domain for flow models, Software and Systems Modeling 4:209-231. [3] Gruninger, M. (2003) Ontology of the Process Specification Language, pp. 599-618, Handbook of Ontologies and Information Systems, S. Staab (ed.). Springer-Verlag, Berlin. [4] Gruninger, M. (1993) Grouping Assumptions in Shape-Based Object Recognition, pp. 32-37, Working Notes AAAI Spring Symposium Series 1993: AI and NP-Hard Problems, Stanford. [5] Gruninger, M. and Kopena, J. (2005) Semantic Integration through Invariants, AI Magazine, 26:11-20. [6] de Sam Lazaro, A., and Engquist, D., and Edwards. D.B. (1993) An Intelligent Design for Manufacturability System for Sheet-metal Parts, Concurrent Engineering: Research and Applications,1:117-123. [7] Soman, A. and Padhye, S. and Campbell, M. (2003) Toward an automated approach to the design of sheet metal components, Artificial Intelligence for Engineering Design, Analysis and Manufacturing, 17:187-204.

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

34

Formal Ontologies Meet Industry R. Ferrario and A. Oltramari (Eds.) IOS Press, 2009 © 2009 The authors and IOS Press. All rights reserved. doi:10.3233/978-1-60750-047-6-34

Parts, Compositions and Decompositions of Functions in Engineering Ontologies Pieter E. Vermaas1 Philosophy Department, Delft University of Technology, The Netherlands

Abstract. In this paper I explore the possibility of introducing in engineering ontologies a generic part-whole relation for functions of technical artefacts by functional composition or functional decomposition. I show by means of the postulates of mereology that general functional compositions and decompositions cannot define such a relation. Yet, one can argue that functional decompositions that are acceptable in engineering may define a proper generic part-whole relation for functions. This possibility requires that (i) the part-whole relation is relative to the specific organisation of functions in decompositions, (ii) there is no strict symmetry between functional composition and decomposition, and (iii) functional decomposition is not transitive. Keywords. Technical function, part-whole relation, mereology, functional composition, functional decomposition, engineering ontology

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

Introduction Functional composition and decomposition are engineering techniques for relating the technical functions of artefacts and their components, e.g., [1-7]. These techniques are used in the conceptual phase of engineering designing when a required overall function of some product-to-be is decomposed into a number of other functions. They are used also in reverse engineering and engineering analyses when an overall function is derived from a series of functions. And functional composition and decomposition are included in engineering knowledge bases for capturing how functions are related to one another [8,9]. Functional composition and decomposition seem thus to define relations between functions that should be included in engineering ontologies, and for this, these relations are to be characterised. Future engineering computer tools, ranging from the next generation of CAD-CAM systems to engineering knowledge bases, can then facilitate engineering reasoning about functions. There is already work done on analysing the relations generated by functional composition and decomposition. But these analyses proceed typically in a roundabout way. Authors analyse the relations between the structural parts and wholes of artefacts that are singled out by the functions that figure in compositions and decompositions, e.g., [10-16]. The direct relations between the functions themselves2 are considered less 1

Corresponding Author: Philosophy Department, Delft University of Technology, Jaffalaan 5, 2628 BX Delft, The Netherlands; [email protected]. 2 A wall, for instance, may be a functionally-defined structural part of a house but the wall is in the strict meaning used in this paper not a functional part of the house. Rather the to support function of the wall is in this case a functional part of the function to provide shelter of the house. Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

P.E. Vermaas / Parts, Compositions and Decompositions of Functions in Engineering Ontologies

35

often [9,17-19]. But all analyses have a starting point in common, which is that these relations are mereological ones concerning parts and wholes. In this paper I report results [20]3 about an analysis of the direct relations between functions as defined by functional composition or decomposition, and about the possibility to capture these relations by means of generic mereological part-whole relations for functions. With generic part-whole relations I here mean part-whole relations that hold for all instances of types of technical functions. I start by introducing functional composition and decomposition in section 1 and list in section 2 the possibilities for introducing a part-whole relation for functions by means of functional composition or decomposition. For evaluating these possibilities I give the minimal postulates for a part-whole relation in section 3. In section 4 it is proved on the basis of a number of examples that functional composition or decomposition cannot define a part-whole relation. Some of the examples of functional decomposition can however be rejected from an engineering point of view, leaving the possibility, reviewed in section 5, that only engineering acceptable functional decompositions provide a proper part-whole relation for functions. This possibility is available only if one (i) accepts that the part-whole relation is relative to the specific organisation of the functions in decompositions, if one (ii) gives up on a strict symmetry between functional composition and decomposition, and, as is shown in section 6, if one (iii) rejects transitivity of decomposition. Section 7 contains conclusions.

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

1. Technical Functions and Their Composition and Decomposition Laying down my assumptions about technical functions in this paper has the immediate drawback of loosing generality of the arguments; there does not exist consensus in engineering about what technical functions are (e.g., [8,21]), and the resulting leeway in determining “the” engineering concept of technical function gives developers of engineering ontologies sufficient room to disagree. My assumptions about technical functions in this paper build on the engineering view that technical functions can be modelled as operations on matter, energy and signal flows, or as operations on operands (e.g., [1,3,17,19,22-25]). Engineers may take issue with these assumptions by taking this modelling as too limited: it may be argued that there are functions that do not fit the modelling (e.g., [26,27]) or it may be argued that the modelling is too sallow by suppressing, say, the relation between functions and the purposes of designers and users. The arguments given in this paper are restricted to those approaches towards technical functions for which my assumptions hold. My assumptions and notation are similar to those in [9]. Let φ and Φ represent functions in the set F of all functions of technical systems. Assume that functions have functional input and/or functional output in terms of physical systems like matter, forces and fields. This input or output may represent information, in which case the input or output are signals. Functions with different input and/or output are instances of different types of function; functions with similar input and output are instances of the same type of function. An example of a function φ is ‘convert electricity to blue light’, where electricity is the functional input and blue light the functional output. 3

This paper is a shortened and adjusted version of [20].

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

36

P.E. Vermaas / Parts, Compositions and Decompositions of Functions in Engineering Ontologies

Two functions φ1 and φ2 are said to be functionally ordered as “φ1 and then φ2” if and only if the functional output of φ1 is functional input of φ2. Such an ordering is written down as φ1→φ2. The functions ‘convert electricity to white light’ and ‘filter from white light blue light’ are ordered as ‘convert electricity to white light’→‘filter from white light blue light’ when the white-light output of the first function is input to the second function. The functional organisation of a set of functions φ1, φ2, … φn is given by the pairwise orderings φi→φj of those functions (i,j=1,2,…n). The notation for this organisation is Org(φ1,φ2,…φn). The organisation Org(‘convert electricity to white light’, ‘filter from white light blue light’) of the just mentioned functions is given by the single ordering {‘convert electricity to white light’→‘filter from white light blue light’}. Functions φ1, φ2, … φn with organisation Org(φ1,φ2,…φn) compose another function Φ, which is the function with input and output similar to the overall input and output of the functions φ1, φ2, … φn in their organisation Org(φ1,φ2,…φn). This functional composition is captured by the relation Comp(Org(φ1,φ2,…φn),Φ). The functions ‘convert electricity to white light’ and ‘filter from white light blue light’ with the organisation {‘convert electricity to white light’→‘filter from white light blue light’}, for example, compose the function ‘convert electricity to blue light’ (see figure 1b).4 A function Φ is decomposable in a set of functions φ1, φ2, … φn with an organisation Org(φ1,φ2,…φn), only if these latter functions φ1, φ2, … φn in their organisation Org(φ1,φ2,…φn) compose Φ. The notation of functional decomposition is Decomp(Φ,Org(φ1,φ2,…φn)), and one thus has the following necessary condition for functional decomposition:

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

Decomp(Φ,Org(φ1,φ2,…φn)) ⇒ Comp(Org(φ1,φ2,…φn),Φ)

(1)

The function ‘convert electricity to white light’, for instance, can be decomposed in the set of functions {‘separate electricity’, ‘convert electricity to blue light’, ‘convert electricity to red light’, ‘convert electricity to yellow light’, ‘mix blue, yellow and red light in white light’} with organisation {‘separate electricity’→‘convert electricity to blue light’, ‘separate electricity’→‘convert electricity to red light’, etc.} (see figure 1a).

2. A Part-Whole Relation for Technical Functions The relations between the functions Φ, φ1, φ2, … φn, as defined by functional composition Comp(Org(φ1,φ2,…φn),Φ) and decomposition Decomp(Φ,Org(φ1,φ2,…φn)), may now be analysed as defining a generic part-whole relation for functions: if a function φi, i=1,2,…n, occurs in a composition or decomposition of a function Φ, then one may try to take all instances of the type φi as parts of all instances of the type Φ. A number of possibilities are available for capturing this relation in detail, depending on two choices. The first choice is to capture this relation with functional composition or with functional decomposition. This choice does not lead to different options if the relation (1) between functional composition and decomposition is actually symmetric in the sense that each composition provides also a decomposition: 4

When I speak of white light in this paper, I assume that it is a mixture of blue, red and yellow light.

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

P.E. Vermaas / Parts, Compositions and Decompositions of Functions in Engineering Ontologies

Decomp(Φ,Org(φ1,φ2,…φn)) ⇔ Comp(Org(φ1,φ2,…φn),Φ)

37

(2)

Yet this symmetry does not need to hold and there are actually engineering reasons to deny it (see section 5). This first choice thus gives in principle two different options. A second choice concerns the generality of the part-whole relation for functions. One may take this relation as a general one that holds between all the instances of the types of functions φi, i=1,2,…n, and Φ independently of the specific organisation Org(φ1,φ2,…φn) that figures in the functional composition or decomposition, or one may take the part-whole relation as a special relation that holds between all the instances relative to this organisation Org(φ1,φ2,…φn). Let PP(φ,φ') be the notation for the general relation that the function φ is a proper part of the function φ', and let PPOrg(φ1,…φn)(φ,φ') be the notation for the special relation that φ is a proper part of φ' relative to the organisation Org(φ1,…φn). Eventually I use X as shorthand for an organisation, such that PPX(φ,φ') can refer to the special relation. The different possibilities for defining the part-whole relation for functions are then as follows: General composition criterion: (Comp(Org(φ1,…φn),Φ) ∧ n≥2) ⇒ ∀φi∈{φ1,…φn}: PP(φi,Φ)

(3)

Special composition criterion: (Comp(Org(φ1,…φn),Φ) ∧ n≥2) ⇒ ∀φi∈{φ1,…φn}: PPOrg(φ1,…φn)(φi,Φ)

(4)

General decomposition criterion: (Decomp(Φ,Org(φ1,…φn)) ∧ n≥2) ⇒ ∀φi∈{φ1,…φn}: PP(φi,Φ)

(5)

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

Special decomposition criterion: (Decomp(Φ,Org(φ1,…φn)) ∧ n≥2) ⇒ ∀φi∈{φ1,…φn}: PPOrg(φ1,…φn)(φi,Φ)

(6)

The condition n≥2 is added to rule out that functions are their own proper parts by the trivial compositions and decompositions Comp(Org(φ),φ) and Decomp(φ,Org(φ)). In the next section I switch to mereology and give the minimal postulates that hold for a part-whole relation. In section 4 I use these postulates for rejecting all possibilities for defining PP(X)(φ,φ') except the special decomposition criterion (6) under the condition that composition-decomposition symmetry (2) is rejected.

3. Mereology The general theory of part-whole relations is mereology [28,29]. Different theories are distinguished in mereology, yet these theories share a common core, which is called a ground mereology, and that captures the very minimum of the part-whole relation. The postulates of this ground mereology depend on a choice between two different concepts

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

38

P.E. Vermaas / Parts, Compositions and Decompositions of Functions in Engineering Ontologies

of part: a concept by which entities are by definition also always part of themselves, and a concept by which entities are by definition never part of themselves. This second concept is called proper part, and is the one I adopt in this paper by, for instance, adding the condition n≥2 to the four criteria for PP(X)(φ,φ'). If one prefers the first concept of part, this condition can be dropped, and the arguments in the sections 4 to 6 should be rephrased (starting with adopting in this section the relevant alternative postulates). A part-whole relation should satisfy at least three conditions. These conditions are for the proper part concept, respectively, irreflexivity, asymmetry and transitivity ([29], formulae 28-30): ¬PP(X)(φ,φ)

(7)

PP(X)(φ,φ') ⇒ ¬PP(X)(φ',φ)

(8)

(PP(X)(φ,φ') ∧ PP(X)(φ',φ")) ⇒ PP(X)(φ,φ")

(9)

The irreflexivity and transitivity conditions (7) and (9) can be taken as postulates; asymmetry (8) is a theorem that follows from these two postulates.

4. Shedding Light

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

I introduced functional composition and decomposition in section 1 by means of examples in which light is generated, mixed and filtered. In this section I present these examples in detail, and use them to show that many of the possibilities for introducing a part-whole relation for functions are ruled out by the postulates of mereology [20]. Consider the following functions: − − − − − − −

φ1: convert electricity to white light φ2: separate electricity φ3: convert electricity to blue light φ4: convert electricity to red light φ5: convert electricity to yellow light φ6: mix blue, yellow and red light in white light φ7: filter from white light blue light

White light can be created by firstly creating blue, red and yellow light in parallel, and then mixing these colours. And blue light, in turn, can be obtained by creating white light and then filtering the blue light from it. These possibilities give two examples of functional decomposition: Decomp(φ1,Org(φ2,φ3,φ4,φ5,φ6)) and Decomp(φ3,Org(φ1,φ7)). The relation (1) between functional composition and decomposition yields two additional examples of functional compositions: Comp(Org(φ2,φ3,φ4,φ5,φ6),φ1) and Comp(Org(φ1,φ7),φ3) (see figure 1, which also gives the organisations Org(φ2,φ3,φ4,φ5,φ6) and Org(φ1,φ7)). If one now adopts one of the general criteria (3) or (5) for defining the part-whole relation for functions, one obtains from these four examples that PP(φ3,φ1) and PP(φ1,φ3). These results violate the

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

P.E. Vermaas / Parts, Compositions and Decompositions of Functions in Engineering Ontologies

39

asymmetry condition (8), hence, the general composition criterion (3) and the general decomposition criterion (5) do not define a part-whole relation that meets the minimal postulates of mereology. These two criteria can thus be ruled out.

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

Figure 1. Functional compositions and decompositions of white light (a) and blue light (b); el. stands for electricity, w.l. for white light, b.l. for blue light, r.l. for red light and y.l. for yellow light.

This first result that the general criteria (3) and (5) do not give proper part-whole relations for functions may not be that surprising because by these criteria the partwhole relation is independent of the specific functional composition or decomposition that generates it. If one instance of a type of function φi figures in one functional composition Comp(Org(φ1,…φn),Φ) or decomposition Decomp(Φ,Org(φ1,…φn)), not only that instance of φi becomes a proper part of a specific instance of the function Φ, but each instance of φi becomes unconditionally a proper part of each instance of Φ. Each instance of converting electricity to blue light thus become unconditionally a proper part of each instance of converting electricity to white light by the decomposition Decomp(φ1,Org(φ2,φ3,φ4,φ5,φ6)). This seems too much: an instance of converting electricity to blue light may be a proper part of instances of converting electricity to white light relative to the decomposition Decomp(φ1,Org(φ2,φ3,φ4,φ5,φ6)) but relative to another decomposition Decomp(Φ,Org(φ1,…φn)), say, of converting electricity to purple light, converting electricity to blue light may be reasonably supposed not to be a proper part of instances of converting electricity to white light. A way to accommodate this supposition is to let the part-whole relation of functions be relative to the organisations in which these functions figure. So let us try the special criteria (4) and (6). With the special criteria (4) and (6) one avoids the above violation of the asymmetry condition (8) because with these criteria one merely has PPX(φ3,φ1) and PPY(φ1,φ3), with X given by Org(φ2,φ3,φ4,φ5,φ6) and Y given by Org(φ1,φ7). And since X and Y are different organisations, one cannot conclude that φ1 and φ3 are relative to one organisation simultaneously parts of each other. Yet, new complications follow when the two organisations X and Y are combined to form a new organisation Z of the functions φ2, φ3, φ4, φ5, φ6 and φ7 as given in figure 2. This organisation Z defines a decomposition Decomp(φ3,Org(φ2,φ3,φ4,φ5,φ6,φ7)) and, via relation (1), a composition Comp(Org(φ2,φ3,φ4,φ5,φ6,φ7),φ3). The special criteria (4) and (6) then give PPZ(φ3,φ3), which violates the irreflexivity postulate (7). Hence, the special criteria (4) and (6) seem also not to define a part-whole relation for functions that meets the minimal postulates of mereology.

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

40

P.E. Vermaas / Parts, Compositions and Decompositions of Functions in Engineering Ontologies

Figure 2. A second functional composition and decomposition of blue light.

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

5. Shedding Engineering Light An engineering response to the negative results derived so far could be that a too wide a class of functional compositions and decompositions is considered – if only compositions and decompositions are considered that make engineering sense, the examples in figure 2 may possibly be ruled out. So let us proceed with developing this response and see if a restriction to a class of engineering acceptable functional compositions and decompositions saves the possibility of defining a part-whole relation for functions by means of compositions and decompositions. A first observation is that with this restriction one can still argue that the general criteria (3) and (5) lead to a violation of the asymmetry condition (8) of mereology since the examples given in figure 1 seem acceptable functional compositions and decompositions: engineers may use them in, say, designing. But the examples given in figure 2 seem awkward from an engineering point of view: in the composition Comp(Org(φ2,φ3,φ4,φ5,φ6,φ7),φ3) the function φ3 of converting electricity to blue light is combined with other functions merely to compose itself; the decomposition Decomp(φ3,Org(φ2,φ3,φ4,φ5,φ6,φ7)) seems self-refuting since including φ3 in a description of how one can obtain this function suggests that more than a few corners may be cut. So, with a restriction to engineering compositions and decompositions the special criteria (4) and (6) may give us a part-whole relation for functions after all. It can now be argued that a composition like Comp(Org(φ2,φ3,φ4,φ5,φ6,φ7),φ3) is acceptable from an engineering point of view. If a light source is reverse engineered, and the functions φ2, φ3, φ4, φ5, φ6 and φ7 are identified with an organisation Org(φ2,φ3,φ4,φ5,φ6,φ7) as given in figure 2, then engineering reasoning does lead to the conclusion that this source has the overall function φ3. The engineer involved may reasonably note that s/he can think of another light source that can more straightforwardly realise this function, but this observation is not a reason for the engineer to reject the conclusion that the functions of the reverse-engineered more complicated light source compose the overall function φ3 of converting electricity to blue light. The special composition criterion (4) thus again leads to PPZ(φ3,φ3) and to a violation of postulate (7) so this criterion should still be ruled out. A reason why it is difficult to reject a functional composition from an engineering point of view is that a specific set of functions φ1, … φn with a specific organisation X composes to exactly one function Φ. A distinction between compositions

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

P.E. Vermaas / Parts, Compositions and Decompositions of Functions in Engineering Ontologies

41

Comp(OrgX(φ1,…φn),Φ) that are acceptable within engineering, and compositions Comp(OrgX(φ1,…φn),Φ') that are not, is therefore not available since Φ is identical to Φ'. This is different for functional decomposition. A specific function Φ may be decomposed in more than one set of functions with specific organisations, so one can distinguish between decompositions Decomp(Φ,OrgX(φ1,…φn)) that are acceptable from an engineering point of view and decompositions Decomp(Φ,OrgY(φ'1,…φ'm)) that should be rejected. In designing, for instance, one may reject the decomposition as given in figure 2 by arguing that if one includes the function φ3 in a functional organisation to exactly obtain that very function φ3, one more efficiently can select φ3 directly. So let us proceed in this way. Define for each function Φ the set of possible functional decompositions Decomp(Φ,Org(φ1,…φn)) that are in engineering acceptable and reject all other decompositions of Φ as unacceptable. How this set of engineering functional decompositions is defined is not analysed here, but at least decompositions Decomp(Φ,Org(φ1,…φn)) in which Φ is one of the decomposing functions φ1, … φn are not acceptable. One can then adopt the special decomposition criterion (6) for defining the part-whole relation for functions without falling prey to the argument that Decomp(φ3,Org(φ2,φ3,φ4,φ5,φ6,φ7)) as given in figure 2 leads to the conclusion that PPZ(φ3,φ3); the decomposition of figure 2 is not an engineering functional decomposition. A consequence of rejecting some decompositions is however that one also has to reject symmetry (2) between functional composition and decomposition. The composition Comp(Org(φ2,φ3,φ4,φ5,φ6,φ7),φ3) makes engineering sense, and gives with the symmetry relation (2) again the unacceptable decomposition Decomp(φ3,Org(φ2,φ3,φ4,φ5,φ6,φ7)). Symmetry is thus lost.

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

6. Further Asymmetry between Engineering Composition and Decomposition Functional decomposition may thus define a relation between functions that can be taken as a part-whole relation for those functions if some decompositions are rejected as unacceptable. And this rejection requires giving up symmetry (2) between functional composition and decomposition. In this section I show that this symmetry is lost in a second way: rejecting some decompositions in engineering requires also rejecting transitivity of functional decomposition, whereas functional composition may be transitive. Consider for instance a designer of an artefact that is to create blue light. This designer may decide to decompose this function φ3 in terms of a sequence of the functions φ1 and φ7 to produce white light and then to filter out blue light (figure 1b). This decomposition is acceptable. Then this designer may choose the cheapest lamp that produces white light, which is also an engineering acceptable choice. This lamp may however produce white light by mixing blue, red and yellow light. One may thus obtain an artefact in which the functions are related to one another by a sequence of functional decompositions that, individually and as a sequence, are acceptable in engineering; the observation that the function φ3 may be realised more directly, may in this case be overruled if using a lamp that only creates blue light is more expensive than using the existing cheap lamp for white light and adding a filter. Hence, the sequence of functional decompositions as given in figure 3 may sometimes be acceptable from an engineering point of view. To now avoid that this sequence

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

42

P.E. Vermaas / Parts, Compositions and Decompositions of Functions in Engineering Ontologies

amounts to the unacceptable decomposition Decomp(φ3,Org(φ2,φ3,φ4,φ5,φ6,φ7)), it has to be denied that functional decomposition is transitive. The compositions in this sequence may, however, be combined by transitivity, since that amounts to the composition Comp(Org(φ2,φ3,φ4,φ5,φ6,φ7),φ3) that cannot be rejected as unacceptable in engineering.

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

Figure 3. A functional composition and decomposition of blue light in two steps.

This argument might not be convincing since there may actually be cheap lamps that create blue light only. Yet other examples provide additional support that it may make engineering sense to obtain a function with limited output by a function that provides this output plus excess output. Consider, for instance, filters that limit Internet access, say for children. The function φ3 in figure 3 is then to download web pages with content that satisfies some conditions. This function may be realised not by directly downloading only the allowed web pages, but by downloading all pages – function φ1 – and then filtering for the allowed pages – function φ7. This downloading of all pages consists in turn of downloading the allowed pages – function φ3 – and the pages that are not – functions φ4, etc. Also if one wishes one specific computing function φ3 only – say, text editing – it typically may be cheaper to get a standard computer that realises φ3 and other computing functions φ4, etc., and then to filter for only the desired function φ3. A way to express the lack of transitivity of functional decomposition is by means of the organisation of the involved functions. A sequence of decompositions Decomp(Φ,OrgX(φ1,φ2,…φn)) and Decomp(φi,OrgYi(φi1,φi2,…φim(i))), with i=1,2,…n, combines to an engineering decomposition Decomp(Φ,OrgZ(φ11,φ12,…φnm(n))) only if the composite organisation OrgZ(φ11,φ12,…φnm(n)) is acceptable in engineering. Engineering analyses of functional decompositions (e.g., [1,3]) typically do not explicitly take the organisation of the composing functions into account, and thus do not provide explicit means to contrast engineering acceptable functional decompositions with decompositions that should be rejected. The notable exception is given by the work by Riichiro Mizoguchi and Yoshinobu Kitamura [17,19,23]. In that work the functional structure of artefacts is captured primarily by an is-achieved-by relation that – in the current notation – for a given function Φ, represents the sequence

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

P.E. Vermaas / Parts, Compositions and Decompositions of Functions in Engineering Ontologies

43

of subfunctions φ1, … φn that can achieve Φ. These is-achieved-by relations single out what I have called engineering functional decompositions. The upshot of the argument given in this section is thus that some of these relations may be combined to other isachieved-by relations, and that other is-achieved-by relations cannot be combined.5

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

7. Conclusion and Further Work In this paper I explored the possibility to introduce a generic part-whole relation for functions of technical artefacts by means of functional composition or functional decomposition. I proved by means of the postulates of mereology that general functional compositions and decompositions cannot define such a relation. I argued that from an engineering point of view one may reject a number of the general functional decompositions, and it was shown that decompositions that are acceptable in engineering might provide a basis for a proper part-whole relation for functions. In short the part-whole relation becomes then as follows: an instance of a type of function φi in a set {φ1,…φn} is a proper part of an instance of another function Φ if and only if instances of Φ actually decomposes into instances of the functions φ1, … φn. This possibility is available only if one (i) accepts that the part-whole relation is relative to the specific organisation of the functions {φ1,…φn} in decompositions, if one (ii) gives up on strict symmetry between functional composition and decomposition, and if one (iii) rejects transitivity of decomposition. The presented results show that the relations between functions of artefacts can be included in engineering ontologies by means of a part-whole relation. Yet it is fair to say that these results do not yet provide a full understanding of this relation. It still should be determined, for instance, whether the described part-whole relation meets other postulates of mereology beyond the three considered in this paper; this partwhole relation for functions can only then be categorised with the taxonomy of mereology relations as given in [30]. It can be proved [20] that the described relation can be taken as an extensional mereology by meeting also the strong supplementation postulate ([29], formula P.5). On the basis of these positive results, it may be expected that also sums and products ([29], §4) can be defined for the part-whole relation for functions given by the special decomposition criterion (6), turning it into a rather straightforward mereology relation.6 Another way of improving on the presented results is to capture the functions involved in the arguments not only in qualitative terms but also in quantitative ones. A first step in this direction is to add quantitative efficiencies for the functions given at the top of section 4. A choice that is compatible with the arguments presented is to assume that φ1 has an efficiency of 0,3 (meaning that 30% of the input electricity is actually converted to white light), that the other conversion functions φ3, φ4 and φ5 have an efficiency of 0,1, that the separate electricity and mixing light functions φ2 and φ6 have an efficiency of 1,0, and that the filter function φ7 has an efficiency of ⅓. 5

Private communication with Yoshinobu Kitamura, March 1, 2009. In [9] it is argued that a part-whole relation for functions induced by functional composition and decomposition may not satisfy Leśniewski’s standard mereology. From the current perspective this argument made use of the general criteria (3) and (5) without making distinctions between engineering acceptable and unacceptable compositions and decompositions. The part-whole relation proposed in this paper is defined by the special decomposition criterion (6), hence it may still satisfy Leśniewski’s mereology. 6

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

44

P.E. Vermaas / Parts, Compositions and Decompositions of Functions in Engineering Ontologies

A comparison with similar work has to be a step away from technical functions since to my knowledge not much other work has been done on part-whole relations defined directly on the level of functions. Similar work that comes to mind is on the mereology of processes as given in [31] since functions as considered in this paper may be seen as special types of processes. Furthermore, given that work has been done on the relations for functionally-defined structural parts and whole of artefacts (e.g., [1016]), it can be determined how these structural part-whole relations are related to the part-whole relation for functions. A preamble for this further work is a more extensive analysis of the part-whole relation for technical functions than given in this paper.7

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

References [1] G. Pahl, W. Beitz. Engineering Design: A Systematic Approach, Springer Berlin, 1996. [2] Y. Umeda, M. Ishii, M. Yoshioka, Y. Shimomura, T. Tomiyama, Supporting conceptual design based on the Function-Behavior-State-Modeler, Artificial Intelligence for Engineering Design, Analysis and Manufacturing 10 (1996), 275–288. [3] R. Stone, K. Wood. Development of a functional basis for design, Journal of Mechanical Design 122 (2000), 359–370. [4] A. Chakrabarti, T.P. Bligh, A scheme for functional reasoning in conceptual design, Design Studies 22 (2001), 493–517. [5] Y.-M. Deng, Function and behavior representation in conceptual mechanical design, Artificial Intelligence for Engineering Design, Analysis and Manufacturing 16 (2002), 343–362. [6] J. Bell, N. Snooke, C. Price, Functional decomposition for interpretation of model based simulation, in M. Hofbaur, B. Rinner, F. Wotawa (eds.), Proceedings of the 19th International Workshop on Qualitative Reasoning (QR-05), Graz, Austria, 18-20 May 2005, 2005, pp. 192–198. [7] B. Far, A. Halim Elamy, Functional reasoning theories: problems and perspectives, Artificial Intelligence for Engineering Design, Analysis and Manufacturing 19 (2005), 75–88. [8] D. Van Eck, D.A. McAdams, P.E. Vermaas, Functional decomposition in engineering: a survey, in 2007 ASME International Design Engineering Technical Conferences and Computers and Information in Engineering Conference DETC2007, September 4-7, 2007, Las Vegas, Nevada, 2007, DETC200734232. [9] P.E. Vermaas, P. Garbacz, Functional decompositions and mereology in engineering, in A.W.M. Meijers (ed.) Handbook Philosophy of Technology and Engineering Sciences, Elsevier Amsterdam, 2009, forthcoming. [10] P. Simons, C. Dement, Aspects of the mereology of artefacts, in R. Poli, P. Simons, (eds.) Formal Ontology, Kluwer Dordrecht, 1996, pp. 255–276. [11] I. Johansson. On the transitivity of the parthood relations, in H. Hochberg, K. Mulligan, (eds.) Relations and Predicates, Ontos Verlag, 2004, pp. 161–181. [12] L. Vieu, M. Aurnague. Part-of relations, functionality and dependence, in M. Aurnague, M. Hickmann, L. Vieu (eds.), Categorization of Spatial Entities in Language and Cognition, John Benjamins, 2005, pp. 483–509. [13] I. Johansson. Formal mereology and ordinary language: reply to Varzi, Applied Ontology 1 (2005/2006), 157–161. [14] A.C. Varzi. A note on the transitivity of parthood, Applied Ontology 1 (2005/2006), 141–146. [15] L. Vieu. On the transitivity of functional parthood, Applied Ontology 1 (2005/2006), 147–155. [16] P. Garbacz, A first order theory of functional parthood, Journal of Philosophical Logic 36 (2007), 309– 337. [17] Y. Kitamura, R. Mizoguchi, Ontology-based description of functional design knowledge and its use in a functional way server, Expert Systems with Applications 24 (2003), 153–166. [18] P.E. Vermaas, Promises of a philosophical analysis of technical functions, in M. Cristiani, R. Cuel (eds.) FOMI 2005: Proceedings of the first International Workshop Formal Ontologies Meet Industry, 2005.

7 I am grateful for comments by Stefano Borgo, Dingmar van Eck, Yoshinobu Kitamura, Riichiro Mizoguchi and two anonymous reviewers. Research for this paper was supported by the Netherlands Organisation for Scientific Research (NWO).

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

P.E. Vermaas / Parts, Compositions and Decompositions of Functions in Engineering Ontologies

45

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

[19] Y. Kitamura, Y. Koji, R. Mizoguchi, An ontological model of device function: industrial deployment and lessons learned, Applied Ontology 1 (2005/2006), 237–262. [20] P.E. Vermaas, The mereology of technical functions, 2009, Delft University of Technology manuscript. [21] M.S Erden., H. Komoto, T.J. Van Beek, V. D’Amelio, E. Echavarria, T. Tomiyama, A review of function modeling: approaches and applications, Artificial Intelligence for Engineering Design, Analysis and Manufacturing 22 (2008), 147–169. [22] M. Lind, Modeling goals and functions of complex plants, Applied Artificial Intelligence 8 (1994), 259283. [23] M. Sasajima, Y. Kitamura, M. Ikeda, R. Mizoguchi, FBRL: a function and behavior representation language, Proceedings of IJCAI-95 (1995), 1830–1836. [24] M. Modarres, S.W. Cheon, Function-centered modeling of engineering systems using the goal tree– success tree technique and functional primitives, Reliability Engineering and System Safety 64 (1999), 181–200. [25] V. Hubka, W.E. Eder, Functions revisited, International Conference on Engineering Design, ICED 01C586/102, Glasgow, Scotland, August 21-23, 2001, 69–76. [26] B. Chandrasekaran, J.R. Josephson, Function in device representation, Engineering with Computers 16 (2000), 162-177. [27] B. Chandrasekaran, Representing function: relating Functional Representation and Functional Modeling research streams, Artificial Intelligence for Engineering Design, Analysis and Manufacturing 19 (2005), 65–74. [28] P. Simons. Parts: A Study in Ontology. Clarendon, 1987. [29] A. Varzi, Mereology, The Stanford Encyclopedia of Philosophy (Summer 2003 Edition), E.N. Zalta (ed.), http://plato.stanford.edu/archives/sum2003/entries/mereology/. [30] C.M. Keet, A. Artale, Representing and reasoning over a taxonomy of part-whole relations, Applied Ontology 3 (2008), 91–110. [31] J. Seibt, Free process theory: towards a typology of occurings, Axiomathes 14 (2004), 23–55.

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

46

Formal Ontologies Meet Industry R. Ferrario and A. Oltramari (Eds.) IOS Press, 2009 © 2009 The authors and IOS Press. All rights reserved. doi:10.3233/978-1-60750-047-6-46

Towards An Ontology Infrastructure for Electromagnetism Alessandra ESPOSITO, Luciano TARRICONE, Laura VALLONE, Marco ZAPPATORE

University of Salento, Lecce, Italy

Abstract. Ontologies are more and more adopted to provide knowledge sharing and reuse, and to promote cooperation between several scientific fields. An increasing need of cooperation and resource sharing is emerging in the electromagnetic arena as well. The range of potential semantic-based applications seems to be wide and the knowledge to be codified is complex and diverse. Therefore the implementation of a well structured ontology in the electromagnetic domain is important. This paper describes a proposal for an electromagnetic ontology framework. The main design criteria were reusability and shareability, in order to make electromagnetic knowledge embeddable in larger and more general frameworks. The implementation of an ontology for aperture-antenna array design is provided as a specific example of a practical application.

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

Keywords. Ontology, electromagnetic, top level, domain, integration, knowledge.

Introduction Recently a strong emphasis has been put on collaboration among large groups, potentially geographically dispersed and heterogeneous: projects are often developed by teams which interact remotely by exploiting a tighter connectivity. Semantic-driven tools enable the integration and sharing of ideas, results and observations among participants who are dispersed and differently skilled. Such an opportunity is very appealing when the problems to be solved are complex and multidisciplinary: this is the case for many electromagnetic (EM) problems. Indeed, due to the complexity of EM problems, scientists could take a substantial advantage from the acquisition of knowledge about previously developed solutions. Unfortunately it often happens that existing solutions are either 1) poorly documented or 2) adequately documented in not widely shared knowledge sources. Therefore, an ordinate and intelligent repository of

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

A. Esposito et al. / Towards an Ontology Infrastructure for Electromagnetism

47

data, coming from scientific articles and studies, widely shared among the EM community, could be useful for EM researchers and professionals. Besides, there are a number of practical complex and multi-disciplinary problems may be afforded by composing available software codes. One of them is the computeraided design (CAD) of EM circuits and antennas [1] which is commonly solved by rigorously identifiable sequences of tasks, each deserving a variety of methods to be solved. Another meaningful example is the optimum planning of wireless networks [2], where the selection of the best suited radiopropagation model given the problem geographical area may be partially automated by exploiting an adequate knowledge base. All the above considerations lead to a single requirement: the codification of EM knowledge according to an agreed and sharable format, possibly expressed in a Web compliant language. In this paper, we propose an EM ontology framework, fully compatible with similar efforts in other research fields, and embeddable in larger and more general frameworks. The paper is structured as follows. Section 1 depicts the layered high level architecture of the ontology framework. Section 2, 3 and 4 are devoted to the description of the three layers of the framework. Section 5 draws some conclusions.

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

1. High Level Architecture An ontology can be defined as the formal specification of a vocabulary of concepts and the relationships among them in a specific domain [3]. Ontologies support cooperation among large distributed teams of domain experts. Therefore, the risk of terminological and conceptual ambiguities is high. One way to circumvent this is to agree on a foundational (also called top level) ontology [4], whose role is the provision of highlevel domain-independent categories, such as object, attribute, event, spatial and temporal connections, etc. Substantially, foundational ontologies provide a set of ontological entities to be reused when building new ontologies. This leads to a hierarchical view of ontologies, where top level ontologies are specialized into the so called lower level ontologies which reflect the practical point of view of application engineering. Often, when the domain is large and diverse, it is useful to introduce an intermediate level, which codifies terms being at the same time specific of the domain and reusable throughout all the lower level ontologies. This is the case of the EM domain, for which we have sketched a modular and layered design as follows (Figure 1): x a top level layer codifying higher level concepts; x a middle level layer representing concepts related to the EM scientific domain. This layer is further partitioned into two levels: a scientific layer including concepts common to scientific domains and an EM layer, containing terms specific of the EM domain; x a bottom level layer codifying concepts strictly related to single EM applications. As described in the following section, in order to enhance and validate interoperability and shareability of such a framework, the above mentioned layers were populated with ontologies taken from publicly available ontology repositories whenever possible.

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

48

A. Esposito et al. / Towards an Ontology Infrastructure for Electromagnetism

2. The Upper Level Ontologies

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

2.1. Upper Level Ontology Selection Currently, several upper ontologies exist [5], differing greatly in terms of user community, topical focus and logico-philosophical foundations. In order to choose the best suited for our framework, we made a selection based on several criteria, substantially centred around usability and applicability. Such criteria were based on Klein work [6] and on the IEEE Standard Upper Ontology Working Group proposal in [7], and focus on: x modelling style – we prefer simple models, being easily interpreted by technicians not expert in ontologies; x mapping with WordNet [8] lexicon, the de-facto standard in the linguistics world; x coverage/scope – adequate coverage of general terms useful for EM; x finding alignment – each time a term useful for EM was not found, we estimated the effort needed to extend available categories for including the missing term; x availability in the OWL format – as OWL [9] is the W3C standard for implementing ontologies exploitable through the Web; x reduced loading times; x modularity; x open source availability. According to the field trials we performed, Cyc [10] and SUMO [11] seemed appropriate ontologies, as they matched most of the listed requirements (such as OWL and open source availability, simplicity of the model). Even though Cyc featured the maximum level of coverage of EM concepts, other parameters (mapping with WordNet lexicon, loading times, easy alignment) definitely led to the choice of the SUMO ontology option.

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

A. Esposito et al. / Towards an Ontology Infrastructure for Electromagnetism

49

2.2. How we Reused SUMO

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

The Suggested Upper Merged Ontology (SUMO) is the most prominent proposal under consideration by the IEEE Standard Upper Ontology (SUO) working group [12]. It is an attempt to link categories and relations coming from different top-level ontologies in order to improve interoperability, communication and search in the Semantic Web area. A sketch of SUMO taxonomy is depicted in Figure 2. The topmost concept in SUMO is Entity, which is further split into Physical and Abstract. Physical entities are further divided into Objects and Processes. Other general topics include structural concepts, general types of objects and processes, abstractions (including set theory, attributes, number, measures, temporal concepts etc.). The taxonomy is large and includes many domain concepts, such Radiating and RadiatingElectromagnetic. Several concepts were reused in our framework, such as AngularDegree, ComplexNumber, FieldOfStudy, TemperatureMeasure, GeographicArea, Procedure, Engineering Component. Some of them will be recalled in the remaining part of the paper.

Figure 2: An excerpt from SUMO taxonomy.

SUMO is provided in the Standard Upper Ontology Knowledge Interchange Format (SUO-KIF) language, a variation and simplification of the Knowledge Interchange Format (KIF) [13], and in OWL-Full. The Web Ontology Language (OWL) is a family of knowledge representation languages for authoring ontologies, endorsed by the World Wide Web. It is available in three flavors: OWL-Full, OWL-DL and OWL-Lite, being OWL-Full the most expressive one and OWL-Lite the least one. OWL-DL was designed to provide the maximum expressiveness while retaining computational completeness and decidability. OWL-Full, instead, is not tractable. Moreover, every OWL-DL ontology can be considered a OWL-Full ontology as well, the reverse is not true. This is due to some differences between the two languages. For example, OWL-Full may treat a class simultaneously as a collection of individuals and as an individual in its own right. As OWL-DL is a good compromise between expressiveness and tractability, we adopted it for our framework. Such a decision implied some effort to convert OWL

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

50

A. Esposito et al. / Towards an Ontology Infrastructure for Electromagnetism

SUMO codification into OWL-DL [14]. Our intervention consisted substantially in modifying (patching) the ontology where 1) concepts were instances at the same time, 2) relations (properties) were instantiated between concepts, and 3) relations (properties) were modeled as concepts. The ontology was validated step by step by using OWL Validator [15]. Some of the modifications we performed on SUMO are shown in Table 1. Table 1 . Operations performed to convert SUMO to OWL-DL. Error Type

Patch Type

Elements involved

Missing Type Information (without a specific URI reference, classes can be seen as classes and instances at the same time)

Insertion of the correct URI reference

All the classes

Properties used as object properties without a specific declaration

Two strategies adopted: 1) substitution of the untyped property with available OWL DL constructs

9 Properties

2) cancellation (useless properties) Properties used as individual and as datatype property (i.e. the same datatype property is both given a type and used as a type)

Properties used as individual and as property

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

(i.e.the same object property is both given a type and used as a type)

Two strategies adopted: 1) rdf:type elimination 2) conversion of the datatype property into an object property and of the types into subproperty relations

28 Datatype Properties

Different strategies adopted: 1) rdf:type elimination 2) conversion of the types into subproperty relations

185 Object Properties

3) conversion of the types into OWL DL constructs

3. The Middle Level Ontologies 3.1. Scientific Layer Once that the top level ontology was chosen, the domain ontologies were defined. The design of domain ontologies followed a layered approach, based on reusing available intermediate ontologies when possible. To do this, we analyzed publicly available domain ontologies related to the scientific domain. Selection criteria were modularity and coverage of the EM domain. According to these criteria, SWEET ontologies [16] and the ontologies published by the Astronomical Department of the University of Maryland (UMD) [17] were the best suited. SWEET ontologies have an extensive set of concepts, related to earth and space sciences. They include terms defining earth realms, physical phenomena, physical processes properties, substances, time space etc. Unfortunately, the current stable version (1.1) has a poor coverage of the EM domain.

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

A. Esposito et al. / Towards an Ontology Infrastructure for Electromagnetism

51

The UMD ontologies (‘pre-International Virtual Observatory Association Thesaurus’ version), instead, have a quite good coverage of terms useful for the EM area (see Table 2) thus being amenable to fill in the “scientific domain layer” of our framework. Unfortunately, none of the examined domain level ontologies include a top level ontology. Therefore, we had to perform manually the integration of UMD ontologies with SUMO foundational ontology. Table 3 shows some of the integration efforts made during the framework implementation. Table 2. Some of the EM terms found in UMD ontologies

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

Physics.owl

EM_Absorption, EM_Emission, EM_Transition, EMradiation, ElectricField, FarInfrared, FarUV, Field, Frequency, IndexOfRefraction, Infrared, MagneticField, Microwave, MillimiterWave, Wavenumber Microwave, SpectralMeasurement, Wave, WaveMeasurement, Wavelength

Instrument. owl

DispersionDevice, EMInstrument, EMresolution, ElectricFieldInstrument, FieldOfView, MagneticFieldInstrument, Magnetometer, Polarimeter, Resolution, Spectrometer, SpectralTransmittance, SpatialResolution

Process.owl

Some EM Relevant Concepts

Absorption, Adsorption, Attenuation, Backscatter, Conduction, Dispersion, Emission, Excitation, Feeding, MediumWaveInteractionProtocol, RadiativeTransfer, Reflection, Scattering, Transmission, WaveAbsorption

3.2. EM domain layer In order to provide a codification of EM concepts, UMD ontologies required a further specialization, thus giving birth to the “EM domain ontology layer” . EM domain ontologies were designed according to a modular structure, based on a number of design criteria: 1. modules should coincide with well-known EM subdomains, e.g. compatibility vs. remote sensing; 2. module design should follow the same principles as the two higher levels, paying attention to usability, modelling style, etc.. 3. the size of modules should not be huge in order to facilitate their loading and processing; 4. modules covering neighbouring sub-domains should exhibit a limited (and documented) degree of overlap.

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

52

A. Esposito et al. / Towards an Ontology Infrastructure for Electromagnetism

Table 3. Some of the operations having been performed to carry out the integration of UMD with SUMO. Class

UMD Ontology

New position in OntoCEM

Creation

Science.owl

Specializes the “Artifact” class of SUMO

Location

Science.owl

Specializes the “Region” class of SUMO

FieldOfStudy

Science.owl

Event

Science.owl

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

TemperatureScale Science.owl MathEntity

Science.owl

PhysicsEntity

Physics.owl

Process

Process.owl

Equivalent to SUMO FieldOfStudy” class Specializes the “Proposition” class of SUMO Specializes the “Process” class of SUMO Equivalent to SUMO “TemperatureMeasure” class Specializes SUMO “ConstantQuantity” class Specializes the “Abstract” class of SUMO Specializes the “Abstract” class of SUMO Equivalent to the “UnitOfMeasure” class of SUMO Equivalent to the “Process” class of SUMO Specializes the “Physical” class of SUMO

GeometricalEntity Geometry.owl

Specializes the “Abstract” class of SUMO

Operation

Operation.owl

Specializes the “Calculation” class of “Science.owl” UMD ontology

Analysis

Operation.owl

Specializes the “MathEntity” class of “Science.owl” UMD ontology

InstrumentEntity

Instruments.owl

Specializes the “Abstract” class of SUMO

Domain ontologies were based on the analysis of several knowledge sources and indicators as well: x EM experts interviews; x topics and committees adopted in leading EM conferences and books; x organization of Societies, Councils and Technical Communities active in the EM area [18]; x IEEE suggested list of keywords for papers [19]. The result was the EM ontology structure shown in Figure 3. Each module codifies an EM sub-domain, except for the EM_Spectrum ontology which was defined in order to collect terms common to several sub-domains (criterion 4).

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

SC IE D O N TI MA FIC IN

A. Esposito et al. / Towards an Ontology Infrastructure for Electromagnetism

Units.owl

Operation.owl

N AI DO M

Instruments.owl

Physics.owl Geometry.owl PhysicalTheory.owl

Statistics.owl

EM

53

UnitsInstances.owl

EM_Compatibility.owl EM_Spectrum.owl RemoteSensing.owl

Quantity.owl

Antenna.owl

EM_Analysis.owl

MW_Devices.owl

EM_Propagation.owl EM_Measurement.owl

Figure 3. Middle level partitioned into scientific and EM domain layers. The scientific domain layer is populated with UMD ontologies, whilst the EM domain layer contains ontologies strictly related to the EM domain.

4. Bottom Level Ontologies The objective of this section is to provide an example of a bottom level ontology. More in detail, it briefly describes the ontology codifying the problem of “design of apertureantenna arrays”. This ontology was obtained by specializing concepts from ontologies belonging to upper levels. Specifically, it reuses concepts from two EM domain level ontologies, namely the “EM_Analysis” and the “Antenna” ontologies, which, in turn, reuse concepts belonging to SUMO top level ontology and to the UMD scientific ontology. In order to illustrate this, the section is partitioned into three subsections, respectively describing the “EM_Analysis”, the “Antenna”, and the “design of aperture-antenna arrays” ontology.

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

4.1. EM_Analysis domain ontology The “EM_Analysis” middle level ontology provides a classification of EM analysis methods, based on two well known analysis method classifications. The former was proposed by Sadiku in [20], the latter has been introduced by Stutzman and Thiele in [21]. Such classifications divide the analysis methods into two main categories: analytical and numerical. Analytical methods give an exact, mathematical solution of the problem. Numerical methods adopt discretization techniques which simplify the problem to a finite number of unknowns. Analytical methods are partitioned into separation of variables, series expansion, conformal mapping and integral methods. Numerical methods, instead, are divided into three main sub-classes depending on the working frequency. According to [21] high frequency and microwave methods have been specialized as well. The ontology (Figure 4) is integrated inside the OntoCEM framework by simply specializing the “Procedure” class of the SUMO top level ontology with an “AnalysisMethod” class.

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

54

A. Esposito et al. / Towards an Ontology Infrastructure for Electromagnetism Top Level Class EM_Analysis Ontology Class

SeparationOfVariables Procedure

SeriesExpansion AnalyticalMethod

Is-A

IntegralMethod AnalysisMethod ConformalMapping NumericalMethod LowFrequencyMethod MicrowaveNumericalMethod TimeDomainMethod FrequencyDomainMethod IntegralEquationsMethod DifferentialEquationsMethod HighFrequencyMethod CurrentBaseMethod FieldBaseMethod

Figure 4. A fragment of the EM_Analysis ontology.

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

4.2. Antenna Domain Ontology A small fragment of the Antenna ontology is depicted in Figure 5. This ontology is centred around the concept of “Antenna”, standing for any device used to receive or radiate electromagnetic waves. The “Antenna” class has been defined by extending the “EngineeringComponent” class of the SUMO top-level ontology. Based on the typology of the radiating element, antennas are partitioned in different categories, such as aperture antennas, loop antennas, slot antennas, etc. For instance, aperture antennas radiate by means of an aperture. To express such a concept, the “Aperture” class of the Instrument.owl UMD ontology has been used. 4.3. The Bottom Level: “Design of Aperture-Antenna Arrays”Ontology This section shows how the conceptualization of the ontologies belonging to higher levels has been reused to codify a focused application area, the analysis and design of aperture-antenna arrays. The analysis and design of aperture-antenna arrays starts from the study of the physical elementary structure (i.e. the geometry of the single radiator, in our example the aperture-antenna) and produces a full insight into the properties of the whole device (the aperture-antenna array). An effective method to attack the problem is its partitioning into simpler units [22]: 1) analysis and design of the single radiator 2) analysis and design of the apertures 3) analysis of the overall behaviour of the system and finally 4) analysis of the radiating properties of the array. The bottom level ontology conceptualizes the four units above mentioned. Figure 6 illustrates the codification of methodologies suited for the second unit, namely, the analysis and design of the apertures. Figure 7, instead, shows how the analysis and

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

A. Esposito et al. / Towards an Ontology Infrastructure for Electromagnetism

55

design of aperture-antennas reuses concepts belonging to higher level ontologies, such as the previously described EM_Analysis and Antenna ontologies.

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

Figure 5. A fragment of the Antenna ontology. The “Antenna” class specializes the SUMO “EngineeringComponent” class, whilst “ApertureAntenna” is linked to the “Aperture” UMD class by the “radiatesBy” property.

Figure 6. EM_Analysis domain ontology concepts are specialized in the CAD bottom level ontology to codify analytical methods specifically related to the application.

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

56

A. Esposito et al. / Towards an Ontology Infrastructure for Electromagnetism

Figure 7. A fragment of the CAD bottom level ontology codifying the concepts “the analysis and design of aperture-antennas is performed on aperture-antennas” and “the analysis and design of aperture-antennas is solved by spectral or space domain methods”.

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

5. Conclusions This paper presented a proposal for an EM ontology framework. The proposal had the objective of maximizing knowledge reusability and integrability. Therefore, a modular, layered architecture was designed, based on a robust sharable conceptualization, given by a publicly available well known top level ontology. The domain vocabulary was codified in a layered fashion as well. General scientific concepts were codified by exploiting a publicly available knowledge representation. EM formalization was designed by mediating between several well known EM sources. The paper, of course, shows results coming from a start-up phase of a long-term, heavy and ambitious task, which requires 1) a large involvement of the EM research community in order to render the knowledge representation adequately shared and agreed, and 2) a continuous monitoring of the state-of-the-art of the Semantic Web, in order to keep the framework aligned with up-to-date shared knowledge infrastructures.

References [1] [2] [3] [4]

L. Tarricone et al. A Parallel Framework for the analysis of metal-flanged Rectangular aperture Arrays. IEEE Trans. on Ant. &Prop. (2001), 1479-1484. F. Mori, R. Sorrentino, M. Strappini, and L. Tarricone, “A Genetic GIS-based approach for the Optimization of Radiobase-station Sizing and Location”, Proc. of EMC2002, Vol. 1, pp. 467–471, 2002. R. Gruber. A translation approach to portable ontologies. Knowledge Acquisition, 5(2):199-220, 1993. N. Guarino. Formal Ontology and Information Systems. In N. Guarino, editor, Proc. of the 1st Int. Conference on Formal Ontologies in Information Systems, FOIS'98, Trento, Italy, pages 3-- 15. IOS Press, June 1998.

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

A. Esposito et al. / Towards an Ontology Infrastructure for Electromagnetism

[5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17]

V. Mascardi, V. Cordi, P. Rosso. "A Comparison of Upper Ontologies". 2006 (http://woa07.disi.unige.it/papers/mascardi.pdf) M. Klein, Combining and Relating Ontologies: Problems and Solutions. IJCAI Workshop on Ontologies, Seattle, 2001. http://suo.ieee/org/SUO/Evaluations/evaluation_questions.htm A. Gangemi et al.. “Conceptual Analysis of Lexical Taxonomies: The Case of WordNet Top-Level”. In: Proc, of the Int. Conf. on Formal Ontology in Information Systems FOIS 2001 Ogunquit, Maine, Oct. 17-19, 2001. http://www.w3.org/TR/2002/WD-owl-ref-20020729/ J. Curtis, D. Baxter, J. Cabral. "On the Application of the Cyc ontology to Word Sense Disambiguation". Proc. of the 19th Int. Florida Artificial Intelligence Research Society Conf., pp.652657, 2006 Niles, I., and Pease, A. Towards a Standard Upper Ontology. In Proc. of the 2nd Int. Conf. on Formal Ontology in Information Systems (FOIS-2001), Ogunquit, Maine, October 17-19, 2001. http://suo.ieee.org http://logic.stanford.edu/kif/dpans.html S. Bechhofer and R. Volz. Patching Syntax in OWL Ontologies. In Proc. of 3rd Int. Semantic Web Conf. (ISWC’04), Hiroshima, Japan, 2004. http://www.mygrid.org.uk/OWL/Validator http://sweet.jpl.nasa.gov/ontology B. Thomas and E. Shaya, ”A User Interface for Semantically Oriented Data Mining of Astronomy Repositories”, Astr. Data Analysis Software and Systems, ASP Conf. Series, Vol. XXX, 2008. http://www.ieee.org/web/societies/home/index.html www.ieee.org/organizations/pubs/ani_prod/keywrd98.txt N. Matthew, O. Sadiku, Numerical Techniques in Electromagnetics, Second Edition - CRC, 2001 W.L. Stutzman, G.A. Thiele, Antenna theory and design, Chp.10, Wiley & sons Inc., 1998. A. Esposito, L. Tarricone, L. Vallone, M. Vallone, "Cooperative Computer Aided Engineering of Antenna Arrays", in Advances in Information Technologies for Electromagnetics, A.Esposito, L.Tarricone,Eds. Springer, (2006).

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

[18] [19] [20] [21] [22]

57

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

58

Formal Ontologies Meet Industry R. Ferrario and A. Oltramari (Eds.) IOS Press, 2009 © 2009 The authors and IOS Press. All rights reserved. doi:10.3233/978-1-60750-047-6-58

Ontological Representation for Algerian Enterprise Modeling Sabria HADJ TAYEB 1 and Myriam NOUREDDINE Department of Computer Science, University of Science and Technology of Oran, BP 1505 El M’Naouer, Oran, Algeria

Abstract. The Algerian enterprises are faced to the international industrial and economic competition which requires a complete reorganization of their structure. The modeling concept can satisfy this need and led to many techniques. However, none of them is complete and allows modeling all the aspects of a system. From a theoretical study and real cases of Algerian enterprises, we developed a knowledge framework through a list of criteria related to characteristics of enterprise modeling. In this context, this article deals with a knowledge representation by a domain ontology realized with the Protégé tool. Keywords. Modeling, Algerian enterprise, criteria, ontology, Protégé

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

Introduction Currently, the Algerian enterprises must reorganize their structures to improve their internal and external functioning. These enterprises must have a framework for modeling to describe the structure and dynamics that modify this structure. We have seen in practical, that Algerian enterprises don’t use a method or in the best case, they use the MERISE method, old but well known in information systems. In recent years, the modeling concept has become fundamental. Structured techniques and related tools have emerged to carry out a project of industrial reorganization, to improve the institutional and operational processes and to assess their performance. Our goal is to provide to managers of Algerian enterprises an overview of techniques for enterprise modeling, with its characteristics. To achieve this aim, we develop a knowledge framework through a list of criteria in order to observe a wide overview of characteristics on techniques. This article deals with a formalization of this knowledge representation following ontological domain of different modeling techniques and their criteria. The second section proposes the concept of enterprise modeling and modeling techniques. In the next section, we present the development of our model through the approach description and the list of criteria identified. The construction of the ontology using the Protégé tool is described and a validation of our model is given by a real case of an Algerian enterprise. 1

Sabria Hadj Tayeb: Department of Computer Science, University of Science and Technology of Oran, BP 1505 El M’Naouer, Oran Algeria; E-mail: [email protected].

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

S. Hadj Tayeb and M. Noureddine / Ontological Representation for Algerian Enterprise Modeling

59

1. Criteria for the Enterprise Modeling Techniques 1.1. Techniques of Modeling Enterprise There are several definitions of the enterprise modeling and we hold that the enterprise modeling is the representation of the structure and operation of the enterprise to improve its performance. Each approach of enterprise modeling is based on scientific concepts in order to improve performance of enterprise. We are interested in this paper in some of these approaches as methods [8] MERISE and GRAI, CIMOSA, GERAM and PERA architectures. We group methods and architectures under the generic term of technique. Those techniques have been chosen for the following reasons [6]: MERISE is a classic technique issue of information systems. It is widely used in our Algerian enterprises. PERA and GRAI are used for the construction of ISO 15704. CIMOSA and GERAM represent the basis of ENV 40003 and ISO 15704 and they are issues of ENV 12204. We present briefly those modeling techniques: • • •

•

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

•

MERISE: The MERISE method provides both of process, models, formalisms and standards for the design and implementation of information systems. GRAI: The GRAI method [3], [9] was developed in the laboratory GRAI3 of the University of Bordeaux (France). Its objective is the modeling of decision aspects in the enterprise during the analysis or design phases. CIMOSA: The architecture CIMOSA is considered as one of the modeling approaches that received the most research works. Its purpose is to define precisely the objectives of the enterprise and manufacturing strategies and to manage the system in a context of perpetual change. PERA: PERA is a complete architecture engineering environments developed by the Purdue Laboratory of Applied Industrial Control in the United States. Its purpose is the design of large systems. GERAM: GERAM [2] is reference architecture generalizes CIMOSA, GRAI, PERA and other architectures to represent the whole enterprise.

1.2. Identification of Criteria Each modeling technique has a corporate name and has particular and general characteristics. We group these features under the term criteria. From the study of modeling techniques and actual cases of Algerian enterprises, we extracted a list of criteria that we have grouped into five groups of criteria, allowed a hierarchical structure of criteria. We obtain the family F = { f 1, f 2, f 3, f 4, f 5} where f i is the meta criterion standard, with: 1. 2. 3. 4. 5.

f 1 = { f 11, f 12, f 13, f 14} lists the criteria for the model. f 2 = { f 21, f 22, f 23} is said general criterion. f 3 = { f 31, f 32} makes reference to the structure. f 4 = { f 41} contains the criteria for the resources. f 5 = { f 51, f 52, f 53, f 54 } includes different views.

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

60

S. Hadj Tayeb and M. Noureddine / Ontological Representation for Algerian Enterprise Modeling

Table 1. Evaluation of each criterion Criterion

Value

Generic model Formalism

Yes, no, unknown Graphical and textual, graphical All, part Yes, no, unknown Easy, difficult, , affordable Easy, difficult, , affordable Yes, no, part Yes, no, part Yes, no, unknown Yes, no Yes, no, unknown Yes, no, unknown Yes, no, unknown Yes, no, unknown

Lifecycle Software support Learning Usability Time Function decision Flow decision Human resource Functional view Organizational view Resource view informational view

We identified fourteen criteria: • • • • • • •

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

• • •

Generic model (f 11), if the model is applicable to a wide type of enterprise. Formalism (f 12), if the formalism is adaptable to enterprise employees. Lifecycle (f 13), supported by the technique. Software support (f 14), represents the generation or not of support to facilitate the implementation of the model. Learning (f 21); it reflects the control or not of the technique. Usability (f 22), it is the ability to assimilate the technique of enterprise modeling. Time (f 23), this criterion is used to describe the properties of states and their changes over time. Function (f 31) and flow (f 32) decision, these criteria are necessary for decision making. Human resource (f 41), this criterion is chosen for the inclusion or otherwise of the human aspect, it is necessary to describe the roles, responsibilities and knowledge of human actors in the production process. Functional (f 51), organizational (f 52), resources (f 53) and informational views (f 54), this set of criteria representing the types of views offered by technique.

Each criterion f ij refers to the criterion j belonging to the family i and is assigned to specific values and the following table (Table 1) presents the evaluations of each criterion.

2. Ontological Representation of Modeling Techniques 2.1. Proposed Approach The proposed approach is described through the approach given in the Fig. 1.

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

S. Hadj Tayeb and M. Noureddine / Ontological Representation for Algerian Enterprise Modeling

61

Figure 1. Proposed approach.

To represent the knowledge criteria in a formal manner, we construct an ontology [4] using the Protégé tool. This ontology will enable us to develop a system to aid in choosing a technique for modeling enterprise. The choice of technique is based on the outranking method PROMETHEE II [1] and the proposed model, through its implementation has led to a result in software with a graph as a histogram.

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

2.2. The Ontology Building To formally represent and exploit our knowledge constituting the different techniques of enterprise modeling and their characteristics to choose the best of enterprise needs, we opted for the ontology. Several definitions have been proposed for the concept of ontology; the most commonly cited is of Gruber “An ontology is an explicit specification of a conceptualization” [10]. For building ontologies, it is often useful to adopt techniques and formal tools that ensure a better use. We adopted the Protégé. Protégé release 3.1 [5] is a java tool, it is produced and made available by the Stanford Medical Informatics laboratory. We summarize in the next section, the stages of creation of our ontology. •

Creating concepts: The first step is the creation of concepts. Recording to the Section 2.2, a technical term (T ) and a group of criterion (F ) will be translated by creating of two concepts at the same level. Each meta criterion f i will be represented by a concept of the meta concept F. Figure 2 illustrates this hierarchy.

The name of the term f ij represents the respective property of T and fi. For example, to create the attribute learning (f 21), we must specify the concept of membership (f 2) in a Domain and define the possible values for this attribute in Allowed values. The Fig. 3 illustrates this example.

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

62

S. Hadj Tayeb and M. Noureddine / Ontological Representation for Algerian Enterprise Modeling

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

Figure 2. Hierarchy of concepts.

Figure 3. Creation of attributes.

•

Creation of forums: The exploitation of the ontology will be done by giving values to any attribute. The Fig. 4 presents the instance of CIMOSA with the meta model (f 1): The study of the technique has allowed assigning the following attributes f 1j: Generic model (f 11): Yes; Formalism (f 12): graphics and text; Lifecycle (f 13): all; Support software (f 14): yes;

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

S. Hadj Tayeb and M. Noureddine / Ontological Representation for Algerian Enterprise Modeling

63

Figure 4. Creation of forums.

Once the ontology created, we generated HTML documents that we used to develop a database of the criteria and all the techniques of enterprise modeling. These data will be exploited in the multi-criteria analysis.

3. Experimentation and Result The validation was made on an example of Algerian enterprise.

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

3.1. Specification of the Enterprise We have choice an local Algerian enterprise [7] which is dedicated to the manufacture of glass objects in particular bottles. The manufacture of these objects passes through four stages: • • • •

Composition step: It is a mixture of different components (sand, limestone…). This mixture is then stuffs. This step is done automatically with computerguided machines. Fusion step: A liquid obtained is transmitted in form of drops to machines connected to the furnace. These drops are blown to take the form of a mold provided in advance. Shaping step: This step is done automatically by machines to get the final form of the bottle. Annealing step: It allows lowering the tension of the final product.

The product undergoes a test for quality control automatically (machines Guided by computer) or visually (by workers qualified). In case of doubt on human labor, the product is transferred for automatic control.

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

64

S. Hadj Tayeb and M. Noureddine / Ontological Representation for Algerian Enterprise Modeling

Figure 5. Graph for the choice of technique.

3.2. Criteria for the Enterprise Following a dialogue with the patrons, we were able to identify nine criteria: Time, functional, informational view, organizational and resource views, formalism, lifecycle and support software. These criteria will be support for multicriteria decision analysis.

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

3.3. Final Result We adopt the outranking method PROMETHEE II [1] to choice a technique for the enterprise modeling. Taking into account the above criteria, PERA is the most appropriate technique for the modeling of this enterprise and MERISE ranks fourth (Fig. 5). PERA methodology is most appropriate for modeling ALVER as it defines all stages of an industry (production of bottles). In addition, it really takes into account the human aspect and places it clearly in the architecture. The enterprise is already automated, so MERISE ranked fourth.

Conclusion To formally represent our knowledge, we created ontology of the various modeling techniques and list of criteria using the Protégé tool. We generated an HTML document to develop a decision aid system based on a multicriteria approach. The system offers to heads of Algerian enterprise an overview of techniques provided with their characteristics. A dialogue will be done between the managers, who will express the needs of their enterprises and researchers who have scientific knowledge about techniques. It will then be up to the bosses to choose the modeling technique suitable close to the expressed requirements. We believe that this approach represents a first step towards an effective reorganization of the enterprise leading the development of industrial production.

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

S. Hadj Tayeb and M. Noureddine / Ontological Representation for Algerian Enterprise Modeling

65

References

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

[1] A. Schärlig, Pratiquer Electre et Prométhée, un complément à Décider sur plusieurs critères. Presses polytechniques et universitaires romandes (1996), Lausanne (Suisse). [2] B. Vallespir, L’intégration en modélisation d’entreprise: les chemins d’U.E.M.L, 4e Conférence Francophone de Modélisation et Simulation MOSIM’03, 140-145, France. [3] C. Tahon, Evaluation des performances des systèmes de production, Lavoisier, Paris, 2003. [4] D. Guinness, An Environment for Merging and Testing Large Ontologies, 7th International Conference on Principles of Knowledge and Reasoning KR2000 Breckenridge (2000), Colorado (USA). [5] H. Knublauch, The Protégé OWL Plugin: An Open Development Environment for Semantic Web Applications, 3rd International Semantic Web Conference ISWC2004 (2007), Hiroshima (Japan). [6] J. Vieille, Intégration Production – Entreprise : La norme ANSI/ISA S95, 10èmes journées des CPIM de France -Certified In Production and Inventory Management (2000), p. 17. [7] M. Noureddine, S. Ferhane, Proposition d’un environnement de modélisation pour les systèmes de production, 3ème Conférence Internationale sur la Productique (CIP’05), CD-Rom Art.73 (2005), p. 8, Tlemcen (Algeria). [8] M. Noureddine, S. Hadj-Tayeb, Contribution à la modélisation d’entreprise algériennes- état de l’art, Colloque national sur la synergie université- entreprise (2008), p. 7, Ouargla (Algeria). [9] P. Girard, Engineering: a method to model, design and run engineering design departments. International journal of Computer Integrated Manufacturing, 2004, 716-732. [10] T.R. Gruber, Toward Principles for the Design of onthologies Used for Knowledge Sharing, International journal Human Computer Studies, vol. 43 (1993), 907-928.

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

66

Formal Ontologies Meet Industry R. Ferrario and A. Oltramari (Eds.) IOS Press, 2009 © 2009 The authors and IOS Press. All rights reserved. doi:10.3233/978-1-60750-047-6-66

Modular Ontologies for Architectural Design Joana HOIS, Mehul BHATT, and Oliver KUTZ SFB/TR 8 Spatial Cognition University of Bremen, Germany {joana, bhatt, okutz}@informatik.uni-bremen.de Abstract. Designs of architectural environments have to take into account various sources of heterogeneous information. Not only quantitative spatial constraints and qualitative relations but also functionally-dependent and abstract conceptualizations are relevant aspects for an architectural design. We aim at a modular ontological approach based on the theory of E-connections to formally present and bring together these different perspectives on the domain. Modularity here allows a flexible integration of the various sources while keeping their thematically different aspects apart. We show how modular ontologies reflect the domain for architectural design and how they can be applied. Keywords. Ontologies, Modularization, Heterogeneous Information, Architecture

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

Introduction Architecture design tools are primarily concerned with the ability to develop models of spatial structures at different levels of granularity. They range, for example, from lowfidelity planar layouts to complex high-resolution 3D models that accurately reflect the end-product, the building plan. For instance, using a CAD tool to design floor plans for office buildings, one may model various spatial elements representing doors, windows, rooms, etc. These elements are based on primitive geometric entities that collectively reflect the building plan. However, such an approach using contemporary design tools lacks the capability to incorporate the semantic content associated with the structural elements that characterize the model [3]. Furthermore, and partly as a consequence, these tools also lack the ability to exploit the cognitive expertise that a designer is equipped with. Our approach utilizes ontologies to incorporate semantics at different layers, e.g., qualitative abstractions of quantitatively modeled designs. In this paper, we propose an approach that enables design tools to represent information about architectural environments by combining different perspectives on the domain. In particular, we endeavor to formalize these different and heterogeneous aspects involved in the architectural design process by applying methods from ontological modularity. Here, we address the various aspects described from thematically different perspectives that provide a particular view on the domain [20] given by different ontologies. In the case of architectural design systems, ontologies can particularly provide a formalization of their domain-specific views.

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

J. Hois et al. / Modular Ontologies for Architectural Design

67

One major topic in the field of formal ontology developments concerns ontological modularity and ontology mapping [24]. We pursue this modularity strategy by taking into account the content of ontologies. Here, modular ontologies support design clarity by specifying the different perspectives on a domain. i.e., each modularly-designed ontology provides the semantics for a particular view. As a result, the different ontologies can be connected by integration and can interact in a meaningful way. In essence, we apply modularity inspired by the theory of E-connections to modular perspectives on architectural design. The ontologies are not only separated into modules logically, but they also reflect different thematic modules. They are applied to industrial standards for architectural design and tools, which are introduced in the next section. Section 2 gives an overview of modularization in ontological engineering. Our contribution to architecture-specific modular ontologies is presented in Section 3, followed by an application scenario for architectural design in ambient intelligence in Section 4.

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

1. Architectural Design Tasks As the design of specialized spatial environments, such as smart buildings and homes, starts to become mainstream and economically viable for a larger consumer base, it is expected that architecture design projects involving the design and implementation of such environments will adopt a radically different approach involving the use of formal knowledge representation and reasoning techniques [2]. It is envisioned that a smart environment will be designed from the initial stages to aid and complement the requirements that characterize its anticipated functional or intelligent behavior [1]. A crucial element that is missing in architectural design pertains to formal modeling, i.e., representation and reasoning of architectural structures. Indeed, as all architectural design tasks pertain to a spatial environment, formal representation and reasoning along terminological and spatial dimensions can be a useful way to ensure that the designed model satisfies key requirements that enable and facilitate its intended form and function. Among other things, we propose the ontological modularization as being a key aspect to operationalize this design approach in the initial architectural modeling phase. We demonstrate our approach in the context of an industrial standard for data representation and interchange in the architectural domain, namely the Industry Foundation Classes (IFC) [6]. This data exchange format is based on object classes, such as IfcDoor, IfcRoof, or IfcShapeAspect. The classes provide information on metrical data, relations, and dependencies between classes. While 2D or 3D CAD models are based merely on metrical data for geometric primitives, e.g., points and lines, IFC’s object classes add semantics to these primitives, i.e., object classes are not only defined by their metrics but also by their inherent relationships to other classes. Section 3 presents the way IFC’s data model is integrated into our modular approach.

2. Ontological Modularity Modularity has become a key issue in ontology engineering. Research into aspects of modularity in ontologies covers a wide spectrum. [24] give a good overview of the breadth of this field. Three orthogonal questions define the research area of modularity for ontologies:

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

68

J. Hois et al. / Modular Ontologies for Architectural Design

• How can large and complex ontologies be built up from parts, possibly being formulated in different logical languages, and in what ways can those parts be related? (modular combination problem) Conversely, given a large ontology, how can we decompose it into ‘meaningful’ modules? (modularization problem) • How can the structure of a modular ontology be represented, and how can various logical (or structural, topical) properties of the parts (modules) be preserved? • How can we perform (automated) logical reasoning over such structured ontologies, and how, or when, can we reduce reasoning in the overall ontology to the ontology’s component modules? The main research question is how to define the notion of module and how to re-use such modules. We briefly summarize some of these aspects and present details relevant to the notion of module, modularity, and reasoning problems. We finally outline the kinds of modules and modular reasoning problems that we encounter in architectural design. 2.1. The Dimensions of Ontological Modularity and Formal Reasoning Main dimensions of ontological modularity and respective (automated) reasoning challenges are:1

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

The Language Layer and Semantic Heterogeneity: Whenever we want to combine two ontologies (or formal theories), we run into the problem of syntactic and semantic heterogeneity. Indeed, even if we stay in the same formal logic, we run into the problem of reconciling the joint vocabulary of the ontologies. The most general solution to this problem is to provide a family of logic translations that allows to seamlessly move from one logic to another along the translation, based on a general definition of logic and logic translation as provided by institution theory [7]. Structuring, Extension, and Refinement: The mere size of ontologies can make the design process quite hard and error prone (at least for humans). This issue has been only partly cured in OWL by the imports construct, which essentially copies the axioms of one ontology into another. Natural operations are, for instance, union, intersection, ‘hiding’ certain symbols, and extension. The semantics, however, of such operations is in general non-trivial. Methods developed for (algebraic) specification, for instance, can be applied to ontology engineering, as they provide systematic structuring techniques [16]. Apart from such structuring concepts, another natural relationship between ontologies is that of a refinement: O2 refines O1 if all of O1 ’ s axioms are entailed by O2 (possibly along a translation). Essentially, this means that we need to provide a theory interpretation of O1 into O2 [15]. Another kind of ‘extension’ is provided by the idea of concrete domains. They extend an ontology language by constructs that allow to ‘import’ computations in specific structures, such as the natural numbers or time intervals [11]. Logical Independence: One of the most important logical concepts of modularity is given by the notion of conservativity. An ontology O2 is a conservative extension of O1 if all assertions made in the language of O1 that follow from O2 already follow from O1 . Essentially, this means that O1 completely and independently specifies its vocabulary, with respect to O2 . This concept can, for instance, be used to extract logically independent modules from a large ontology. While this notion of module therefore is important, 1 This

is, of course, not an exhaustive analysis.

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

J. Hois et al. / Modular Ontologies for Architectural Design

69

it is also computationally difficult. Although proving conservativity is undecidable for first-order logic and many expressive description logics (DL) [24], there are general algorithmic solutions for less expressive DLs [15]. The simplest case of a conservative extension is a definitonal extension, as it extends the vocabulary of an ontology O by new terms, whose meaning is entirely determined by the axioms given in O.

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

Matching and Alignment: Matching [5] and aligning [25] ontologies focus on the identification of (thematically) overlapping parts of two ontologies (matching problem) and on systematically relating terms across ontologies that have been identified as, for instance, synonymous (alignment problem). As opposed to structuring and conservativity, such relationships are often established by using statistical methods and heuristics, employing, for instance, similarity measures and probabilities. Integration and Connection Informally, an integration of two ontologies (O1 and O2 ) into a third ontology O is any operation by which O1 , O2 are ‘re-interpreted’ from the (global) point of view of O. This has been utilized in the approach of [23] (called semantic integration), which integrates two ontologies by mapping (or translating) them into a common reference ontology. The main feature here is that semantic consequence is preserved upwards to the reference ontology. Intuitively, the difference between integrations and connections is that in the former, we combine two ontologies O1 and O2 using an often large and previously-known reference ontology O, where the models of O are typically much richer than those of O1 and O2 . In the latter, we connect two ontologies in such a way that the respective theories, signatures, and models are kept disjoint, and a (usually small and flexible) bridge theory formulated (in a bridge language) over a signature that goes across the sort structure of the components is used to link together the two ontologies. In E-connections [17], specifically, a finite number of formalisms, typically talking about distinct domains or distinct views on the same domain, are connected by relations between entities in the different domains, capturing different aspects or representations of the ‘same object’. For instance, an ‘abstract’ object o of a description logic DL1 can be related via a relation R to its life-span in a temporal logic T (e.g., a set of time points) as well as to its spatial extension in a spatial logic S (e.g., a set of points in a topological space). Essentially, the language of an E-connection is the (disjoint) union of the original languages enriched with operators capable of talking about the link relations. The possibility of having multiple relations between domains is essential for the versatility of this framework, the expressiveness of which can be varied by allowing different language constructs to be applied to the connecting relations. E-connections have also been adopted as a framework for the integration of ontologies in the Semantic Web [8], and, just as DLs themselves, offer an appealing compromise between expressive power and computational complexity: although powerful enough to express many interesting concepts, the coupling between the combined logics is sufficiently loose for proving general results about the transfer of decidability. But as follows from the complexity results of [17], E-connections in general add substantial expressivity and interaction to the components. Note that the requirement of disjoint domains is not essential for the expressivity of E-connections. What is essential, however, is the disjointness of the formal languages of the component logics. What this boils down to is the following simple fact: while more expressive E-connection languages allow to express various degrees of qualitative identity, for instance, by using number restric-

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

70

J. Hois et al. / Modular Ontologies for Architectural Design

tions on links to establish partial bijections (which we will use below), they lack means to express ‘proper’ numerical trans-module identity. Fig. 2 displays an example of the connection of two ontologies, with a single link relation E. 2.2. Modularity in Architectural Design Specification The aspects of ontological modularity that we employ in order to realize the envisioned application to architectural design are manual alignments, conservative (definitional) extensions, E -connecting thematically different ontologies, and global extension. Thematic module: A thematic module for a domain D is an ontology that covers a particular aspect of or perspective on D. The main impact of this notion is that we assume that two thematically different modules for D need to be interpreted by disjoint domains. An example, that we will elaborate on later, is the conceptual space of materials of objects and qualitative representations of topological relationships between such objects: these interpretations clearly should not overlap. Definitional Extensions: New concepts are, for instance, added to the DOLCE-Lite ontology in the Physical Object ontology (see Section 3.2) by a definitional signature extension. Moreover, we add new concepts to the spatial relations in the Building Architecture ontology, again, in a definitional manner.

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

Linking thematic modules: Alignments are given by the human expert (the architect), identifying certain relationships between thematically different modules. An overall integration of these thematic modules is then achieved by E -connecting the aligned vocabulary along newly introduced link relations and appropriate linking axioms. As dedicated reasoners for E-connections are not available at the moment, we realize this scenario by using E-connections in a way that allows a complete encoding of the semantics of E-connections into OWL DL (compare also [8]): (1) disjointness of thematically different domains is enforced by introducing new ‘local’ top concepts for each ontology (2) domain and range of link relations are accordingly restricted; (3) as E-connection operators we use DL existential and universal restrictions for these link relations. Global extensions of integrated representations: New constraints are added on top of the integrated representation by E-connections. Moreover, the process of building integrated representations might be iterated at a later stage of the specification process, integrating further ontologies whilst treating the previously built representation as a new ‘monolithic’ building block (see Section 4).

3. Modular Ontologies for Architectural Design In the following, we present different ontologies developed for architectural design.2 First, we outline general ontological formalization aspects underlying their development. We subsequently introduce the modularly specified ontologies, in particular from metrical, quantitative, and different conceptual perspectives, while taking into account formal ontological design criteria and modularity issues. Finally, results with respect to expressivity and reasoning aspects are evaluated. 2 www.ontospace.uni-bremen.de/ontology/modSpace/ArchitecturalDesign.html

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

71

J. Hois et al. / Modular Ontologies for Architectural Design Task−Specific Requirements (building automation, access restriction, ...)

Integrated Representation and Axioms based on E−Connections

DOLCE

Physical Object

Spatial Relations

Conceptual Layer

Building Architecture

Building Construction

Qualitative Layer

IFC data model

Quantitative Layer

Figure 1. Different ontological layers contributing heterogeneous aspects of architectural design.

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

3.1. General Specification Aspects Although ontologies can be defined in any logic, we focus here on ontologies as theories formulated in DL, supported by the web ontology language OWL DL 2 [19]. DL distinguishes between TBox and ABox. The TBox comprises all class and relation definitions, while the ABox comprises all instantiations of these classes and relations. Ontologies may be formulated in more or less expressive logics, however, DL ontologies have the following benefits: they are widely used and a common standard for ontology specifications, they provide constructions that are general enough for specifying complex ontologies, and they provide a balance between expressive power and computational complexity in terms of reasoning practicability [14]. In case of architectural design, the domain of buildings and their characteristics and constraints are defined in the TBox. Reasoning over the TBox, e.g., allows to check its consistency. The requirements for classes and instances of concrete buildings can then be axiomatized. Building plans can be instantiated as an ABox accordingly. Reasoning over the ABox, e.g., allows to prove whether the model satisfies the TBox constraints. In addition, spatial reasoning is of particular interest for architectural design. Here, we apply a specific feature of the reasoning engine RacerPro [12] that supports region-based spatial reasoning directly by the so-called SBox [10], which provides reasoning based on the region connection calculus [22], in particular RCC-5 and RCC-8 (see 4.2). 3.2. Thematically Different Ontologies for Architectural Design In the architectural design process different criteria can be grouped closely together into modules that reflect different topics. An overview is illustrated in Figure 1. Basic information with respect to architectural design is metrical data, i.e., a quantitative layer. It particularly reflects metrical data of construction elements if building plans, such as Wall, Door, or Window. From this constructional perspective, entities are specified on the basis of their metrical aspects, such as size, position, or opening angle. In contrast, more abstract spatial relations are indicated by a qualitative layer, which specifies dependencies and spatial constraints. Even though it defines similar entities to those from quantitative information, its characteristics focuses on spatial relations between entities. While, for instance, quantitative aspects of a Wall merely determine the wall’s height and length, qualitative aspects of a Wall describe its bounding of rooms or its connection to ceilings

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

72

J. Hois et al. / Modular Ontologies for Architectural Design

and floors. Finally, a conceptual layer in architectural design specifies architectural entities and phenomena as such. All entities are given by their intrinsic features, e.g., a Wall is described by its material, color, or style. All layers are specified by ontologies. Some of them extend or re-use existing ontologies. They are connected with each other by formalizing link relations across ontologies, which results in the Integrated Representation. This ontology is extended for task-specific aspects that provide a certain purpose. 3.2.1. Quantitative Layer In our use case, we describe information on construction plans on the basis of industrial building components specified by the IFC data model (cf. Section 1). Sample classes from IFC are mirrored by this first layer of architectural information. Constructional elements of buildings are specified by BuildingElement in the Building Construction ontology of the quantitative layer. The properties, they have to specify, are at least their height, length, and width. The formulation (in Manchester Syntax [13]) is: Class: SubClassOf:

BuildingElement height exactly 1 Float, length exactly 1 Float, width exactly 1 Float

Information provided by this quantitative layer is related to concrete floor plans and instantiated accordingly. It can also constrain minimal or maximal sizes of certain entities (e.g., rooms or corridors) on a metrical basis. Another dimension to this representational layer could also be added, e.g., data properties specified with concrete domains [11]. For instance, the values of the size of certain entities, such as windows and doors can be bound to certain upper and lower limits, in order to ensure accessibility.

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

3.2.2. Qualitative Layer Information on spatial-functional entities in the architectural domain and their qualitative spatial relationships are formalized in the qualitative layer. Region-based relationships are defined for entities in this module. The entities are similar to those specified in the Building Construction ontology, but they specify non-metrically determined entities, e.g., functional aspects of entities. A Door, for instance, is specified as a connection between rooms and corridors, it is spatially connected to adjacent entities, and it is also provided by a spatial region that indicates its spatial-functional access. The Building Architecture ontology of this module offers region-based spatial relationships, as indicated by the region connection calculus RCC-8 [22], by re-using the RCC ontology that is introduced in [9]. The eight region-based relations are defined in a property hierarchy. These properties are applied to the entities specified in the qualitative layer. For example, the types of adjacent entities of a Door are constrained as follows: Class: SubClassOf:

Door BuildingElement, rcc:externallyConnectedTo some (Wall or Window), ...

As outlined in Section 3.1, RCC-8 relations and their implications are provided by the SBox of RacerPro. It allows spatial reasoning over these relations. Region-based dependencies on certain architectural entities can be analyzed and architectural requirements can be proven accordingly (cf. [4]).

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

J. Hois et al. / Modular Ontologies for Architectural Design

73

3.2.3. Conceptual Layer In the conceptual layer, architecture-related entities are based on their idiosyncratic characteristics, i.e., they are specified by their properties and axioms without any contextual or embedded aspects. For example, the layer specifies particular subclasses of Door, such as SwingDoor, RevolvingDoor, or SlidingDoor. Entities from the conceptual layer can be related to building plans but also more abstract entities, e.g., functions, costs, or actions. This layer extends a foundational ontology, which provide an abstract foundation for specifying specific domain entities and relations, namely DOLCE [18]. In particular, DOLCE-Lite3 , formulated in OWL DL, is extended in order to provide a categorization of architectural entities. The resulting Physical Object ontology refines physical endurants of DOLCE [18]. Specific qualities can also be described, e.g., the material type of a Window is here defined as one of aluminum, steel, wood, or plastic: Class: SubClassOf:

Window BuildingConstruct, DOLCE-Lite:has-quality exactly 1 Material

3.2.4. Integrated Representation The previous three modules are combined and linked in the Integrated Representation ontology. As described in Section 2, the theory of E-Connections is encoded by providing an additional layer of axioms. Even though all three ontologies are imported into the Integrated Representation, only link relations between classes from different ontologies are defined here. The classes from the different ontologies are made disjoint, as the different perspectives given by the ontologies are not supposed to be ‘matched’:

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

DisjointClasses:

DOLCE-Lite:particular, buildingArchitecture:Functional_Structure, buildingConstruction:Architectural_Feature

Note, that these three classes are ‘artificially constructed’ top node classes from the respective ontologies (namespaces indicate their different origins). As a result, the classes from different ontologies are defined as disjoint sets, i.e., no instance can be specified in more than one of them. In addition, the Integrated Representation ontology does not define any new classes or subclasses. Instead, it specifies correspondences between classes from different ontologies by defining link relations, which are specified as bijective mappings using cardinality constraints. For example, a specific class in the quantitative layer, such as Door in the Building Construction ontology that reflects metrical information about doors, has its correspondence in the qualitative layer, such as Door in the Building Architecture ontology that reflects connections to spatially adjacent entities. The link relation compose specified in the Integrated Representation ontology defines the relation from the quantitative layer to the qualitative layer, and its inverse relation isComposedOf vice versa: 3 http://www.loa-cnr.it/ontologies/DOLCE-Lite.owl

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

74

J. Hois et al. / Modular Ontologies for Architectural Design

ObjectProperty: Domain: Range: InverseOf: Class:

compose buildingConstruction:Architectural_Feature buildingArchitecture:Functional_Structure isComposedOf buildingConstruction:Door

SubClassOf: Class:

compose exactly 1 buildingArchitecture:Door buildingArchitecture:Door

SubClassOf:

isComposedOf exactly 1 buildingConstruction:Door

A similar link relation is defined for entities from the qualitative to the conceptual layer. This relation is again bijective and specified by conceptualizedBy and conceptualize accordingly. An example of this kind of link relation is illustrated in Fig. 2. Given this integrated layer that defines the link relations between the different ontologies, extensions of this layer can then be used to axiomatize specific requirements to model certain design criteria.

Figure 2. A two-dimensional E-Connection with examples from the qualitative and quantitative modules.

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

3.2.5. Task-Specific Requirements Based on the integrated representation layer, the task-specific requirements layer adds additional definitions and constraints to architectural information. It formulates requirements that have to be satisfied by a concrete building plan. These requirements describe certain tasks or purposes that a specific design has to meet. They can, for instance, determine regulations for user access, automatic control of electronic devices, or monitoring and surveillance tasks. As an example, one ontology of this task-specific layer aims at specifying requirements for ambient intelligence (see Section 4). It therefore defines constraints on assistance systems as well as motion sensors and visual sensors that are supposed to monitor movements inside buildings. This kind of information is formalized as ontological constraints on classes from the different modules re-using link relations of the integrated representation. For instance, the requirement that all buildings have an intelligent navigation terminal that provides building information for visitors is defined by the following requirement constraint: Class: SubClassOf:

buildingArchitecture:Building rcc:inverseProperPartOf min 1 (buildingArchitecture:Display and (integratedRepresentation:isConceptualizedBy some physicalObject:NavigationTerminal))

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

J. Hois et al. / Modular Ontologies for Architectural Design

75

4. Application Scenario: Ambient Intelligence The field of ambient intelligence (AmI) has become a rapidly evolving area in research, business, and administration [21]. Its aim is to enrich the environment of daily life aspects by providing automatic control and decision support with respect to environmental conditions. AmI is intended to support humans in achieving their everyday activities. Areas of applications in ambient intelligence focus on work and home scenarios, they address issues of transportation and navigation, accessibility, tourism, elderly and health care [21]. The architectural design process of AmI environments therefore has to take into account relevant information on these criteria. Given our modular ontological representation, design criteria for these different aspects can be formalized accordingly. The thematic modules support the design process by analyzing whether design criteria are satisfied, i.e., in this case whether particular AmI requirements are satisfied.

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

4.1. Use Case: AmI Requirements The modular ontologies for architectural design described above can be used in the design process of AmI environments for defining requirements. In Figure 3, an example of an architectural design is illustrated. The underlying IFC data of this example is instantiated as an ABox in the Building Construction ontology. ABox reasoning over their metrical information analyzes whether the instances satisfy the ontological constraints. The integrated representation requires the instances in Building Construction to be linked to instances in the Building Architecture ontology. The ABox of this latter ontology has to be instantiated accordingly. Region-based spatial relations between these instances are calculated on the basis of metrical information. The qualitative layer then defines the topological relationships. Examples of such calculations are outlined in [4]. The qualitative relations are specified by the RCC-8 relations as pre-defined in the SBox of RacerPro, and SBox consistency can be proven. Moreover, the integrated representation requires the instances of the qualitative layer to be linked to instances in the conceptual layer. Particular PhysicalObject entities are instantiated and related to the qualitative layer. Global consistency can be proven by ontological reasoning in the integrated representation. In addition, particular requirements for AmI environments in the task-specific requirements ontology can be analyzed. In our example, sensors have to cover certain regions around doors. These are functional regions that are defined by the doors and instantiated in the qualitative layer. The region of

Figure 3. Example of an architectural design, in which motion sensors (indicated by arrows in the left image) provide monitoring support (sensor ranges are illustrated as red areas in the right image).

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

76

J. Hois et al. / Modular Ontologies for Architectural Design

the sensor range has to be an inverse proper part of this functional region. Whether such requirements are satisfied by a model can be proven. 4.2. Preliminary Results: Expressivity and Reasoning We have developed and applied the modular ontologies for architectural design to ensure consistency requirements in AmI environments. Several axioms that have to be satisfied by concrete floor plans can be refined in the task-specific ontologies. Link relations can be used to constrain relations between instances from different ontologies, and ABox consistencies can be easily analyzed. In detail, we use ontological alignments, conservative and global extensions, and E-connections for the ontological representation. Furthermore, we use spatial reasoning provided by RacerPro for analyzing RCC-8 relations. Hence, floor plans can be analyzed with respect to their consistency. They are directly given by ontological consistency proofs. In cases that only affect RCC-8 relations and instances in the Building Architecture ontology, a consistency proof exclusively in this module may be sufficient. If constraints in the task-specific requirements ontology use link relations of the integrated representation, only global consistency can be proven, although the integrated representation is based on modularly developed ontologies. Furthermore, as the ontologies are formulated in OWL DL, their expressiveness is limited to axioms formulated in description logics. Even though this is sufficient for our purpose and the representation of architectural design, actual E-connections in the integrated representation are encoded by defining the classes from different ontologies as disjoint sets. Also, the task-specific requirements layer has to import all other ontological modules, which increases complexity and which is unnecessary in most cases.

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

5. Conclusions and Future Work In this paper, we have presented modularly developed and E -connected ontologies for architectural design. Ontological reasoning is used for proving consistency of task-specific requirements. Applying formal methods for modularity supports the ontological design and the representation of different perspectives on the domain. The application scenario outlines the way a metrically modeled work-in-progress design is enriched with ontologically specified requirement constraints. The integration of different architectural information, however, may be extended in several directions: concrete domains can be used in the quantitative layer in order to provide detailed axiomatizations of certain entities, e.g., the minimum and maximum size of steps of a staircase. Several ontologies can be defined for task-specific requirements, e.g., energy saving, navigation, home entertainment, or emergency situations. The qualitative layer can be extended by further spatial relations than region-based relations, e.g., orientation, distance, or shape-based relations. These extensions depend on specific reasoning support and are left for future work. Acknowledgements We gratefully acknowledge the financial support of the DFG through the Collaborative Research Center SFB/TR8, projects I1-[OntoSpace] and R3-[Q-Shape]. The second author also acknowledges funding by the Alexander von Humboldt Stiftung, Germany. Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

J. Hois et al. / Modular Ontologies for Architectural Design

77

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

References [1] Ö. Akin. Architects’ reasoning with structures and functions. Environment and Planning B: Planning and Design, 20(3):273–294, 1993. [2] J. C. Augusto and C. D. Nugent, editors. Designing Smart Homes, The Role of Artificial Intelligence, volume 4008 of LNCS. Springer, 2006. [3] M. Bhatt and F. Dylla. A qualitative model of dynamic scene analysis and interpretation in ambient intelligence systems. Int. Journal of Robotics and Automation, 2009. (to appear). [4] M. Bhatt, F. Dylla, and J. Hois. Spatio-terminological inference for the design of ambient environments. In Conference on Spatial Information Theory. Springer, 2009. [5] J. Euzenat and P. Shvaiko. Ontology matching. Springer, 2007. [6] T. Froese, M. Fischer, F. Grobler, J. Ritzenthaler, K. Yu, S. Sutherland, S. Staub, B. Akinci, R. Akbas, B. Koo, A. Barron, and J. Kunz. Industry foundation classes for project management - a trial implementation. ITCon, 4:17–36, 1999. www.ifcwiki.org/. [7] J. A. Goguen and R. M. Burstall. Introducing institutions. In E. Clarke and D. Kozen, editors, Logics of Programming Workshop, pages 221–256. Springer, 1984. [8] B. C. Grau, B. Parsia, and E. Sirin. Combining OWL ontologies using E-Connections. Journal Of Web Semantics, 4(1):40–59, 2006. [9] R. Grütter, T. Scharrenbach, and B. Bauer-Messmer. Improving an RCC-derived geospatial approximation by OWL axioms. In A. Sheth, S. Staab, M. Dean, M. Paolucci, D. Maynard, T. Finin, and K. Thirunarayan, editors, 7th Int. Semantic Web Conference, pages 293–306. Springer-Verlag, 2008. [10] V. Haarslev, C. Lutz, and R. Möller. Foundations of spatioterminological reasoning with description logics. In 6th Int. Conference on Principles of Knowledge Representation and Reasoning, pages 112–123. Morgan Kaufmann, 1998. [11] V. Haarslev and R. Möller. Description logic systems with concrete domains: Applications for the semantic web. In F. Bry, C. Lutz, U. Sattler, and M. Schoop, editors, 10th Int. Workshop on Knowledge Representation meets Databases, volume 79. CEUR-WS.org, 2003. [12] V. Haarslev, R. Möller, and M. Wessel. Querying the semantic web with Racer + nRQL. In Int. Workshop on Applications of Description Logics, 2004. [13] M. Horridge and P. F. Patel-Schneider. Manchester OWL syntax for OWL 1.1. In OWL: Experiences and Directions, 2008. [14] I. Horrocks, O. Kutz, and U. Sattler. The Even More Irresistible SROIQ. In 10th Int. Conference on Principles of Knowledge Representation and Reasoning, pages 57–67. AAAI Press, 2006. [15] R. Kontchakov, F. Wolter, and M. Zakharyaschev. Can You Tell the Difference Between DL-Lite Ontologies? In 11th Int. Conference on Knowledge Representation and Reasoning, pages 285–295. AAAI, 2008. [16] O. Kutz, D. Lücke, and T. Mossakowski. Heterogeneously Structured Ontologies—Integration, Connection, and Refinement. In T. Meyer and M. A. Orgun, editors, Knowledge Representation Ontology Workshop, pages 41–50. ACS, 2008. [17] O. Kutz, C. Lutz, F. Wolter, and M. Zakharyaschev. E-Connections of Abstract Description Systems. Artificial Intelligence, 156(1):1–73, 2004. [18] C. Masolo, S. Borgo, A. Gangemi, N. Guarino, and A. Oltramari. Ontologies library. WonderWeb Deliverable D18, ISTC-CNR, 2003. [19] B. Motik, P. F. Patel-Schneider, and B. C. Grau. OWL 2 Web Ontology Language: Direct Semantics. Technical report, W3C, 2008. http://www.w3.org/TR/owl2-semantics/. [20] W. Pike and M. Gahegan. Beyond ontologies: Toward situated representations of scientific knowledge. Int. Journal of Man-Machine Studies, 65(7):674–688, 2007. [21] C. Ramos. Ambient intelligence - a state of the art from artificial intelligence perspective. In J. Neves, M. F. Santos, and J. Machado, editors, 13th Portuguese Conference on Aritficial Intelligence, pages 285– 295. Springer, 2007. [22] D. A. Randell, Z. Cui, and A. G. Cohn. A spatial logic based on regions and connection. In 3rd Int. Conference on Knowledge Representation and Reasoning, pages 165–176. Morgan Kaufmann, San Mateo, 1992. [23] M. Schorlemmer and Y. Kalfoglou. Institutionalising Ontology-Based Semantic Integration. Journal of Applied Ontology, 3(3), 2008. [24] H. Stuckenschmidt, C. Parent, and S. Spaccapietra, editors. Modular Ontologies - Concepts, Theories and Techniques for Knowledge Modularization. Springer, 2009. [25] A. Zimmermann, M. Krötzsch, J. Euzenat, and P. Hitzler. Formalizing Ontology Alignment and its Operations with Category Theory. In Formal Ontology in Information Systems, pages 277–288, 2006.

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

78

Formal Ontologies Meet Industry R. Ferrario and A. Oltramari (Eds.) IOS Press, 2009 © 2009 The authors and IOS Press. All rights reserved. doi:10.3233/978-1-60750-047-6-78

Ontology-strength Industry Standards The Case of the Telecommunication Domain 1 Pierre GRENON a,2 and David DE FRANCISCO b a KMI, The Open University, United Kingdom b Telefónica R&D, Spain Abstract. This paper discusses an ontologisation of a subset of the Telemanagement Forum’s New Generation Operational Support Systems standards. The result is a set of ontologies, each corresponding to a particular element in the set of standards, covering complementary aspects of the telecommunication domain. These ontologies are articulated in a modular way according to the standards they target. The ontologies are also mapped, again, in ways adapted from the correspondences put forward within the set of standards itself. We show that ontologisation of informally or semi-formally laid standards can involve non-trivial ontological engineering choices. In addition to sharing lessons learnt, our aim is to support the view that the production of ontologies issued from standards is an endeavour that has the potential of furthering and enhancing standards’ development, their dissemination, and operationalisation. Ontologisation of standards, we believe, should be part of the standard development lifecycle.

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

Keywords. Ontology, Industry Standards, Telecommuncation, NGOSS

Introduction The Telemanagement Forum (TMF hereafter) [1] brings together a large number of telecommunication enterprises and IT providers. TMF develops and maintains the New Generation of Operation Support Systems framework (NGOSS hereafter) [2,3]. NGOSS is intended to support better integration, sharability, reusability and flexibility within and between telecommunication enterprises. For this purpose, the NGOSS framework defines a set of standards, technology neutral architecture, development practices and best design guidelines. Our interest is narrower and revolves around the 3 main following elements of NGOSS: the enhanced Telecommunication Operations Map (eTOM) [4], the Telecommunication Applications Map (TAM) [5], and the Shared Information Data model (SID) [6]. We will take as a basis a number of documents issued by the TMF describing and guiding the use of a standard terminology that supports the specification of a number of aspects of business processes and of what is relevant to managing them in the telecommunication domain. The work we are reporting here is concerned with producing 1 The

work presented here was funded by the European Commission under the project SUPER (FP6-026850). Author: Pierre Grenon, KMI, The Open University, Walton Hall, Milton Keynes, MK7 6AA, United Kingdom; E-mail: [email protected]. 2 Corresponding

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

P. Grenon and D. de Francisco / Ontology-Strength Industry Standards

79

an ontology that follows the understanding of the telecommunication domain which is embodied in the selected parts of the NGOSS framework. Our interest in NGOSS is due to its potential usefulness in describing business process models in the telecommunication sector in ways that may be common to a number of enterprises. The context is that of the development of an application for that very purpose, the management of business process models with semantic technologies between and within enterprises, in the SUPER project.3 In that context, the recourse to NGOSS standards answers the need for specific knowledge in the telecommunication domain (the domain in which SUPER has its use cases) in order to guide the development of ontologies used in annotating ontology-based business process models with domain knowledge. Ontologies are indeed useful in laying out the characteristics of entities—in our case, business process models—found in a domain of interest [7]. As formal artifacts, ontologies support, in the present context, the machine-readable representation of business knowledge used during the design as well as the implementation of business processes. In that respect, NGOSS provides the resources of a common domain vocabulary (through the SID standard), a registry of relevant business functions realised in telecommunication business process or some of their elements (through the eTOM standard) and the resources to link these with IT systems involved in the execution of such processes (through the TAM standard). Using NGOSS standards is thus in first analysis an appealing idea. But NGOSS is provided by the TMF as a set of documents and standard specifications with both descriptive and normative characters. The caveat here is that NGOSS requires processing in order to be integrated within an ontology-based information system, it is precisely the contribution of ontologisation to operationalise standards in this way. In this paper we show that the process of ontologisation of NGOSS is non trivial; carrying it out is not merely to produce a formatted version of sets of concepts issued from standard documents in a language with an XML serialisation. If that much is obvious, all the better, and then we hope our discussion may prove useful to similar endeavours. We hope also that the present paper will contribute to fostering a synergy between industry standards and ontologies. It is widely recognised that ontologies should thrive towards standardisation, not least as an approach to solving the problem of interoperability; it is less obvious that the creation of ontologised versions of standards is a good thing for nonontological standards themselves. But considering that i) ontologies are tools dedicated to the specification of clear and unambiguous resources for the representation of knowledge in any domain and that ii) ontologies are independent component in information systems, it takes little consideration to realise that ontologising standards triggers formidable opportunities for both the development and the deployment of industry standards given the increasing deployment of semantic technologies and ontology-based information systems. We hope that the present paper suggest more reasons to share these intuitions and to propound the underlying methodological disposition. The rest of the paper is structured as follows. Section 1 discusses the informal or semi-formal (diagrammatic) but non ontological models forthcoming from NGOSS specifications of eTOM, SID, and TAM. Section 2 discusses salient technical issues arising from the transformation of these models into ontological artifacts. Section 3 presents the set of ontologies resulting from the ontologisation of our NGOSS sample. The conclusion discusses future work in the light of the benefits of bootstrapping ontological representation in the telecommunication domain with ontologised industry standards. 3 http://www.ip-super.org

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

80

P. Grenon and D. de Francisco / Ontology-Strength Industry Standards

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

1. Informal Model Standards represent a take on a domain. The perspective they embody is thus prospectively one of many. NGOSS’ are industry standards and biased towards specific needs and interests. According to TMF, the main bias is towards a business apprehension of the telecommunication sector in which enterprises operate [4]. Moreover, they are orientated towards enterprises operating both on the side of service provision to end users (telecommunication companies) and of business to business infrastructure procurement. While these aspects are important when debating whether or not to adopt a standard, they are, in first analysis, not obviously relevant to the technical problem of ontologising the standards. For such an ontologisation consists in seeing the domain covered by the standards through the standards themselves. On the one hand NGOSS is a telecommunication industry standard which supports in different ways numerous aspects of business process management from both the operational and business perspectives. On the other hand, it is little more than a set of documents that includes guidelines of applications and that are intended to be adopted and consistently followed by all enterprises committed to the standards.4 The portion of NGOSS discussed here is made of eTOM, TAM, and SID. eTOM is the part of NGOSS that exhibits the business bias mentioned earlier and that establishes a common vocabulary for sorting processes along business operations and functional dimensions. It describes enterprise process areas relevant to telecommunication service providers and is intended to facilitate the analysis of process execution, the integration of processes, and the improvement of existing processes as well as the development of new ones. It is primarily a tool for managing processes within the enterprise although, emphatically, from a business perspective. eTOM, like other NGOSS artifacts, is highly structured and consists in a series of layered decompositions of ever finer grains, from the most abstract to the most concrete (Figure 1). It relies on groupings of different levels that fit one another as process management matriochka. These are eTOM’s functional areas which come in a series of 4 designated levels. In addition, eTOM cuts the cheese in two orthogonal superimposed dimensions. The first is the business activity dimension that defines so-called ‘end-to-end relationships’ between processes insofar as they are complementary in carrying out a business activity (process) from beginning to end. The second is the functional area dimension that recollects processes dealing with similar aspects of the business domain and defines ‘functional relationships’. Thus while the variety of concrete processes covered by eTOM forms the flesh of the standard, the skeleton, the organising structure, is dealt with primarily through the concepts of i) functional area, ii) functional area level (0, 1, 2, 3), iii) functional relationship, and iv) end-to-end relationship. SID is a part of NGOSS that is, in addition to contributing to NGOSS’ business bias, driven towards information. Hence it consists in an information model intended to enable unification across systems and implementations. It is thus an abstract descriptive model defining relevant concepts, their characteristics and relationships. It is enormous, 4 A brief procedural digression is in order. For the adoption of TMF’s standards requires the membership of an enterprise in TMF as only members have access to full standard documents, specifications, and derived products. Thus, the dissemination of TMF material is not straightforward. For our purpose here, we will primarily point towards publicly available sources and essentially to portions of the TMF’s website. These documents are nevertheless enough to go a long way in carrying out our discussion as we do not ambition to present a fully ontologised version of NGOSS but only the main coarser lines emerging from such a tentative exercise. It is as if the ontologisation presented here was only that of the publicly visible part of the NGOSS iceberg.

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

P. Grenon and D. de Francisco / Ontology-Strength Industry Standards

     

 

  

81

    

 

  

  

 

        

   

      

 

           

        

        

        

                 

           

          



Figure 1. eTOM’s levels 0 and 1 structure (adapted from [8]). Level 0 concepts (shaded) are decomposed into level 1 concepts. Boxes traversing the picture horizontally are ’functional’ groupings and those traversing the picture vertically are ‘end-to-end’ groupings.

  

  

 

  
 

 

  



 

 

 

  






 



 

 

  




  

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

Figure 2. SID’s domains and ABEs of first level structure (adapted from [9]) coverage and structure. Shaded boxes represent domains and unfilled ones some of their first level component ABE. The further decomposition of the latter is not shown. The relative position of domains in this picture is not significant.

however, as, in fact, it gathers every concepts that are more or less directly relevant to any of business activity in the telecommunication domain, some of which, sometimes, are hardly domain specific at all (for example, duration or geographical location) and some of which are rather high-level as domain concepts go (for example, software and hardware but also country, money or employee). In many cases they are prolonged in ways that are more obviously tailored to the telecommunication domain (service bundle, grid location, ...). But overall SID provides rather abstract background means for making specifications of concrete applications in enterprise-specific ways, suggesting that enterprises may have to extend or complement SID with their own specifics. As to its structure, SID is a layered framework with, moreover, different kinds of layers (Figure 2). The coarser grain is that of ‘domains’ which are decomposed into so-called ‘aggregate business entities’ in which partake ‘entities’. SID is thus structured using the following high-level concepts: i) Domain, ii) Aggregate Business Entity (ABE), iii) ABE levels (1 to 5), iv) (SID-)Entity, and v) relationships of decomposition or specialisation. TAM is geared towards applications and is intended to facilitate the sorting of business functions and data and information into application areas allowing functional analysis

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

82

P. Grenon and D. de Francisco / Ontology-Strength Industry Standards

within a given enterprise operating against the IT background. Each company adopting TAM follows it in laying out its so-called ’system map’ used to characterise the company’s pool of IT systems and their relationships. TAM provides structure and is in a way an IT system or application oriented equivalent of eTOM. Also, within NGOSS, TAM provides a mean for characterising software applications in functional terms (relative to SID) and business terms (relative to eTOM). Because of this, and also for lack of room, we don’t provide a depiction of TAM, moreover not needed for our present purpose. TAM includes a very large number of concepts and these are made part of areas which themselves come in a number of levels characterized by different degrees of requirements, specification, and implementations of designed functionality. The main structural concepts in TAM are thus: i) application and ii) application level (1 to 3) . As seen from this rapid presentation, these elements are not just independent elements. There are in fact built-in correspondences and mappings such as between SID and eTOM. There are also elements such as TAM that are designed to support the integration of objects within the domain (for example, software solutions) according to dimensions provided by other elements.

2. Towards an ontologisation of informal standards Ontologising the parts of the telecommunication domain that are covered by the NGOSS elements we are focusing on consists in answering the following basic questions:

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

1. 2. 3. 4. 5.

What are the entities in the domain? What are the properties of these entities? How do these entities relate one to another? What are the different kinds or types under which fall these entities? How do these types relate one to another?

In our circumstances, all of these questions are not equally challenging nor critical. Moreover the stakes in answering them is lowered in the case of ontologisation of a standard when that standard is well worked out. For the answers to questions 2 and 3, in particular, are usually a matter of specification. Thus, how good a job we do answering them depends more or less directly on how good a specification the standard provides. Consider the descriptions in section 1, they offer elements for answering the questions just put forward. The trick, however, is that NGOSS stands in the first place as a terminology system—for there is little difference between the conceptual system it pretends to put forward and a mere terminology. NGOSS elements primarily consist in vocabularies. Surely there is some structure to them, so in truth we have something that overall approximates a thesaurus, listing terms (standing for concepts) and sketching conceptual structures associated to them. The stake in ontologisation is, however, to represent entities in a domain and the ontological structure of that domain (including the types under which fall entities and relations between them), it is not to lay a terminological system alluding to concepts [10]. This shows that how to answer some of the questions laid just now is not entirely obvious, in particular the correlated or dual questions 1 and 4. We have a number of terms provided by NGOSS. Our first question can be put in relation to the correlated semantics of this set of terms as follows: Are the entities in the domain denoted by the terms used in NGOSS (roughly put, are these terms names) or

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

P. Grenon and D. de Francisco / Ontology-Strength Industry Standards

83

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

are the entities in the domain things which fall under kinds whose label is provided by NGOSS (roughly put, are these terms predicates)? In the first case, NGOSS terms are terms for instances (falling under kinds) and in the second case NGOSS terms are terms for kinds (under which fall instances)—we will use ‘kind’ and ‘type’ interchangeably. Another way of putting the question is to put the forthcoming NGOSS ontologies in an applicative context. Suppose an enterprise, for example, Telefónica decides to use NGOSS ontologies to provide an umbrella over their application specific ontologies, making NGOSS ontologies domain ones in the sense of [7]. Perhaps there is all that is needed already in the NGOSS ontologies, but suppose there is not. Suppose Telefónica has to extend NGOSS ontologies. Should Telefónica’s ontologies introduce specialisations (of kinds) or should they introduce variants (of instances) of the constructs from the NGOSS ontologies? For example, is refining SID’s notion of Resource Domain with Telefónica Resource Domain a case of introducing an instance or a kind? Is Resource Domain an instance or a kind in the first place? If a kind, is it to be instantiated or subtyped? Although we consider this scenario for the sake of illustrating the importance of recognising the distinction between kinds and their instances, in a concrete application context, an enterprise like Telefónica would expect to find a ready ontologisation of SID leading the way to extension. Moreover, there are too many parameters in application for relying only on a prospective bottom-up approach. But it is only once it has been ontologised that a standard would allow us to decide the relevant issues. At the time of ontologisation, however, our recourse is to track hints and symptoms of the distinction between kinds and instances within the standard itself. The first thing we can do then is to look over the terms in the standards’ vocabularies and then proceed with an ontologisation of these as concepts in their own rights. This strategy is dubious, however, for although in some cases we may have good independent understanding of how to represent ontologically certain concepts such as, for example, software or company, it becomes more difficult the closer we approach the intimate structure of the standard itself. Indeed, while technical NGOSS concepts come with a gloss, the documentation is not designed to answer ontological questions. Consider the following glossary entry for Aggregate Business Entity on the TMF’s website: “A well-defined set of information and operations that characterize a highly cohesive, loosely coupled set of business entities. ABEs are used in the SID Business Framework to represent business concepts.” [11] As can be seen, this description is less than helpful for it is neither comprehensible nor informative. The only thing we can extract from it is that ABEs are groups with allegedly clear membership criteria, although we know nothing of these criteria. The second sentence is also particularly disconcerting as it says nothing. But then, for our purpose here, we do not need to grasp the fundamental nature of the ABE concept, nor even, we submit, that of the variety of ABEs found in SID nor, to take another example, the business functions found in eTOM. The notion of ABE is an NGOSS construct, but it is difficult to extract from the definition cited earlier anything more meaningful than that an ABE is a group. The notion of business function is, of course, more pervasive. But although the question of what a business function is precisely and intelligibly may be fascinating, it is not one that requires being decided for the ontologisation of eTOM. What matters to the ontologisation of NGOSS elements is that there are things that NGOSS puts under the ‘ABE’ heading, on the one hand, and that

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

84

P. Grenon and D. de Francisco / Ontology-Strength Industry Standards

the variety of business functions it considers may be arranged in areas and according to relationships of different kinds. For, again, the prospective ontologies are not thought out here in order to check or correct the standards, but in order to apply the prescriptions of the standards through the corresponding forthcoming ontologies when describing business processes with unambiguously referenceable semantics. What matters to us then is that NGOSS puts concepts under headings and that it makes statements about those things in virtue of them falling under such headings. Our interest is with two correlated types of descriptions provided by the standards: i) the population of a variety of kinds and ii) the representation of knowledge about those kinds. Their intrinsic and intimate nature is something that for our purpose may remain primitive and undefined, hidden in a black box, left to the standard developers to decide.

3. The ontology of NGOSS standards

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

What are the kinds and instances that show in the models forthcoming from the application of NGOSS’ perspective to the telecommunication domain? We hope that the foregoing sections provide enough preparations and offered enough caution to get up to speed at this point. We submit that the ontologisation of NGOSS standards should be primarily a representation of the layered structures of the set of concepts that properly support domain knowledge representation. In first analysis, this process should be seamless. eTOM is about business functions and its building blocks are specialisations of (the kind) business function. These are the domain specific kinds. But then they are organised in groups, or ‘areas’, of a number of levels and bunched together in similarity classes generated from the two structural relationships mentioned in section 1, end-to-end and functional. SID is about all sorts of things (arguably, SID ought to be many ontologies) that correspond to domain specific kinds (software, user account, card processing, and so on). The SID’s structure, imposed over the former, is more than its tentative coverage the defining characteristics of SID. It is made of the kinds of layers identified by SID, namely: domain, ABE, and also (SID-)entity; these and their subsumption relationships. TAM is about connecting software solutions and IT Systems to both eTOM’s business functions and their groupings and SID’s domain kinds and their groupings. It uses domain kinds for the representation of software and their characteristics and it should be sensitive to the sort of structuring that goes on in both eTOM and TAM. It’s contribution is also to organise applications into groups of layered levels. The parts of the standards made of domain-specific kinds deserve an ontology in their own rights, but not differently than if they were considered outside of the standard. Hence we shall leave them aside and focus our attention on the ontologisation of structural elements (the main prospective abstract kinds and the relationships they or their instances partake in). This may be, however, more difficult than it appears in first analysis. For there are two questions that have to be addressed now, and these prove to be thorny: 1. How do domain kinds stand to structural kinds? 2. What sort of subsumption is there in the structures illustrated in section 1? Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

P. Grenon and D. de Francisco / Ontology-Strength Industry Standards

85

These two questions are not independent and they should be taken in that order. Answering the first question faces us with an alternative. Either domain kinds are specialisations of the structural kinds or they are not. To say they are is to take at face value the decomposition shown in Figures 1 and 2 as a hierarchical decomposition (the same is true for TAM’s application levels). In that case, the answer to the second question is forthcoming as the subsumption relation in the structures exhibited is just subtyping. Subtyping or hierarchical subsumption among types can be defined as the instantiation of the higher type by instances of the lower one. Formally, for two types A and B,

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

subtypeOf (A, B) ≡def ∀ x (instanceOf (x, A) → instanceOf (x, B)) If domain kinds are not subkinds of structural ones, we have to find in which relations they stand and the answer to the second question may bring about a number of relations. Making domain kinds subkinds of structural kinds is appealing. It is simple and the forthcoming representation is prospectively simple as well since it merely relies on the subtypeOf relation. Only, this answer to the first question, and by way of consequence to the second, is wrong. Consider eTOM’s Billing. It is a kind of function—a subkind of the general business function kind whose instances are all similar in essential ways but may vary in minute characteristics. Billing is also a level 1 eTOM group. But then it also falls under the subsuming level 0, Operation. But what of an instance of Billing? It is a function performed by a specific process, perhaps, perhaps something more generic, but it is not an instance of Operation. Rather, it is subsumed in a way yet to be specified under a (functionally broader) instance of Operation. Here, the subsumption involved between kinds is definable in terms of that between instances in the same way that subtypeOf is, but it is a different subsumption. Similarly, in SID, an instance of what we labelled ‘Bill’ in Figure 2 is not an instance of what we labelled ‘Customer’. The latter is a domain while the former is an ABE and the subsumption relation is one of decomposition or constitution. We can multiply examples, in the end the structural decomposition in NGOSS standards proves to be not the hierarchical subsumption of subtypeOf. How do we cope with this though? We have kinds through and through and nevertheless we have to account for the fact that Billing in eTOM is a level 1 area but we do not want to make this a subkind relation. One strategy could consist in constraining all instances of Billing to have a level attribute set to the adequate value. This is because instances are commonly regarded as terms and thus as the subjects of assertions whereas kinds are regarded as predicates, making up assertions about instances although not as subjects. This remains manageable as long as we can find a proper work around and schematic representations for making assertions which would be in first analysis about kinds. But what if we mean to make assertions about kinds themselves? What if, for example, we want to associate them (not their instances) with an ID? The solution is simple. We only need to turn kinds into instances of other kinds making the latter higher order kinds. This is the solution we propose to the problem of ontologising the structure of NGOSS standards. The end-to-end and functional relations in eTOM, the relations between domains and their component ABEs as well as the relations between these and their subcomponents are relations between instances of higher-level kinds. The foregoing discussion leads to an ontology for the NGOSS standards that can be recollected in two tables. Table 1 presents the main second order kinds in each of the NGOSS standards. Table 2 recollects the main relations between first order kinds.

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

86

P. Grenon and D. de Francisco / Ontology-Strength Industry Standards

Table 1. Second order kinds with examples of instances in NGOSS ontologies. NGOSS eTOM

SID

TAM

Second order kind

Example of instance (first order kind)

Area Level 0, 1, 2, 3 Functional group area

0: Operation, 1: Fulfillment, 2: Selling, . . . Customer Relationship Management

End-to-end group area

Fullfillment, Billing

Domain ABE Level 1, 2, 3, 4, 5

Customer 1: Customer Bill, 2: Customer Billing Credit, . . .

Entity

Customer Account, Billing Period

Application Level 1, 2, 3

1: Bill Calculation, 2: Affiliate, 3: Sales Workflow

Table 2. Examples of structural relations between first order kinds in NGOSS ontologies. NGOSS eTOM

SID

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

TAM

Relation

Domain

Range

Example

end-to-end

Area

Area

Selling — Order Handling

functional part of

Area Area

Area Area

Selling — Problem Handling Selling — Fulfillment, Selling — CRM

in Domain

ABE type

Domain type

Customer Bill — Customer

in ABE sub-ABE of

Entity type ABE type

ABE type ABE type

Customer Bill — Customer Customer Billing Credit — Customer Bill

part of

Application

Application

Price — Solution Management

In the context that motivated our implementation of the ontology sketched here, we worked with two languages, namely: WSML [12] and OCML [13, pp. 47-66]; because tools were being developed that supported ontologies and ontology-based descriptions of business process models in these languages. In the end, however, the main integrating software, WSMO Studio [14] worked with WSML ontologies—allowing to import OCML ones via a plug-in. For this reason, our brief discussion of modelling specifics will use WSML. Also, in addition to illustrating the code (Figure 3), we want to make a brief point better made with WSML. For in order to formalise the ontology put forward here we need a knowledge representation language that allows treating both kinds and instances as terms in a logical sense. This means in particular that we need languages that go beyond Description Logic languages in which the distinction between an individual (instance) and a class (type) is hardwired. OCML does not make that distinction and allows, like CycL [16], for types of any order. But WSML does, or rather WSML is a family of linguistic variants among which WSML-DL does not allow for a term to be both instance and concept (type in WSML)—in short, it keeps separated namespaces for these two sorts of structures. Hence our implementation located us on the linguistic scale at the level of WSML-flight which allows what we needed and for which there is still an efficient reasoner (important for validation queries which we have no room to show here). In contrast, had we used OWL, we would have had to locate ourself within OWL-full so as to use no longer the OWL fragment but RDF(S) [15]. This is the only point we want to make regarding implementation as it is telling of the requirements that ontologisation imposes on the knowledge representation and that are not always in line with the principle of ontology language minimalism that is often considered as best practice and that also often influences the design of applications in which ontologies are intended to play a role. We are, of course, only pointing at a

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

P. Grenon and D. de Francisco / Ontology-Strength Industry Standards


 
 



 

  

 


 
 
  


 
 
 
 

 

 
 

 


 
  
 


  

 



87

  

 
  



  
 
  
  

    

  

   

   
 ,
  

 
  

 



    

   


   
  
 

 
 
  

   

 




Figure 3. Excerpt of WSML code from SID. Structural kinds are WSML concepts (left) whose instances are domain kinds which are also concepts in an orthogonal first order hierarchy (right).

mainstream tendency for there are ontology-based information-systems, such as Cyc [16], which successfully do not embarrass themselves with minimalism.

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

Conclusions and future work We have presented the main lines of an ontologisation of elements in the NGOSS set of standard. We focused on fundamental issues we met in the process showing that such an endeavour is not merely a matter of formatting and involves hard knowledge representation problems. Some of these problems are ontology engineering ones which are not specific to the task of a standard’s ontologisation. But others, those involved with the representation, not of the domain knowledge included in the standard, but of the very structure which the standard imposes upon this knowledge are peculiar to such task. There are other aspects of the work which we have not covered here but that are nevertheless significant to ontologisation at large. Two of these are i) the structure of the resultant ontology itself and ii), correlated to the first, the implementation of mappings between standards. The ontology that we implemented is in fact a set of ontologies, modular, but integrated, not least through mappings between modules. The initial rationale came from the fact that the targeted standards are themselves modules in the overall NGOSS framework. Also, they are intended to serve different purposes and covering different areas of the telecommunication domain or rather taking different perspectives on it. However, these views are also complementary and interlinked (in the NGOSS framework itself). Thus although the ontologisation of NGOSS standards was modular in our work, the ontologies required taking into account interdependencies and crossreferences. To this end we strived to preserve each standard element in correspondence with an ontology and to add bidirectional pairwise mappings between them, thus we developed overall 6 ontological artifacts (3 ontologies and 3 mappings). Future work involves further modularisation and articulation, in particular in relation to the SID ontology that we have alluded to as being tentatively a set of ontologies itself. A reason for this is that SID is a grab bag of a multitude of concepts covering aspects of the telecommunication domain that can be abstracted into independent modules (currency, time, location, and so on). A significant motivation for this is that modularisation affords handling smaller ontology artifacts and is thus more friendly to software environments. This is particularly significant in relation to SID because of its sheer size (our ontology

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

88

P. Grenon and D. de Francisco / Ontology-Strength Industry Standards

contains more than 1200 terms5 ) This, however, poses problems of modularisation hitherto unaddressed in our work in relation, in particular, to the cross-references involved in the definition of attributes of the various kinds in the ontology. Another somewhat related problem that we have not discussed here is the apparent ‘flattening’ of domain-kind hierarchies in our ontologies. Because we have represented the layered structures of standards with subsumption relations that were not mere subtyping (for the reasons we gave), the resultant hierarchy is shallow in comparison to the (mistaken) interpretation of layers as subtyping. One part of the solution which prolonges the idea of ontologising structural elements of standards is provided by ad hoc defined classes under which domain kinds falls. This is, for example, the case with SIDAbe which (hierarchically) subsumes CustomerBillAbe in our excerpt of Figure 3. This kind is a kind that any instance of SIDAbeType (that is to say any ABE kind) is specialising, and it is thus definable. This strategy can be applied to any second-order kind supporting the standard’s structural classification. In that way, depth can be added to the ontology using the correct representation of the structure of the standard to define kinds of entities in the domain corresponding to second-order kinds. In our analysis, some belong to SID already, so our approach has the advantage of providing clear and operational semantics to these SID concepts. Other can be defined opportunistically, but it is an open question whether they should be included since following the standard to the letter would seem to require leaving them aside. Or, perhaps, we would need to put in place yet another layer in which to express that a given ontological object is in correspondence with a SID concept. Another part of the solution, we alluded to already. It is that domain kinds deserve an ontological treatment on which we have not focused here. Of course, domain kinds are parts of standards and a robust ontologisation of standards ought to include their robust ontologisation as such. However, in comparison to ontologising the definitive structure that standards impose on domain knowledge the stake are lower. In practice also, this task was not our priority and it was performed minimally sometimes, opportunistically (in relation to use cases needs within the project forming the context of our work), and thus unsystematically. But this just shows that our work requires so much refinement in order to obtain a robust ontologisation of the standards. Standards, however, don’t always include such a level of refinement regarding domain kinds. Thus, again, we are faced with hesitation in face of the possibility of generating ontologies going beyond standards’ specifications—for then they no longer reflect, but extend standards. Our ontologisation of NGOSS was driven by the requirements of an application developed in a project. Our aim was not to ontologise the NGOSS standards for the sake of it. NGOSS provided a sharable elaborate and broad resource for a mid-level ontology of the telecommunication domain. Thus our ontologisation was opportunistic and not orthodox in relation to the alignment of our ontological artifacts and the sources used and to some extent their versions. Hence, rather than successive ontologisations of evolving standards, we maintained an evolutive (modular) ontology, first based on versions of standards that became eventually superseded and progressively amended to cope with standards’ evolution. The risk with this strategy was for the product of our ontologisation to become an hybrid between successively targeted versions and not strictly compliant with any of the specific targets. This is a trade-off that, in the context of our application-driven ontologisation, we accepted and coped with. Overall we didn’t 5 Technically

it is twice as large in the WSML version as terms are declared twice, once for each kind of structure (instance and concept). That phenomenon doesn’t occur with OCML. Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

P. Grenon and D. de Francisco / Ontology-Strength Industry Standards

89

experience problems due to this, because the use of our ontologies was shared within the project (in particular by 4 use cases each driven by a different company). Nevertheless, NGOSS consists in standards that are under incessant development, refinement, and update. Thus, the ontologisation of NGOSS—if it is to be properly aligned with the standards—needs to follow corresponding cycles and perhaps also unforeseen rework. We believe that such work ought to be facilitated once ontologisation has reached a sufficient degree of modularisation and the mechanism for the coherent handling of modularisation is in place. But then, perhaps, the prospective modularisation of resultant ontologies may be finer grained than NGOSS’. In that respect then an ontological version of NGOSS may be susceptible to more sophistication than the originating standards themselves. It seems that this is a dilemma that can only be resolved if any addition triggered by ontologisation is fed into the standardisation process itself for its evaluation. These points, the last in particular, suggest that not only is there value in ontologies benefiting from standardisation efforts but in return ontologies may also be useful in contributing to the development and navigation of the standards themselves, including, in particular, across different versions of the standards. Hence, while it is intuitively more or less obvious that standards are good candidates for ontologisation, we hope to have at least provided a good motivation for extending standardisation efforts so as to include ontologies as one among possibly many final outputs of standardisation.

References [1] [2]

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

[3] [4] [5] [6] [7]

[8] [9] [10]

[11] [12] [13] [14]

[15] [16]

Telemanagement Forum, http://www.tmforum.org M.J. Creaner and J.P. Reilly, NGOSS Distilled: The Essential Guide to Next Generation Telecoms Management. The Lean Corporation (2005). NGOSS, http://www.tmforum.org/SolutionFrameworks/1911/home.html eTOM, http://www.tmforum.org/BusinessProcessFramework/1647/home.html?catID=1647 TAM, http://www.tmforum.org/ApplicationsFramework/2322/home.html?catID=2322 SID, http://www.tmforum.org/InformationFramework/1684/home.html?catID=1684 N. Guarino. “Formal Ontology and Information Systems” in N. Guarino (ed.), Formal Ontology in Information Systems. Proceedings of FOIS’98, Trento, Italy, 6-8 June 1998, IOS Press: Amsterdam, 1998, pp. 3–15. “Enhanced Telecom Operations Map (eTOM), Business Process Framework Release 7.0”, http://www.tmforum.org/browse.aspx?linkID=35431&docID=8862 “SID model”, http://www.tmforum.org/sdata/images/catheader/2417.gif B. Smith, “Beyond Concepts, or: Ontology as Reality Representation” in A. Varzi and L. Vieu (eds.), Formal Ontology and Information Systems. Proceedings of the Third International Conference (FOIS 2004), Amsterdam: IOS Press, 2004, pp. 73–84. ABE, http://www.tmforum.org/OpenGlossary/4716/home.html?catid=4716&gid=29 N. Steinmetz and I. Toma (eds.), “D16.1v1.0 WSML Language Reference WSML Final Draft 2008-0808”, http://www.wsmo.org/TR/d16/d16.1/v1.0/ E. Motta, Reusable Components for Knowledge Modelling: Case Studies in Parametric Design Problem Solving, Frontiers in Artificial Intelligence and Applications, Vol. 53, Amsterdam: IOS Press, 1999. M. Dimitrov, A. Simov, V. Momtchev, and M. Konstantinov, “WSMO Studio — A Semantic Web Services Modelling Environment for WSMO”, in ESWC ’07: Proceedings of the 4th European conference on The Semantic Web, Berlin and Heidelberg: Springer-Verlag, 2007, pp. 749–758. M. Dean and G. Schreiber, (eds.),“ OWL Web Ontology Language Reference”, W3C Recommendation, 10 February 2004, http://www.w3.org/TR/2004/REC-owl-ref-20040210/ D. B. Lenat and R.V. Guha., Building Large Knowledge Based Systems, Reading, Massachusetts: Addison Wesley, 1990.

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

90

Formal Ontologies Meet Industry R. Ferrario and A. Oltramari (Eds.) IOS Press, 2009 © 2009 The authors and IOS Press. All rights reserved. doi:10.3233/978-1-60750-047-6-90

Disentangling Knowledge Objects Stefano BORGO a,1 and Giandomenico POZZA b for Applied Ontology (LOA), ISTC-CNR, Italy b Istituto Zooprofilattico Sperimentale delle Venezie, Legnaro (PD), Italy a Laboratory

Abstract. We discuss a framework for knowledge management based on ontological techniques. The central notion we study, called knowledge object, is provided as a first attempt to provide a new approach to model the relationship between the material, the information and the organization’s perspectives. The paper comprises two parts. In the first part it gives an introduction to knowledge objects, their elements and some basic properties. In the second part, it shows how knowledge objects help in understanding and clarifying knowledge creation and evolution in the scenario of a veterinary public health institute, namely the Istituto Zooprofilattico Sperimentale delle Venezie (IZSVE). Keywords. knowledge object, knowledge management, ontology, veterinary public health

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

Introduction The modeling of the activities of manufacturing enterprises and service providers requires the introduction of specific notions and the development of dedicated techniques. As it is often the case in application domains, any modeling effort has to take into account complex interactions among different entities that present distinct characteristics: physical, informational, agentive, epistemic, contextual. For this reason, one should take an interdisciplinary approach and try to organize a modular system to which new elements can be added and alternative perspectives related. This should be, in our view, the strategy to follow in knowledge management in the attempt to overcome the fragmentation of research and the (perhaps only apparent) incomparability of results. In this paper we investigate the notion of knowledge object, recently introduced in [Pozza et al., 2009], by exploiting it in the formalization of a specific case study. This notion is studied here in isolation since we are still at an evaluation phase: we need to understand if knowledge objects, as here understood, properly identify a key element for the modeling of the enterprise and if they are fruitful in knowledge management in general. The paper is organized as follows: Section 1 introduces knowledge objects and the ontological elements on which they rely; Section 2 discusses some general property of knowledge objects; Section 3 discusses two real cases of knowledge objects and studies their internal (across elements) and external (across knowledge objects) relationships; Section 4 sums up our work and adds further observations. 1 Corresponding Author, E-mail: [email protected]. Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

S. Borgo and G. Pozza / Disentangling Knowledge Objects

91

1. Knowledge objects and their basic elements

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

The proposal we presented in [Pozza et al., 2009] introduces a technical notion, dubbed knowledge object, which we believe is useful in modeling applicative scenarios. Of course, knowledge management requires more than a single formal notion since one needs to talk of important and complex elements like products, agents and activities. A successful framework has to do justice of a series of things, from activities to data, from physical entities to agents’ roles, that require sophisticated frameworks to be properly introduced. Still, it is worth to study fragments of a knowledge management scenario to test single notions and their behaviors as well as to verify their generality and coherence within an overall framework like those provided by foundational ontology. The idea being that, step by step, a full homogeneous picture will slowly emerge. At the core of most approaches to knowledge management there is a fairly wide notion of organization as a complex body. Our work in this paper takes the organization’s perspective as a main player. We use the term organization essentially to refer to structured hierarchies of roles [Masolo et al., 2004] including organization and roles’ goals and interdependences. Of course organizations are much more than this. Nonetheless, for what concerns us in this paper, this view suffices to link knowledge and activities to organizations and to agents. More precisely, in this paper we work within the perspective of a fixed organization, namely the Istituto Zooprofilattico Sperimentale delle Venezie (IZSVE2 ), although we do not really investigate the import of its structure into the entities called knowledge objects. The notion of knowledge object that we model is built out of three elements grounded in ontology: a material entity, an information entity and an actor that here is IZSVE and that, generally speaking, could be any organization, role or human/artificial agent. A material entity, hereafter indicated by M, is any object or amount of matter of interest to the actor, e.g. water, milk, sand, chairs, cows. Note that when we write M we mean a specific amount of matter: the water this morning in container number 23 on my lab desk or the liver I’m holding in my hands now. To be more precise, M is a name for the material entity as Gioele is a name for a given person. We write M(t) when we want to refer to M as it is at time t like one says ‘Gioele at age 11’ to mean how this person is at some period of time. The information entity D is harder to pinpoint. We write D to indicate, roughly, a time-dependent collection of data. Informally, one can think of D as a tuple in a relational database. As seen for M, D can be temporally distinct: when we write D(t) we mean the specific collection of data corresponding to D at time t. To keep the presentation simple, we do not formalize further the nature of entity D. For what concerns us, D could be seen as a collection of elementary assertions or sentences about M. It is important to understand that the sentences in the collection D can vary over time because, e.g., some are added or substituted as new knowledge is acquired over time. Also, our reference to elementary sentences is itself a simplification; on the one hand, we do not commit to a specific language so it is not determined what counts as a sentence and, on the other, in most applications D will be a relational database comprising structurally complex assertions. 2 http://www.izsvenezie.it

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

92

S. Borgo and G. Pozza / Disentangling Knowledge Objects

Definition 1 (Knowledge object) Given a material entity M, an informational entity D and a role, agent or social entity O, a knowledge object is a triple of form (M, D, O) such that at each time t, D(t) collects the data that at time t O has about M.

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

Thus, if O is a manufacturing enterprise and M a product iterm, the idea is that (M, D, O) is a knowledge object where D(t) is the collection of data that the organization O has about product item M at that time. Note that, for the sake of simplicity, in the paper we use expressions of form (M, D, O) to refer to knowledge objects although, more properly, we should talk of entities a defined by dedicated expressions like a = KO(M, D, O) whose intended reading is: “M, D and O, in this order, form the knowledge object a”. Since here we are not attempting to provide an axiomatization of such a notion, our choice is harmless and improves readability. We hasten to point out that our use of the expression “knowledge object” is technical and that in the literature there are other kinds of entities that are called knowledge objects, cf. [Mentzas et al., 2001,Bolisani and Oltramari 2009]. The ontological import of knowledge objects should be fairly clear from the previous discussion. We can use the DOLCE foundational ontology [Masolo et al., 2002] to highlight some category constraints: PED(M), element M must be a physical endurant; APO(O) ∨ SOB(O), element O must be an agentive physical object, e.g., a person, or a social object, e.g., a company.3 Regarding D, as anticipated, its ontological status has not been fully clarified yet. One can see the sentences in D as abstract entities, which is the informal view we take in this paper, or focus on the sentences’ content, which would correspond to qualia. D itself, being a collection that exists and changes in time, is better seen as an endurant. A further observation is needed. Our guiding intuition is that the element D changes over time because O gathers or looses data about M and because it registers some changes that M undergoes. Viewing D as a database tuple, one should think that there is an entry in the table for each kind of data that could be of interest to the agent O.4 This informal reading of D is suggestive but incorrect since it is important to see D as an ontological entity, thus independent of any database structure, formal language or coding procedure. Indeed, the intention is that D collects the meaning of the data that the organization has about M, and the meaning is obviously independently of how and where the data are stored. Even our suggestion of considering D as a collection of sentences presents the same problem. Although there is nothing wrong in viewing D as a collection of sentences, we insist that the data collected by D must be representation independent in order to avoid commitments to languages of one form or another. After all, the choice of a language, including its expressiveness, clearness and lack of ambiguity, is an implementation issue. From this perspective, it is clear that both the database tuple and the sentence collection interpretations fall short of the required representation independence since each brings Information Technology’s (IT) aspects into the picture. IT considerations are important and unavoidable at the implementation phase but they do not belong to the abstract domain of knowledge management we are after. 3 Roles are not defined in [Masolo et al., 2002]. A proposed revision of DOLCE introduces concepts as a distinct category C viewing roles as a subcategory of this, see[Borgo and Masolo 2009]. In this new version of the ontology, we would write APO(O) ∨ SOB(O) ∨ C(O). 4 In principle, this tuple could be infinite as we do not limit or constraint information data.

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

S. Borgo and G. Pozza / Disentangling Knowledge Objects

93

Knowledge objects, being dependent on elements belonging to different ontological categories, are particularly complex entities. In particular, they include distinct forms of knowledge: explicit knowledge, corresponding to the single elements M, D, and O; and tacit knowledge provided by the sophisticated relationships among M, D and O. The first is dubbed explicit because these elements are directly accessible to the agentive element O: the object M, the data contained in the information system as well as the organization structure, its roles, goals etc. The second type of knowledge is dubbed tacit because, although somehow constrained by protocols and standards, in practice it relies on informal procedures and habits developed over time within the organization. A further source of complexity of knowledge objects is inherited from element O. If O is an organization, the knowledge object (M, D, O) is affected by O’s role hierarchy and can be studied in relation to the knowledge objects of type (M, D, O S ), where O S is a sub-organization of O (e.g. a laboratory), of type (M, D, R O ), where R O is a role of O (e.g. a technician), and of type (M, D, A R ), where A R is an agent playing the role R in organization O. We will see other cases in the next section where we concentrate on the first two elements, M and D, and their dependences. Generally speaking, with the introduction of knowledge objects we insist on a change of perspective with respect to traditional IT approaches since we claim that M, D and O must be regarded as equally important elements in knowledge objects and in knowledge management in general.

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

2. The knowledge object entity As we discussed, we use M to indicate some material of interest to the agent O. In general this could be anything, from standard “resource material” like milk, plastic and wood, to sophisticated entities like a cow, a liver, a piece of furniture, a computer. Our approach aims to be general to make clear the possible interactions with important techniques even in domains beyond IT, e.g., the methodologies to managing artifacts and functionality descriptions developed in engineering design [Chandrasekaran and Josephson 2000, Stone and Wood 2000,Kitamura et al., 2006]. Assume we are given some material M at time t. M may change over time because of intrinsic or extrinsic properties or because of external actions. In the first case we have materials like decomposable food (an intrinsic property) and spare parts of systems like a switch for a device (an extrinsic property). An example of changes due to activities is given by a wooden bar cut in two or a piece of liver that is altered by adding some substance for a laboratory test. We need to introduce a dedicated terminology to talk about these different changes. Recall that we write D(t) to mean the collection of sentence relative to M that the organization has at time t. The fact that D provides information about M and that M, being a physical entity, is never fully grasped in informational terms does not prevent us from saying that D(t) can be considered a complete description of M. The point is that we do not seek absolute completeness; completeness is here relative to the information about M that is considered of value to the organization O. A (somewhat informal) definition of KO-completeness is as follows Definition 2 (KO-complete knowledge objects) An knowledge object (M,D,O) is said to be KO-complete whenever D collects all sentences relative to M that are of value for O’s tasks and activities. Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

94

S. Borgo and G. Pozza / Disentangling Knowledge Objects

What exactly counts as a valuable sentence (or valuable data) with respect to M and O depends on the specific organization, role or agent O and on the type of material M. This aspect is clearly contextual and we do not need to dwell into it. We assume then the following principle

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

Principle 1 (KO-completeness of knowledge objects) Let (M,D,O) be a knowledge object. At each time t in which (M, D, O) exists, a KO-complete description of M is defined, reachable and recognizable by O. Principle 1 needs some explanation. It says that each organization knows the data it needs about M (KO-completeness is defined), that it is in the organization’s capacities to collect these data (KO-completeness is reachable) and that the organization is able to recognize when all these data have been collected (KO-completeness is recognizable). In this perspective, D(t) is the best approximation O has of the KO-complete collection of data about M. As the reader may have anticipated, in the real world things are a more complicated than this. Even if we know that D(t) contains all the information needed to have a KO-complete description of M, it might be the case that some of the information in D(t) is incorrect. We have to deal with the following cases: D(t) is KOcomplete and correct, D(t) is KO-complete and incorrect, D(t) is correct and not KOcomplete, D(t) is neither correct nor KO-complete. Since the organization activities are dynamic, we have the further problem of matching the evolution of M and the changes in D over time. Initially, D contains the data received by the organization O from an external source, often the same source that provides the material M (most of the times, as in the case of IZSVE of next section, the data is assumed to be reliable and perhaps only some rough visual or tactile test is performed). Note that, although M may change in unpredictable ways, the way D evolves over time is fairly easy to control: it depends on the verified changes in M and the (explicit or implicit) protocols adopted by O. For the sake of simplicity, let us assume from now on that the data that D collects is given in a fixed language that is formally defined (in order to be fully accessible, clearly understood, free of ambiguities and vagueness) and sufficiently expressive to capture all the needed information. This is an oversimplification that experts in software engineering would immediately point out, but in principle IT has developed methodologies and tools to manage fairly well with these assumptions. Our next goal is to find qualifying relationships between the elements M and D in a knowledge object (M, D, O) and to isolate its ideal behavior. Let us say that (M, D, O) is correct at time t if D(t) describes correctly M, i.e., any sentence collected in D at t is a piece of correct information about M. We assume that sentences in D are grouped in some way. For instance data may be classified depending on the topic it is about: weight, color, temperature, chemical composition etc. Of course, each organization may determine its own peculiar way to classify data in order to optimize its tasks and goals. Definition 3 (Correctness preservation) Assume that the list of topics relevant to a material M and an organization O has been given. The knowledge object (M, D, O) is said to preserve correctness if whenever all sentences relative to a given topic and collected in D are correct, at any later time D is not updated with new sentences relative to that same topic unless these sentences are also correct.

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

S. Borgo and G. Pozza / Disentangling Knowledge Objects

95

For example, if D(t1 ) contains only correct sentences on the topic temperature (e.g. that M at t0 , with t0 < t1 , is at temperature 4.5o C and that at t1 is at temperature 4o C), then any new sentence that D(t1 + h) contains about temperature must be correct of M (e.g. that M kept temperature 4o C from t1 to t1 +h). The idea is that the data added by the organization O is verified via internal procedures and quality checks that are established or enforced by the organization itself, thus any new sentence is reliable (actually, correct) to the organization itself. This property models the basic assumption that an organization trusts itself on managing its own knowledge objects, an observation that also motivates the following definition in which we relativize KO-completeness to single topics: Definition 4 (Faithfully monotone) We say that a knowledge object (M, D, O) is faithfully monotone if the following holds: (a) (M, D, O) preserves correctness and (b) if D is KO-complete for M at time t with respect to a topic P, then D is KO-complete for M with respect to P at any later time t + h.

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

Faithfully monotonicity captures a natural attitude of organizations: once an organization O creates a new knowledge object (M, D, O), from the perspective of the organization the object (M, D, O) is treated as faithfully monotone. Informally, this means that O enforces any change in D to be determined by substituting wrong sentence with correct ones or by adding new correct sentences, and that if D is KO-complete with respect to a topic at some point in time, O supervises D to remain KO-complete with respect to that topic. As said, what counts as a topic depends on the organization’s goals and tasks since topics, in the way we introduced them here, are just collectors of data. Usually material qualities like weight, color and size are taken as separate topics. Also social or relational qualities like market price and compatibility to standards can be included as topics in D. Sometimes topics are complex qualities like the speed (of a car) which depends on other topics (weight, power, penetration coefficient, road pendency and so on). We ignore this issue here. There are several ways to extend the notion of knowledge object. One can study the relationship between D and D’ in knowledge objects (M, D, O) and (M  , D  , O) where M’ is a generic or specific part of M, e.g. when M is an amount of milk and M  is the portion collected for some lab test or when M is an animal and M  its liver. Analogously, one can explain what happens of two distinct knowledge objects (M, D, O) and (M  , D  , O) when M and M  are fastened or merged, like when pieces of steel are welded together or chemicals get mixed. One could also try to generalize this framework by removing the restriction that M is a material entity. This generalization would provide the means to properly model the variety of public and private service organizations for which M is essentially contextualized data. We leave the analysis of these topics for future work and devote the rest of the paper to study a concrete application of knowledge objects in the IZSVE scenario.

3. A class of structured knowledge objects In this section we focus our attention on a class of knowledge objects to highlight interesting features of these entities and to show how to deal with a particularly complex situ-

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

96

S. Borgo and G. Pozza / Disentangling Knowledge Objects

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

ation. Although these knowledge objects may arise in several contexts, our study looks at how they are generated and dealt with in the context of the Istituto Zooprofilattico Sperimentale delle Venezie (IZSVE), a veterinary public health institute that conducts laboratory controls and research activities in three main areas: animal health and welfare, food safety, and environmental protection. The IZSVE receives biological specimens for planned health security monitoring and runs a series of chemical and/or microbiological analyses on them. An official sample for analyses relative to food safety is received by IZSVE already divided in a number of properly identified items (each called aliquota, pl. aliquote): one aliquota is used for the actual test(s) while the others are kept in the eventuality that the laboratory, the competent authority or someone else legally affected by the outcome of the test(s), requests these to be repeat. In other cases, due mainly to practical difficulties to store material or to collect enough substance, the sample consists of a single item and counterpart technicians are allowed to be present during the test to verify the correctness of the procedure and the outcome. The flow of material and data that initiates with the IZSVE’s reception of a sample and ends with a test report delivered to the interested parties, is quite sophisticated. Fortunately, we need to consider only a fragment of it to introduce and formalize the properties of the knowledge objects we are interested in. The knowledge object, as described earlier, is created when a sample M is received and relative information D is collected (in principle D could be empty but in practice at least date and time of reception are recorded). Formally, a receipt is released to state the official generation of the knowledge object (M,D,IZSVE). Note that the adopted protocols ensures that the material M is uniquely identified (practically, the specimen is stored in a sealed container properly identified by the receiver) and that the entity (M,D,IZSVE) is properly generated. We do not question these protocols here. Once the knowledge object (M,D,IZSVE) is created, it changes under the full control of the organization IZSVE, which indeed controls M and D evolution modulo natural phenomena due, e.g., to the natural decay of M. The IZSVE’s knowledge objects are distinct individuals and we have seen that in many cases M is a complex physical entity decomposed in distinct parts (aliquote), say, M1 , . . . , Mn where generally n = 4 or 5.5 Since each Mi is clearly identified and isolated as a consequence of the reception of M, IZSVE actually generates n + 1 knowledge objects, namely, (M,D,IZSVE), (M1 ,D1 ,IZSVE),. . . ,(Mn ,Dn ,IZSVE). The strong relationship between (M,D,IZSVE) and the other knowledge objects related to it is expected due to the parthood relationship between M and each Mi . Less obvious is the existence of strong relationships among the sub-objects (Mi ,Di ,IZSVE) as we are going to see. 3.1. Information permeability The government authorities regulate the procedures to isolate and identify the aliquote M1 , . . . , Mn from the whole specimen M, so that IZSVE can take the aliquote to be perfectly equivalent for any IZSVE’s purpose relative to the national food safety program. The structuring of M in aliquote is motivated by the need to preserve intact and unaltered some parts of the specimen at stake, parts that must be indistinguishable from M 5 The requirement to have the specimen divided in distinct parts is due to IZSVE’s requests and/or to official

regulations. Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

S. Borgo and G. Pozza / Disentangling Knowledge Objects

97

itself in terms of the properties under investigation. Indeed, it would make no sense to apply a test on the physical element of one aliquota, say aliquota (M1 ,D1 ,IZSVE), to get data about the whole M when suspecting that the aliquota under scrutiny is somehow exceptional with respect to the others. It follows that, since M is composed of several parts, all these parts by assumption must have the same properties values. Thus, when a test is performed on M1 to identify, say, the amount of a certain bacteria in it, the result of the test is used to update D1 and, coherently with the assumptions, the same data is used to update D itself. This is somehow expected since there is a parthood relationship between M1 and M. By recalling the same assumption, the very same data is used to update also all elements D2 , . . . , Dn of the remaining knowledge objects. This means that the knowledge objects (Mi ,Di ,IZSVE) are bound by a form of information permeability: the result of any test run on one of these sub-objects immediately extends to all the others although it is obvious that M1 = Mi for 1 = i and that IZSVE has no direct information about (Mi ,Di ,IZSVE). We call information permeability this specific process of information propagation across distinct knowledge objects; information permeability is clearly a crucial feature that must be modeled very carefully. The reader should notice that information permeability is not specific to IZSVE’s knowledge objects. For instance, in manufacturing it is exemplified by the application of quality tests on a restricted subset of the produced items. Information permeability comes in different forms: upward permeability goes from D1 , the collection of data relative to the tested aliquota M1 , to D as in the case of tested biological properties (the checked presence of bacteria in M1 is used to update D1 and, by permeability, D); downward permeability goes from D to each Di (information about origin, ownership, type of material of M is inherited by each Di from D); horizontal permeability goes from D1 to each Di as the result of the transitivity of this propagation process. Finally, permeability is coherent with the ontological dependence relationships among the knowledge objects and among their elements, cf. Figure 1. This observation can be emphasized by ideally dividing D into two parts. One part contains the data originally furnished to IZSVE, like the origin and ownership of M, that is inherited by the knowledge object relative to each aliquota. These data propagate by downward permeability due to the ontological dependence of the aliquota from the specimen. The other part of D contains the information gathered by IZSVE via its laboratories’ activity on the tested aliquota (M1 ,D1 ,IZSVE). The data collected in this way are inherited by (M,D,IZSVE) and by each (Mi ,Di ,IZSVE) in two steps: first upward permeability due to the parthood relationship between M and M1 , and then downward permeability because of the ontological dependence of each aliquota on the specimen, cf. Figure 1. 3.2. Information non-monotonicity The case we have just seen is fairly complex when one aims to formalize the interactions among the different elements. Still, there are more challenging cases. In this last part of the paper we discuss another example in the IZSVE scenario which is more complex and that is again about knowledge objects built out of strictly related physical elements. As before, we do not need to introduce the full picture, so we limit the presentation to some general aspects without lingering over many details. Technically, the term ‘sample’ means “a set composed of one or several units or a portion of matter selected by different means in a population or in an impor-

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

98

S. Borgo and G. Pozza / Disentangling Knowledge Objects

M Specific Constant Dependence

part_of

......

M1

......

Mi

D downward permeability

upward permeability D1 (test results)

......

Di

Specific Constant Dependence

......

horizontal permeability

(M, D, IZSVE) Specific Constant Dependence

Specific Constant Dependence

(M1, D1, IZSVE)

......

(Mi, Di, IZSVE)

......

Mutual Generic Dependence

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

Figure 1. Some relationships among IZSVE’s knowledge objects of specimen and of its aliquote

tant quantity of matter, which is intended to provide information on a given characteristic of the studied population or matter and to provide a basis for a decision concerning the population or matter in question or concerning the process which has produced it” (from [Commission Regulation (EC) No 2073/2005], Art. 2, Lett. (j)). Also, in the same European regulation we read that a ‘batch’ is “a group or set of identifiable products obtained from a given process under practically identical circumstances and produced in a given place within one defined production period” (from [Commission Regulation (EC) No 2073/2005], Art. 2, Lett. (e)). In short, the cited European regulation gives us the technical background for testing multiple samples which are all relative to a single batch. Indeed, under the food safety regulations IZSVE is requested to perform the same test contemporarily on distinct samples, called sample units, and to outcome a single result which is then referred to the whole batch. Formally, we are dealing with at least three types of knowledge objects6 : the knowl6 There is a fourth type of knowledge object which is obtained when the aliquote of the different sample units

are regrouped to ensure all sample units are covered by the test. However, knowledge objects in this group are Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

S. Borgo and G. Pozza / Disentangling Knowledge Objects

99

edge object relative to the batch, a knowledge object relative to each sample unit forming the batch, and a knowledge object relative to each aliquota of each sample unit. Following the previous example, the parthood relationship among the physical elements are not hard to detect and so are the forms of dependences among the knowledge objects. As we said, the results of the tests are given in a unique report which is about the whole batch, i.e., the test results are not divided according to the sample units. In the cases that interest us here, e.g. when testing the presence of histamine, a molecule, a test result may fall within three categories: the test gives a “satisfactory” result (the substance concentration is detected below the given minimal threshold), the test gives a “risk” result (the substance concentration is detected between the given minimal and maximal thresholds), the test gives an “unsatisfactory” result (the substance concentration is detected above the given maximal threshold). The information used to update the knowledge object relative to the whole batch depends on these results in a way that differs from what we have seen in Section 3.1. If the test on a single unit gives result “unsatisfactory”, it triggers an upward permeability process and the whole batch is classified unsatisfactory; if the test on a single unit gives result “satisfactory”, it does not trigger any upward permeability process; if the test on a single unit is within the min/max range, it does not trigger any upward permeability process either. That is, if none of the results on the single units is above the maximal threshold, no information propagation occurs. In this case, IZSVE’s results are interpreted by the responsible authority according to an official regulation that establishes, depending on the substance detected, if the percentage of results that are within the min/max range (risk range) is acceptable: if so, the whole batch is classified satisfactory, otherwise unsatisfactory. Thus, upward permeability takes place after this extra step and, differently from Section 3.1, the data used to update the informational element of the batch is not the data that has been used to update the informational element of a tested knowledge object. The propagated information corresponds instead to a property of the set of all outcomes. In other words, upward permeability does not propagate from a knowledge object to another knowledge object; it goes from a set of knowledge objects to single knowledge object. This phenomenon shows a distinct way to propagate data across knowledge objects since it introduces forms of non-monotonic behavior: if the number of results is above the allowed percentage, the informational element of the whole batch is updated by a sentence like “the batch is classified unsatisfactory with respect to the histamine level” and all the related knowledge objects are similarly updated (i.e., classified unsatisfactory with respect to the histamine level) by downward permeability, no matter if in the actual test they were classified satisfactory; if the number of results is below the allowed percentage, the informational element for the whole batch is updated by a sentence like “the batch is classified satisfactory with respect to the histamine level” and all the knowledge objects are updated accordingly by downward permeability, no matter if in the actual test they were classified within the min/max range, see [Commission Regulation (EC) No 2073/2005]. formally modeled using the same relationships we already introduce for the other types. Thus, for the sake of simplicity, we ignore this further group.

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

100

S. Borgo and G. Pozza / Disentangling Knowledge Objects

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

4. Conclusions We proposed to identify and analyze knowledge management concepts via techniques based on ontological analysis in order to highlight crucial elements, properties and relationships. The introduction of knowledge objects is a first attempt to build knowledge management notions based on ontological analysis. Knowledge objects, in our definition, are new types of entities that emerge from the interaction of the material, the informational and the organization’s perspectives. The properties and the role of these entities have been exploited to some degree. The capacity of knowledge objects to model important aspects of knowledge management and their usefulness to develop and formalize a proper framework will be seen in the future as this approach is studied in more depth. As the study continues, some limitations should be lifted, like the restriction of M to material entities and the restrictive view of organizations as hierarchies of roles and goals. Although further considerations are needed to test and enhance this work, we believe that the approach goes in the right direction in putting knowledge objects at the centre of research and it makes room for the development of other ontologically motivated notions and techniques. Regarding the properties of the IZSVE’s knowledge objects that we studied, we already pointed out some specific properties and discussed a few open issues. We conclude the paper highlighting one more: the real extent of entity M. On the one hand, it is natural to take M to be the full amount of specimen that IZSVE is asked to test. As we have seen, this full amount is sometimes split in parts forming the aliquote that, all things considered, are parts indistinguishable from the whole specimen M. On the other hand, IZSVE receives the target specimen already stored in one or more containers (which IZSVE receives together with the specimen) and it is IZSVE’s responsibility to establish the suitability and appropriateness of the containers themselves. This fact suggests that IZSVE should take M to be both the specimen and the container(s) where it is stored, i.e., the full material entity that is practically delivered to IZSVE. The first view is to be preferred from a theoretical perspective since IZSVE purpose is to collect and analyze the specimen. The second view is correct from the operational viewpoint since IZSVE has to properly manage and consider the whole entity constituted by the specimen plus the container(s) that store it. Acknowledgements The authors thank the reviewers for their comments which helped to improve the presentation of the material. This paper contributes to the IZSVE 19/08 RC project “Analisi e definizione di una metodologia operativa per migliorare la gestione dei dati relativi al conferimento di campioni all’Istituto Zooprofilattico delle Venezie, con particolare riferimento alla sicurezza alimentare” (funded by the Italian Ministero del Lavoro, della Salute e delle Politiche sociali). References [Bolisani and Oltramari 2009] Ettore Bolisani and Alessandro Oltramari. Capitalizing flows of knowledge: Models and accounting perspectives. In 4th International Forum on Knowledge Asset Dynamics – IFKAD. University of Glasgow, February 2009. Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

S. Borgo and G. Pozza / Disentangling Knowledge Objects

101

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

[Borgo and Masolo 2009] S. Borgo and C. Masolo. Foundational Choices in DOLCE. Handbook on Ontologies, 2nd edition, Springer, (to appear). [Chandrasekaran and Josephson 2000] B. Chandrasekaran and J.R. Josephson. Function in Device Representation. Engineering with Computers, 16(3/4):162–177, 2000. [Commission Regulation (EC) No 2073/2005] Commission Regulation (EC) No 2073/2005 of 15 November 2005. Microbiological criteria for foodstuffs. Official Journal L 338, 22/12/2005, 0001-0026. [Kitamura et al., 2006] Y. Kitamura, Y. Koji, and R. Mizoguchi. An ontological model of device function: industrial deployment and lessons learned. Applied Ontology, 1(3-4):237–262, 2006. [Masolo et al., 2002] C. Masolo, S. Borgo, A. Gangemi, N. Guarino, A. Oltramari, and L. Schneider WonderWeb Deliverable D17: The WonderWeb Library of Foundational Ontologies 2002 [Masolo et al., 2004] C. Masolo, L. Vieu, E. Bottazzi, C. Catenacci, R. Ferrario, A. Gangemi, and N. Guarino. Social roles and their descriptions. In D. Dubois, C. Welty, and M.A. Williams, editors, Proceedings of the 9th International Conference on the Principles of Knowledge Representation and Reasoning (KR), pages 267–277, 2004. [Mentzas et al., 2001] G. Mentzas, D. Apostolou, R. Young, and A. Abecker. Knowledge networking: a holistic solution for leveraging corporate knowledge. Journal of Knowledge Management, 5(1):94–107, 2001. [Pozza et al., 2009] Giandomenico Pozza, Stefano Borgo, and Licia Ravarotto. From data to knowledge objects, ontological considerations with inputs from the public health domain. Proceedings of the 10th European Conference on Knowledge Management, Vicenza 3-4 Sept. 2009. [ Stone and Wood 2000] R.B. Stone and K. Wood. Development of a functional basis for design. Journal of Mechanical Design, 122(4):359–370, 2000.

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

102

Formal Ontologies Meet Industry R. Ferrario and A. Oltramari (Eds.) IOS Press, 2009 © 2009 The authors and IOS Press. All rights reserved. doi:10.3233/978-1-60750-047-6-102

Do you still want to vote for your favorite politician? Ask Ontobella! Pawel Garbacz, Marek Lechniak, Piotr Kulicki and Robert Trypuz Department of Philosophy The John Paul II Catholic University of Lublin Abstract. The paper presents the preliminary version of Ontobella - a domain ontology of beliefs. The philosophical assumptions of this system are taken from the philosophy of Roman Ingarden and from the psychological results obtained in the Lvov-Warsaw school. Ontobella is applied as the conceptual framework for a computer system that collects information about political debates. We use it to retrieve and store beliefs expressed during sessions of the Polish parliament.

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

Keywords. ontology, belief, intentionality, politics

Representation of knowledge and beliefs plays a crucial role in Artificial Intelligence, especially in modeling artificial agents being able to carry out actions on the basis of their knowledge about environment and goals [12,8] and in automated planning [7]. Beliefs are largely studied in epistemic logic, where many formal systems of representation of private, public and common beliefs have been proposed. One special interest is a question of belief revision formally described by the well-known AGM model [1]. Finally, the important role of knowledge representation in argumentation and persuasion is evident. Nonetheless, ontological foundations of these approaches remain unclear. The domain of beliefs did not attract a lot of attention in ontological engineering despite the substantial research in psychology and philosophy. The authors of this contribution are not aware of any domain specific ontology of beliefs except for the COM ontology. This paper presents a slightly different approach to modelling the domain of beliefs. Section 1 describes the results we have achieved so far in the form on the Ontobella ontology. Section 2 is devoted to one of the possible applications of such ontologies, while the last section compares Ontobella with the COM ontology. The formal shape of our system is outlined in the Appendix.

1. Main ontological assumptions of Ontobella Our ontology of beliefs, Ontobella, has two main sources of inspiration: 1. R. Ingarden’s ontology of intentional object (cf. [3]), in particular: • the distinction between autonomous and heteronomous entities,

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

P. Garbacz et al. / Do You Still Want to Vote for Your Favorite Politician? Ask Ontobella!

103

• the conception of intentional objects as contents of mental acts, • the definition of beliefs as those mental acts the represent situations together with his theory of situations/states of affairs (Sachverhalten). 2. the psychological legacy of the Lvov-Warsaw school (the so-called ”descriptive psychology”–cf. [10]), in particular: • the conception of beliefs as perdurants, • the thesis that representational object exhibit (mental) content, • the distinction between assertions and rejections as non-reducible propositional attitudes, • the definition of memory and expectation. However, due to the lack of space, we have not elaborated here on these threads and confine our presentation to the results we took over from these traditions. 1.1. Ontological neutrality of Ontobella

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

The Ontobella ontology is a domain-specific ontology, so it presupposes a number of general categories like those of endurant, state of affairs, time etc. Although we share the conviction that the precise understanding of these notions is crucial to the task of ontological engineering, we do not want to restrict our ontology of beliefs to some specific upper-level ontology. We believe that any sufficiently broad upper level ontology has concepts that subsume the most general concepts of Ontobella. The following table shows the correspondence between Ontobella and a sample of the upper-level ontologies. The correspondence shown is either of the type of subsumption (i.e. a given Ontobella category is subsumed by the respective category from an upper-level ontology) or identity. Ontobella SimpleThing Situation occursIn Endurant Perdurant dependsOn ≺ participates Agent a The

DOLCE Particular Situationa Endurant Perdurant specific constant dependence proper parthood constant participation Agentive Physical Object

GFO Individual Configuration Presential Occurent dependence (abstract) part-of relation exhibit -

BFO ontology Entity constituent of (cpart) Continuant Occurent specific constant dependence proper parthood complete participation -

SUMO entity Object Process properPart cognitive agent

category of situations is to be found in an extension of DOLCE named ”DOLCE+” (cf. [6]). Table 1. Ontobella and upper-level ontologies

On the other hand, we use these upper-level notions to define certain less specific ones, usually following the standard patterns (see definitions 1, 2, 3). The only exception to our ”ontological indifference” is the conception of time. The idea we refer to is borrowed from the ISO 15926 ontology where a time object (be it an instant or period) is the mereological sum of a set of perdurants (cf. [11]). In our ontology we assume that there exists a set of time objects (TimeThing) that are perdurants (axiom 6) and are ordered by means of the relation of being earlier than (axiom

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

104

P. Garbacz et al. / Do You Still Want to Vote for Your Favorite Politician? Ask Ontobella!

7).1 This solution allows us to define the ternary temporal relations with the help of their binary counterparts (cf. definitions 4, 36, 31, 34, and 35). We do not, however, accept the four-dimensional perspective of the ISO 15926 ontology. 1.2. Specific categories of Ontobella Ontobella is a first-order theory that is determined by the axioms and definitions given in the Appendix. Here we briefly describe its most important tenets in the natural language. Moreover, there exists an OWL version of Ontobella, available at www.trypuz.ovh.org/ontobella/ontobella.owl. We assume that beliefs are perdurants, i.e. that they extend in time. This ontological categorisation takes into account the following intuitions:

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

• our beliefs start and end at certain points in time, • our beliefs have phases, i.e. they are not ”fully-feathered” when that come into existence, but they develop in the course of time, • our beliefs are parts of our mental lives, • although some of our beliefs are composed of other beliefs, it is not the case that any collection of beliefs corresponds to a belief. Beliefs are those perdurants in which certain agents participate. The current axiomatisation aims at modelling only individual beliefs, so any belief has exactly one agent which participates in this belief. This relation between the agent and his or her belief is represented by the “hasBelief” predicate. Except for axiom 26, we do not characterise the notion of agent in more detail. A belief is a representational object, that is to say, due to its ontological status, it represents other objects. Among other representational objects, beliefs may be isolated as those objects that represent situations/states of affairs (Situation). As all representational objects, beliefs represent something because of their (mental, or even better intentional) content. The relation of representation is divided into two components: hasContent and directRepresents. The first relation links beliefs, and, for that matter, all other representational objects, to their contents.2 Then, each piece of mental content directly represents (directRepresents) the respective piece of reality, i.e. situations/states of affairs in the case of beliefs. This informal description of beliefs as perdurants that represent situations is captured by definitions 13 and 16 together with axiom 13. Figure 1 outlines the main elements of this view on beliefs. In sum, when an agent has a belief, then by means of this belief he or she refers towards a certain mental content (cf. definition 21) and thereby believes that a certain situation is the case (definition 18). Finally, if an object occurs in this situation, then we say that the belief at stake concerns the object (definition 15). 1 The

notion of time object is taken as primitive here, but one can define it provided that one is able to define the relation of simultaneousness on the set of all perdurants. Then each time object is the mereological sum of an abstraction class of this relation. Thus, a time object may be either an instant or a period. Our time objects coincide with the so-called historical closures from [11]. 2 We found that R. Ingarden’s notion of intentional objects is particularly useful to define contents of beliefs, but the proper exposition of his theory goes far beyond the scope of the present paper. Nonetheless, the notion of specific content dependence might serve as the upper-level super-concept for hasContent (cf. axioms 10 and 11).

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

P. Garbacz et al. / Do You Still Want to Vote for Your Favorite Politician? Ask Ontobella!

105

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

Figure 1. Representation of beliefs in Ontobella

Axiom 15 gives the criterion of identity for beliefs: one agent cannot have two different beliefs if these beliefs share content and have the same temporal span. All beliefs are ephemeral objects, but some of them are stored by means of written or spoken utterances. We assume that belief storages (cf. definition 32) are created for the sake of storing beliefs, so the former store the latter as long as they exist (axiom 19). Moreover, since some belief storages are perdurants, we exclude the possibility that a belief is its own storage (axiom 20). Among such storage objects, we distinguish what we call transparent storage objects, i.e. those written or spoken utterances that exhibit such features by means of which certain agents are able to identify those of entities as storage objects. For example, the string “In my opinion Warsaw is more beautiful than London” is a transparent storage object in contrast to the string “Warsaw is more beautiful than London”. We do not define the category TranspBeliefStorage of such transparent storage objects (see, however, axiom, 28), but we need it for the application purposes - see below. If a belief is stored in a storage object, we call the former a stored belief (definition 33). When an agent entertains a certain belief, this belief is either accepted or rejected by him or her (cf. axiom 14). The notion of belief (definition 17) is broad as it includes all propositional doxastic attitudes, including those where we hardly believe or disbelieve something. We are able to order both beliefs and belief storages according to their temporal position in time by means of the relation ⇒T that links our time objects.3 Definitions 11, 7, 8, and 9 extend this relation to all entities that exist in time, including situations. It is a psychological fact the beliefs support (or motivate) one another. Sometimes this relation is based on the deductive inference: I believe that every politician lies because I disbelieve some politicians do not lie. In most cases, however, the psychological fact that I believe so and so is supported by other beliefs (of mine) but the support at stake has little to do with logic. Moreover, it might be the case that my belief is supported 3x

⇒T y means that (time object) x is earlier than (time object) y.

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

106

P. Garbacz et al. / Do You Still Want to Vote for Your Favorite Politician? Ask Ontobella!

not only by my other beliefs, but also by some other mental objects, e.g. emotions, or even external factors, e.g. someone’s behaviour. In order to model this relation of support, we use the relation supports. We assumed the following minimal characterisation of this relation: • all objects in the field of the supports relation are perdurants (axiom 21), ∗ So neither John nor his suit can persuade Ann to buy a new model of Volvo. If she happens to be persuaded by John, we model this by selecting certain perdurant in which John participates, say his eloquent speech or his broad smile, and claim that it is this perdurant that supports Ann’s new belief that she ought to buy this Volvo. • other agents’ beliefs do not directly support my beliefs; if they do, then it is mediated by the behaviour of those agents (axiom 22), • direct supports does not span over time: if one perdurant directly supports a belief, then their temporal spans overlap (axiom 23), • no belief can support itself (axiom 24), • supports is transitive, at least over the belief set of one agent (axiom 25). Axioms 24 and 25 presuppose that the relation of belief support has certain flavour of rationality. If you keep on repeating “p because p”, you do not support anything in this sense. 1.3. Expressivity of Ontobella

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

Using the aforementioned primitive notions, we are able to define the following concepts and relations: 1. the relation that links beliefs with the objects these beliefs are about (definition 15), 2. the concept of second-order belief (definition 19), 3. the relation by means of which two beliefs have the same content (definition 14), 4. relation by means of which two agents believe in the same content (definition 22), 5. the concept of plain, i.e. non-ambiguous, belief (definition 23), 6. the concept of self-awareness (definition 25), 7. the concept of beliefs that represent past, present, and future situations (definitions, respectively, 26, 27, and 28), 8. the concept of memory and expectation (definitions 29 and 30).

2. Ontobella as a conceptualisation of political debate We apply the Ontobella ontology to model the domain of political debate or rather the noetic aspect of this debate, i.e. we are interested in the beliefs expressed during such debates. As a case study, we choose to represent the debates that took place in the Polish Parliament (i.e. Sejm of the Republic of Poland). We built a database that contains the representations of the beliefs expressed during the sessions of the Parliament. The schema of this database is shown in fig. 2. It is

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

P. Garbacz et al. / Do You Still Want to Vote for Your Favorite Politician? Ask Ontobella!

107

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

Figure 2. Schema of the Ontobella database.

in the form of an entity-relationship diagram presented using Barker’s “crow foot” notation. Entities BELIEF_STORAGE, BELIEF and AGENT are directly translated from Ontobella categories. Entity DEPENDENCE is a result of the technically motivated reification of the Ontobella’s concerns relation. Entity NON_AGENTIVE_OBJECT along with entity AGENT plays the role of a dictionary of objects that may occur in situations that beliefs indirectly represent. Entities PARLIAMENTARY_SESSION and PARLIAMENTARY_SPEECH represent the structure of parliamentary processes and their documentations from which the information is retrieved. The schema has been defined on the base of the Ontobella ontology, however, we deployed only a tiny part of the whole theory. The database in question will be part of the computer system that retrieves the information about the beliefs expressed in the political debate, stores them in the database, and allows the user to find the information he or she needs. The architecture of this system is shown in fig. 3. The source of the relevant data are the parliamentary reports published at http://orka2.sejm.gov.pl. In order to automate this process, we defined an algorithm that finds within those reports the pertinent TranspBeliefStorage and stores them in the database. This algorithm is based on the notion of text pattern, which was successfully employed in the automated classification of sentences in legal documents (see [4]). However, the implementation of this algorithm is still ahead of us.

3. Ontobella vs COM Computed Ontology of Mind (in short COM) [5] is a formal ontology developed by two researchers, R. Ferrario and A. Oltramari, from the Laboratory for Applied Ontology.

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

108

P. Garbacz et al. / Do You Still Want to Vote for Your Favorite Politician? Ask Ontobella!

Figure 3. Computer system retrieving the information about the beliefs expressed in the political debate.

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

As far as we know it was the first ontology of mental states built with the intention of application in information systems and, until Ontobella has been created, it remained the only one. Thus it seems appropriate to compare these two ontologies. Let us start with a short introduction to COM. The authors of COM claim that it is inspired by J. A. Fodor’s Computational Representation Theory of Mind and the BDI approach. COM is grounded in DOLCE [9] and shares its ontological choices. Moreover, all predicates present in COM—except two temporal relations of precedence and start taken from the theory of time of Allen and Hayes—come from DOLCE. Thus the authors constructed COM by means of two methods: • by introducing definitions of new predicates using the existing ones, • by proposing axioms “deepening” the characterization of DOLCE categories. In COM the main two categories are Mental State and Mental Object. The second one is a subcategory of Non-physical Object and its main characteristic is such that its instances do not “generically dependent [on] a community of agents” [9]. They are two kinds of Mental Object postulated in COM: Percept and Computed Object. Any instance of the first one does not depend on any mental object. The second category is defined in the following way: x is Computed Object if and only if it is Mental Object and there is always some Percept or another Computed Object y such that x is proceeded by (historically depends on) y. Instances of Computed Object “are indirect, namely they are the result of the computational processes that occur every time that an input (external or internal) is processed” [5]. The four subcategories of Computed Object are Computed Belief, Computed Desire and Computed Intention. All kinds of Mental State—there are four of them: Perceptual State, Desire, Intention and Belief —are defined in COM by means of reference to some subcategory of Mental Object and to relation aboutness. For instance Belief is defined in the following way (see D11 in [5]):

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

P. Garbacz et al. / Do You Still Want to Vote for Your Favorite Politician? Ask Ontobella!

109

x is a belief if and only if there exists the agent a, a time t and a computed belief object y such that the mental state x of the agent a is about y at t. Formally aboutness is a four-place predicate defined by means of DOLCE predicates as follows (see D9 in [5]): “a state y of the intentional agent x is about a mental object z at time t if and only if the agent x participates to the state y at t and the mental object z also participates to y at t, being z one-sided specific constant dependent on x”.

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

The temporal extension and existence of mental objects are bound to the duration of mental states that are about them in such a way that: percepts, computed intentions and desires exist as long as the mental states that are about them exist, whereas the computed belief is required to start at the beginning of the belief that is about it. The main differences between Ontobella and COM: 1. Ontobella is an ontology of beliefs and does not characterize other mental states such as intentions or desires which COM does. On the other hand, there are many interesting categories concerning beliefs in Ontobella such as: past, present and future beliefs, expectations, memory, self-awareness or second order beliefs. None of them are present in COM. 2. Belief in COM is a (Mental) State which is a subcategory of Perdurant. In Ontobella Belief is a direct subclass of Perdurant. 3. COM’s mental states are connected by aboutness with objects internal to the agent. COM does not provide any way to go from the “content” of belief (i.e. computed belief) to the objects represented by this content. Ontobella joins beliefs with their contents by means of primitive relation hasContent. Contents represent something “external” (directRepresents). By the relation indirectRepresents in Ontosbella we link a belief with a situation represented by its content. Contents of beliefs (and indirectly beliefs themselves) can be linked to objects which occur in the situations represented by them. 4. Contents of beliefs are not “private” in Ontobella—in the sense that two beliefs (of two distinct agents) may share the same content (see definition 22)—whereas they are “private” in COM. A common feature of the two ontologies in question is that every belief belongs to one and only one agent. Additionally in COM it is also required that for every agent present at time t there is a computed belief dependent on it. Ontobella does not take a position on this issue. 5. It can be proved in COM that belief is always preceded by another mental state. In Ontobella beliefs may be supported by other perdurants, e.g. percepts, emotions, other beliefs, desires, etc. (cf. axioms 21–25).

4. Further work As the reader can easily appreciate, our research is still at an initial phase. Except for the implementation tasks described above, we envisage to pursuing the following problems: 1. In a number of context, including politics, beliefs are role-dependent: an agent has or expresses a belief because he plays a certain role. For example, a MOP

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

110

P. Garbacz et al. / Do You Still Want to Vote for Your Favorite Politician? Ask Ontobella!

presents a certain opinion not as his or her view, but as a view of his or her party. Thus, the adequate ontology of beliefs should account for this dependence. 2. The philosophical theories of beliefs tend to look at them from the point of view of epistemology or logic. Thus, the results obtained may oversimplify the obscure nature of beliefs. Although we tried to avoid this trap, we perceive the need to compare the assumptions we made to the more empirical psychological type of research. In particular, we want to investigate whether one can build software applications to support psychotherapy. The cognitive therapy developed by Aaron Beck [2] seems to be a promising field in this respect. Nevertheless, we believe that the current state of the art in the ontological analysis of beliefs justifies the certain sketchiness of our results.

5. Appendix - Ontobella’s formal outlook Mereology (D1) Perdurant(x) ∧ Perdurant(y) → [x  y  x ≺ y ∨ x = y] (D2) x ◦ y  ∃z(z  x ∧ z  y) Occurrence (A1) occursIn(x, y) → SimpleThing(x) ∧ Situation(y) (A2) Situation(x) → ∃zoccursIn(z, x) (A3) SimpleThing(x) → ∃zoccursIn(x, z)

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

Participation (A4) (A5) (D3) (D4) (D5)

Endurant(x) → ∃yparticipates(x, y) Perdurant(x) → ∃yparticipates(y, x) Endurant(x) → [life(x)  σ(y)participates(x, y)]4 participatesAt(x, y, z)  participates(x, y) ∧ TimeThing(z) ∧ y ≺ z agentIn(x, y)  Agent(y) ∧ participates(y, x)

Time Thing (A6) TimeThing(x) → Perdurant(x) (A7) ⇒T is a strict partial order in set TimeThing. (D6) hasTemporalSpanOf(x, y)  [TimeThing(y) ∧ x  y] ∧ ∀z[TimeThing(z) ∧ x  z → y  z] (A8) Perdurant(x) → ∃yhasTemporalSpanOf(x, y) (D7) Perdurant(y) → [x = spanperd (y)  hasTemporalSpanOf(y, x)] (D8) Endurant(x) → [spansim (x)  spanperd (life(x))] Perdurant(x) → [spansim (x)  spanperd (x)] (D9) Situation(x) → [spansit (x)  spanperd (σ(y)[Perdurant(y) ∧ occursIn(y, x)]) denotes the mereological sum of the set of all entities α that satisfy the condition φ. The consistency of 3 is guaranteed by the formal properties of relation ≺. 4 ’σ(α)φ’

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

P. Garbacz et al. / Do You Still Want to Vote for Your Favorite Politician? Ask Ontobella!

111

(D10) SimpleThing(x) → [span(x)  spansim (x)] Situation(x) → [span(x)  spansit (x)] (D11) x ⇒ y  span(x) ⇒T span(y) (A9) x1 ≺ x2 → [hasTemporalSpanOf(x1 , y1 ) ∧ hasTemporalSpanOf(x2 , y2 ) → y1 ≺ y2 ] Content and representation (A10) (A11) (A12) (D12) (D13) (D14) (D15)

hasContent(x, y) → dependsOn(y, x). hasContent(x, y) → IntentionalThing(y) ∧ ∃zagentIn(x, z). directRepresents(x, y) → IntentionalThing(x) IntentionalThing(x)  ∃yhasContent(y, x) indirectRepresents(x, y)  ∃z[hasContent(x, z) ∧ directRepresents(z, y)] shareContent(x, y)  ∃z[hasContent(x, z) ∧ hasContent(y, z)] concerns(x, y)  Belief(x) ∧ ∀z[indirectRepresents(x, z) → occursIn(y, z)]

Belief (A13) hasBelief(x1 , y) ∧ hasBelief(x2 , y) → x1 = x2 . (D16) hasBelief(x, y)  agentIn(y, x)∧∃zhasContent(y, z)∧∀z[indirectRepresents(y, z) → Situation(z)] (A14) hasBelief(x, y) ≡ asserts(x, y) ∨ rejects(x, y). (A15) hasBelief(x, y1 ) ∧ hasBelief(x, y2 ) ∧ shareContent(y1 , y2 ) ∧ spanperd (y1 ) = spanperd (y2 ) → y1 = y2 . (A16) hasBelief(x, y) ≡ ∃z[ThatAgentHasBelief(z) ∧ ∀v(occursIn(v, z) → v = x ∨ v = y)]. (D17) Belief(x)  ∃yhasBelief(y, x) (A17) Belief(x) → ¬TimeThing(x). (D18) believesThat(x, y)  ∃z[hasBelief(x, z) ∧ indirectRepresents(z, y)]

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

Second order, self and shared belief (D19) (D20) (D21) (D22)

SecondOrderBelief(x)  Belief(x) ∧ ∃y[Belief(y) ∧ concerns(x, y)] selfBeliefOf(x, y)  hasBelief(y, x) ∧ concerns(x, y) believesIn(x, y)  ∃z[hasBelief(x, z) ∧ hasContent(z, y)] shareBelief(x, y, z)  believesIn(x, z) ∧ believesIn(y, z)

Self-awareness (D23) PlainBelief(x)  Belief(x) ∧ ∃!yindirectRepresents(x, y) (D24) PlainBelief(x) → [y  representedby(x) ≡ indirectRepresents(x, y)] (D25) PlainBelief(x) → {SelfAware(x)  ThatAgentHasBelief(representedby(x)) ∧ ∧ ∀y[hasBelief(y, x) → occursIn(y, representedby(x))]} Past, present and future belief (D26) PastBelief(x)  Belief(x) ∧ ∀y(indirectRepresents(x, y) → y ⇒ x). (D27) PresentBelief(x)  Belief(x)∧∀y(indirectRepresents(x, y) → ¬x ⇒ y ∧¬y ⇒ x).

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

112

P. Garbacz et al. / Do You Still Want to Vote for Your Favorite Politician? Ask Ontobella!

(D28) FutureBelief(x)  Belief(x) ∧ ∀y(indirectRepresents(x, y) → x ⇒ y). Memory and expectation (D29) remembersThat(x, y)  ∃z, v{[hasBelief(x, z) ∧ indirectRepresents(z, y) ∧ PastBelief(z)] ∧ [hasBelief(x, v) ∧ indirectRepresents(v, y)] ∧ v ⇒ z} (D30) expectsThat(x, y)  ∃z[hasBelief(x, z)∧indirectRepresents(z, y)∧FutureBelief(z)] Storage isStoredIn(x, y) → Belief(x) ∧ [Endurant(y) ∨ Perdurant(y)] isStoredIn(x, y) → isStoredAt(x, y, spansim (y)). isStoredIn(x, y) → x = y. isStoredAt(x, y, z)  isStoredIn(x, y) ∧ TimeThing(z) ∧ ∧ [Endurant(y) → life(y) ◦ z] ∧ [Perdurant(y) → y ≺ z]. (D32) BeliefStorage(x)  ∃yisStoredIn(y, x). (D33) StoredBelief(x)  ∃yisStoredIn(x, y)

(A18) (A19) (A20) (D31)

Ternary expansions of binary relations (D34) assertsAt(x, y, z)  asserts(x, y) ∧ TimeThing(z) ∧ y ≺ z (D35) rejectsAt(x, y, z)  rejects(x, y) ∧ TimeThing(z) ∧ y ≺ z (D36) hasBeliefAt(x, y, z)  hasBelief(x, y) ∧ TimeThing(z) ∧ y ≺ z Support

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

(A21) (A22) (A23) (A24) (A25)

supports(x, y) → Perdurant(x) ∧ Belief(y). supports(x, y) ∧ Belief(x) → ∃z[hasBelief(z, x) ∧ hasBelief(z, y)]. supports(x, y) ∧ ¬∃z[supports(x, z) ∧ supports(z, y)] → span(x) ◦ span(y). ¬supports(x, x). Belief(x) → [supports(x, y) ∧ supports(y, z) → supports(x, z)].

Other taxonomical relations (A26) Agent(x) → Endurant(x). (A27) ThatAgentHasBelief(x) → Situation(x). (A28) TranspBeliefStorage(x) → BeliefStorage(x).

References [1] [2] [3] [4]

[5]

C. E. Alchourròn, P. Gardenfors, and D. Makinson. On the logic of theory change: Partial meet contraction and revision functions. Journal of Symbolic Logic, 50, 1985. J.S. Beck. Cognitive Therapy: Basics and Beyond. New York: Guilford, 1995. Arkadiusz Chrudzimski, editor. Existence, Culture, and Persons. The Ontology of Roman Ingarden. ontos-verlag, 2005. Emile de Maat and Radboud Winkels. Automatic classification of sentences in dutch laws. In E. Francesconi, G. Sartor, and D. Tiscornia, editors, Legal Knowledge and Information Systems. Jurix 2008: The Twenty First Annual Conference, pages 207–216, 2008. Roberta Ferrario and Alessandro Oltramari. Towards a Computational Ontology of Mind. In FOIS 2004, Torino, 2004.

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

P. Garbacz et al. / Do You Still Want to Vote for Your Favorite Politician? Ask Ontobella!

[6]

[7] [8] [9]

[10] [11]

Aldo Gangemi, Maria-Teresa Sagri, and Daniela Tiscornia. A constructive framework for legal ontologies. In V. R. Benjamins et al., editor, Law and the Semantic Web, number LNAI 3369, pages 97–124. Springer-Verlag, 2005. Malik Ghallab, Dana Nau, and Paolo Traverso. Automated Planning Theory and Practice. Morgan Kaufmann Publishers Inc., 2004. A. Herzig and D. Longin. C&L Intention Revised. In M-A. Williams D. Dubois, Ch. Welty, editor, Principles of Knowleadge Representation and Reasoning. Menlo Park, California, AAAI Press, 2004. C. Masolo, S. Borgo, A. Gangemini, N. Guarino, A. Oltramari, and L. Schneider. The wonderweb library of fundational ontologies and the dolce ontology. wonderweb deliverable d18, final report (vr. 1.0. 31-12-2003). In The WonderWeb Library of Fundational Ontologies and the DOLCE ontology. WonderWeb Deliverable D18, Final Report (vr. 1.0. 31-12-2003), 2003. Artur Rojszczak. From the Act of Judging to the Sentence. The Problem of Truth Bearers from Bolzano to Tarski. Springer Netherlands, 2005. John Stell and Matthew West. A 4-dimensionalist mereotopology. In C. Masolo and L. Vieu, editors, Formal Ontology in Information Systems, pages 261–272, 2004. Michael Wooldridge. Reasoning about Rational Agents. The MIT Press, 2000.

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

[12]

113

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

114

Formal Ontologies Meet Industry R. Ferrario and A. Oltramari (Eds.) IOS Press, 2009 © 2009 The authors and IOS Press. All rights reserved. doi:10.3233/978-1-60750-047-6-114

Supporting the Development of Medical Ontologies David CORSAR a , Laura MOSS a , Derek SLEEMAN a,b Malcolm SIM b Department of Computing Science, University of Aberdeen, Aberdeen, UK b Section of Anaesthesia, Glasgow Royal Infirmary, University of Glasgow, Glasgow, UK a

Abstract. Ontologies are widely used in the biomedical community, which has built standard reference ontologies for various aspects of medicine. These projects have produced broad descriptions of the medical domain, resulting in large, complex ontologies which can be difficult to reuse as part of a focused application. We describe four ontologies which we developed to support intelligent reasoning about a particular medical sub-domain. We also describe how concepts in these ontologies can be aligned with standard reference ontologies to promote interoperability.. Keywords. Intensive Care Unit, ICU, Ontology, Reasoning

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

Introduction Ontologies have recently become “institutionalized” in the biomedical community [23], serving a range of purposes from knowledge sources in specific, single purpose applications built by one research group, through to providing wide ranging descriptions of domains built by international collaborations (discussed in section 1). These projects have exploited the purpose of ontologies which according to Gruber’s definition is to create an “explicit specification of a conceptualization” [6], thereby providing the community with a common representation and vocabulary for describing and analysing data [23]. However, as also noted in [23], these ontologies have become large and complex, which hinders both their manageability from a maintenance viewpoint, and their (re-)use as components of systems. One technique gaining attention from the Semantic Web community for supporting the maintenance, reuse, and evolution of large ontologies is modularisation. One popular approach to ontology modularisation is the extraction of significant modules of related knowledge (required for a particular use) from existing, large ontologies [13,18,4,2,5]; however, the variety and complexity of proposed approaches have been criticised for hampering the reuse process [3]. We recently needed to rapidly build an ontology/ontologies in conjunction with clinicians from Glasgow Royal Infirmary which was capable of supporting a system reasoning about Intensive Care Unit (ICU) time series data. The ontologies were required to provide knowledge about areas such as disorders, treatments, and patient data. Ideally we would have reused (parts of) existing relevant ontologies; however, although various ICU ontologies have been discussed in the literature (including [9,15,12]), none were

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

D. Corsar et al. / Supporting the Development of Medical Ontologies

115

available to us at that time; so the primary option considered was to use a module extraction technique to produce modules containing appropriate concepts from the various standard biomedical ontologies. Several problems were identified with this approach:

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

• A range of module extraction techniques exist, which would have required significant analysis, and understanding. To our knowledge to date there is no guide as to which extraction method is most effective in particular situations. • It can be difficult to identify the concepts relevant to our planned system in the appropriate biomedical ontologies which contain thousands of concepts. • Module extraction techniques typically focus on extracting all the concepts related to particular concepts; this could result in modules consisting of concepts we did not require. • Having extracted a series of modules, relationships would need to be defined between these modules as appropriate for our application. • There is no guarantee that after performing these steps, the resulting ontologies would contain all the knowledge required by our system. • This process would have to be repeated every time a new concept was required by the system. In the information systems context, the standard biomedical ontologies would be referred to as “domain ontologies” [7], that is ontologies provide the vocabulary related to a generic domain (such as medicine). Our system requires the development of “application ontologies” [7], which specialise the concept descriptions in the domain ontologies and further “task ontologies” (which describe a particular type of task, such as diagnosis) to produce concept descriptions specifically for an application. As our development team was small (one person), and rapid development of the system was vital, we decided not to use the reuse-based process outlined above, instead we opted to build our own (relatively small) ontologies for the task. This produced compact ontologies, consisting solely of the knowledge required by the system. This approach obviously has many benefits when: building, using, and editing the ontologies with standard ontology editors (e.g. Protégé 3), which do not handle large ontologies well; maintaining the ontologies is easier, simply extending the relevant ontology when needed; and inferencing using these ontologies is quicker (important for the system’s end user). We do, however, recognise that standardised biomedical ontologies have a role as an important reference point, particularly beneficial for supporting interoperability, and so should be used whenever possible. To achieve this, we developed a simple alignment meta-ontology which enables the definition of correspondences between concepts in our ontologies with concepts in other ontologies. We believe this approach is consistent with the description of FMA and SNOMED-CT as reference terminologies/ontologies [20,21]. We have since analysed the ontologies we developed, generalising and adapting them to produce, we believe, three ontologies which provide a framework for describing the concepts required to perform various types of knowledge-based reasoning in a medical (sub-)domain. We believe that these generic descriptions can be quickly extended to describe a particular medical domain, and so can be used to support the rapid development of knowledge-based systems. As the extended models can be aligned with appropriate standard medical ontologies/knowledge sources, our approach supports easy integration and use of different standard biomedical ontologies by an application. We believe

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

116

D. Corsar et al. / Supporting the Development of Medical Ontologies

that our ontologies can be viewed as an initial step towards providing an ontology design pattern for developing medical application ontologies. This paper is structured as follows: section 1 summarises existing biomedical ontologies; section 2 outlines our ontologies for alignment, human physiology, and medical domains; section 3 describes the use of these ontologies when developing applications; section 4 outlines an existing application using our ontologies; and section 5 provides some conclusions and outlines future work plans.

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

1. Existing Biomedical Ontologies This section provides an overview of four established biomedical ontologies1 , which provide definitions for various medical concepts for reuse or reference. Unified Medical Language System (UMLS), which is produced and distributed by the US National Library of Medicine for the purpose of aiding “the development of computer systems that behave as if they ‘understand’ the meaning of the language of biomedicine and health”[1]. The UMLS Semantic Network consists of subject categories (derived from many different thesauri, classifications and controlled terms) and defines relationships between these categories. The major groups of categories are: organisms, anatomical structures, biological function, chemicals, events, physical objects, and ideas [1]. The UMLS Semantic Network is provided in a relational table format. SNOMED Clinical Terms (SNOMED-CT) [21] is a widely-used clinical terminology that provides a common reference point for the comparison and aggregation of data. Concepts are organized into a hierarchy from the general to the specific and include the following high level categories: clinical finding/disorder, procedure/interventions, observable entities, body structure, organism, substance, pharmaceutical/biological product, specimen, special concept, physical object, physical force, event, environmental/geographic location, social context, and staging/scales. SNOMED-CT currently contains more than 311,000 concepts; each concept can be associated with a concept description; SNOMED-CT contains 800,000 concept descriptions and approximately 1,360,000 relationships between concepts. The Foundational Model of Anatomy (FMA) Ontology is a comprehensive collection of classes and relationships which describe the structural organisation of the human body from the molecular level to major body parts. The FMA is implemented in a frame-based formalism and stored in a relational database; it contains 70,000 distinct anatomical concepts, 110,000 terms, and 170 relationships [17]. The proposed use of the FMA is as a reference ontology for “correlating different views on anatomy, aligning existing and emerging ontologies in bioinformatics ontologies”. GALEN Common Reference Model The GALEN Common Reference Model (part of the Generalized Architecture for Languages, Encyclopaedias and Nomenclatures in medicine (GALEN) project) contains a large open-source ontology for the medical domain formulated in a specialised description logic based language called GRAIL [16]. The GALEN Common Reference Model represents the core concepts in pathology, anatomy and therapeutics, that have widespread use in medical applications. 1 It

is acknowledged that there are numerous other examples of biomedical ontologies.

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

D. Corsar et al. / Supporting the Development of Medical Ontologies

117

Figure 1. Visualisation of the Alignment meta-ontology.

2. A Framework for the Development of Medical Ontologies As discussed in the introduction, using standard biomedical ontologies for specific applications is not easy, mainly due to their large size. This characteristic is to be expected - the goal of these projects is to supply knowledge for a (wide) range of applications, and so they require general models, which consequently become very large and complex. The task therefore becomes that of harnessing the power of these large general ontologies while still building applications quickly. To support this task, we outline three ontologies (expressed in OWL-DL2 ) which can be used as a base for models of medical domains used in knowledge-based systems; further these models can be aligned with other knowledge sources to support interoperability through reuse of general medical ontologies.

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

2.1. Alignment Annotations The alignment meta-ontology (visualised in figure 1) defines a hasAlignment annotation object property, which can be used to define a correspondence between a concept in an ontology and a concept in another knowledge source. The range of the hasAlignment property is the Alignment class, which refers to the other knowledge source (via the inSource property) and the name of the corresponding concept in that other source (via the conceptName property). The Knowledge_Source class stores both the URI and the name of a knowledge source. This allows alignments to be made to other knowledge sources of any type, such as ontologies (for example, GALEN), semantic nets (for example, UMLS), thesauri (for example WordNet), or Web pages (for example, the NCI online browser for SNOMED-CT3 . In cases where the knowledge source is not an ontology, the URI should be set to either the local file or remote/Web location. For example, an alignment associated with a Steroid class may be with the “Steroid (substance)” concept in SNOMED-CT. The SNOMED-CT knowledge source would be represented as an individual of type Knowledge_Source, with name property value “SNOMED-CT” and an appropriate uri property value; the alignment would be represented by an individual of type Alignment, with the conceptName property value of “Steroid (substance)” and inSource property value of the SNOMED-CT knowledge source individual; this Alignment individual would then be set to the value of the hasAlignment annotation property for the Steroid class. The alignment ontology is imported by the three ontologies described below. 2 http://www.w3.org/TR/owl-features/ 3 http://nciterms.nci.nih.gov/NCIBrowser

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

118

D. Corsar et al. / Supporting the Development of Medical Ontologies

Figure 2. Visualisation of the Human Physiology Ontology.

2.2. The Human Physiology Ontology Knowledge of human physiology is often required when reasoning in the medical domain, and so we have defined a basic human physiology ontology (visualised in figure 2) to represent knowledge of organs and organ systems, clinical features, and physiological effects.

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

2.2.1. Describing Organs and Systems The Organ_System and Organ classes are used to represent basic human physiology, with the hasOrgan property associating the organ system with its primary organ. For example, the “respiratory system” (which could be represented as an individual of type Organ_System) has the “lungs” (which could be represented as an individual of type Organ) as the primary organ. The hasAssociatedParameters is used to associate Physiological_Parameters with an Organ. For example, the parameters “FiO2 ” and “SpO2 ” are physiological parameters associated with the “lungs”. 2.2.2. Clinical Features and Physiological Effects The Clinical_Feature class provides a template for describing physiological states such as ‘low MAP’ and ‘high temperature’. The various subclasses of Clinical_Feature each describe a different level/type of feature, such as a very low, low, high, or very high value. For example ‘low MAP’ is an instance of Low_Feature associated with the parameter MAP with a value in the range 51 to 69. The Physiological_Effect class represents different types of effects that occur in the human body; two types of effects have been defined: parameter changes and symptoms. The Parameter_Change class describes changes in physiological parameters; a string description of the change is provided, and the magnitude of the change

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

D. Corsar et al. / Supporting the Development of Medical Ontologies

119

(via the minimumChange and maximumChange properties). The three subclasses of Parameter_Change define the possible directions of the change. The Symptom class represents the physiological symptoms associated with a disorder. A symptom can have various types of indicators: physiological effect(s), clinical feature(s), and/or disorder(s). As there may be more than one symptom associated with a disorder, a weighting can be applied to each symptom to reflect its overall importance to a disorder. Subclasses of Symptoms can be used to reflect the different types of symptom. 2.3. The Medical Domain Ontology The medical domain ontology provides a system with the knowledge specific to a particular medical specialty. Four aspects of a particular medical domain can be described: disorders, treatments, disorder severity scores, and drugs.

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

2.3.1. Describing Disorders The main purpose of the Disorders class is to provide information about commonly encountered medical conditions in a particular medical domain. Depending on the domain and the application in which the ontology is being used, it may be appropriate to describe various types of disorders; this can be achieved by defining subclasses of the Disorder class. Currently there is one such subclass, Medical_Disorders, which has eleven different direct subclasses representing disorders of different physiological systems, each with appropriate subclasses. The properties requiredSymptoms and additionalSymptoms provide information about the clinical signs and symptoms associated with a particular disorder. All of the required symptoms must be observed before a diagnosis can be made; observation of any additional symptoms strengthens the diagnosis. Casual relationships between different medical disorders are frequently observed in some domains, and can be described using the causedBy and causes (inverse) properties. This allows one to define relationships such as the disorder ‘pulmonary oedema’ which can lead to the disorder ‘heart failure’. Finally, the treatments used for a particular disorder are associated with the disorder via the treatments property. 2.3.2. Describing Treatments The Treatment class represents the commonly used treatments in a particular medical domain. There are many different types of treatments across the whole of medicine, and we do not attempt to model each one; rather we expect developers to define subclasses of Treatment as appropriate for the application they are developing. For example, when building our ICU ontology, we defined four types (subclasses) of Treatment: Drug, Procedures, Nutrition, and Fluids. In this implementation, the definition of the Treatment class is minimal, with only the forDisorder property associated with it in order to link a treatment with a disorder. 2.3.3. Describing Disorder Severity Scoring systems are often used to summarise a patient’s condition (for example, abstract labels such as ‘A’, ‘B’, ‘C’, or standard scoring systems such as Apache IV [8] or GlasFormal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

120

D. Corsar et al. / Supporting the Development of Medical Ontologies

gow Coma Score [22]); scoring systems can include a definition of ‘acceptable’ physiological ranges. We represent scoring systems using the Scoring_Stage class in the medical domain ontology. The Scoring_Stage class is associated with a label property, which has the range Rating, this is used to define a particular score/label (which could be, for example “acceptable”). The Scoring_Stage class is also associated with the parameterRange class, which has the range Score_Element. A Score_Element essentially defines a combination of parameter ranges, which (overall) must evaluate to true for the patient to be considered in that particular Scoring_Stage. For example, a rating ‘B’ in the Glasgow scoring system [19] may require (amongst other parameters) that the patient’s oxygen saturation (SpO2 ) be between 94% and 95%.

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

2.3.4. Describing Drugs and Related Details

Figure 3. Visualisation of the drug class from the Medical Domain Ontology.

We have used the Drug class (a subclass of Treatment) (visualised in figure 3) to describe how drugs are used as treatments in the ICU domain, although we believe the class description is generic and can be applied when modelling other medical domains. Features of a drug, such as activeDrugName, alternativeDrugNames, the anticipated length of time between the drug’s administration to a patient and its effect being observed (the drug’s timeToReact) and any contraindications of the drug (i.e. known scenarios (disorders or other treatments) in which a patient should not be given a particular drug) have been modelled in the drug class; in addition, descriptions of a drug’s effects, interactions, and uses are supported. Various types of drug effects are described: expected effects; possible effects, which are effects that may occur; conditional effects that occur under certain conditions, for example, when the drug adrenaline is administered it would not normally be expected

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

D. Corsar et al. / Supporting the Development of Medical Ontologies

121

that a patient’s SpO2 would increase, however, when the patient also has a low cardiac output an increase in SpO2 would be observed; rarely observed effects are very unusual but still theoretically possible; and (unwanted) side effects. In general, an effect is either a physiological effect or a clinical feature. When drugs are given to a patient they can interact with other drugs to produce anticipated and unanticipated physiological effects, and so having knowledge of drug interactions can be important when reasoning in a medical domain. Due to the uncertain nature of drug interactions, they are, in general, difficult to model in OWL ontologies, however we have developed a specific definition of an interaction that would support the reasoning we required. For a particular drug, it is usually known which other drugs can interact with it and the associated nature of the interaction. This is represented by the Drug_Interaction class. Each Drug_Interaction is associated with a drug (hasDrug), the interacting drug (interactsWith), and the physiological effects observed during the interaction (interactionEffect). Occasionally a physiological condition may affect the anticipated response to a drug, and this can be specified using the conditionsAffectingPhysiologicalResponseToDrug property. For example, severe sepsis can result in a patient being unable to respond to the drug noradrenaline The condition which affects the drug can take the form of a clinical feature, e.g. high MAP; physiological effect, e.g. increase in FiO2 4 ; or another medical disorder, e.g. sepsis. Various properties are used to represent different drug doses, as an individual drug may be given at different doses depending on the severity of the disorder. How commonly a drug is used to treat particular disorders can also be described.

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

2.4. The Patient Data ontology The Patient Data ontology (visualised in figure 4) has been designed to model the time series data which is typically collected in medical domains. The ontology defines a Patient_Data class, which represents the patient’s Sessions and Location. The Session class models a treatment session, which in turn links a series of Timepoints; the later describe the Readings for a particular xsd:dateTime. Each Reading has a Parameter and a value. In our experience the vast majority of the information used by clinicians when treating ICU patients is the data collected by the unit. The exceptions being that lab test data

Figure 4. Visualisation of the Patient Data Ontology.

4 Fraction

of inspired oxygen given to the patient.

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

122

D. Corsar et al. / Supporting the Development of Medical Ontologies

and sometimes a diagnosis (reason for patient being in the ICU) are included; such data could be easily recorded through extensions to the patient data ontology.

3. Supporting Application Development/Research

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

We believe the ontologies described above provide a framework for specifying the types of knowledge a knowledge-based system requires to perform reasoning within a medical domain. However, to enable a system to perform such reasoning in a particular medical sub-domain, these ontologies must be extended and/or instantiated to describe the appropriate domain knowledge. We envisage that this framework will enable the rapid development of a knowledge-based system within a medical domain in the following manner: 1. The knowledge engineer becomes familiar with the types of medical domain knowledge that the system requires. 2. The knowledge engineer uses this knowledge to extend our medical and physiological ontologies with further class definitions. 3. The extended ontologies are then instantiated, providing further domain knowledge for the system5 . 4. Once the knowledge engineer is satisfied with their ontologies, the alignment annotation property can be used to define correspondences between the “local” concepts and those in other knowledge sources (for example, the standard reference ontologies discussed in section 1). 5. If necessary, the patient data ontology is instantiated with data from a medical source. This would normally be performed by conversion software which performs this task at runtime. 6. Having produced a suitable domain model (aligned with other knowledge sources) and patient data, a system can be produced using that knowledge. 7. If necessary, this process (particularly steps 1-4) can be repeated at any time to extend the domain knowledge as required.

4. Example Application - Explaining Anomalous Patient Responses to Treatment in the Intensive Care Unit Within the medical domain there are clear expectations of how a patient should respond to treatments administered. When these responses are not observed, this can indicate a serious condition for the patient, and can also be challenging for clinicians to understand why the anomalous responses have occurred. An ontology-driven system implemented for the Intensive Care Unit (ICU) domain is currently being developed to assist clinicians detect anomalous patient responses to treatment and suggesting hypotheses to explain these anomalies. As part of the development of this tool, we have undertaken a study to determine how ICU clinicians identify anomalous patient responses; we then asked further clinicians to provide potential explanations for these anomalies [10], and analysed the high-level reasoning deployed by the clinicians when determining potential explana5 We appreciate that it is likely the first three steps will be performed recursively as the knowledge engineer’s

understanding of the domain (and consequently, the ontologies) evolve. Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

D. Corsar et al. / Supporting the Development of Medical Ontologies

123

tions. Figure 5 provides an example of an anomaly identified by a clinician and figure 6 provides an example of a hypothesis provided in response to the anomaly shown in figure 5.

Figure 5. An example of an identified anomalous response to treatment.

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

Figure 6. An example of an explanation for an anomaly.

To summarise, during the interviews the clinicians appeared to use domain knowledge about treatments, medical conditions, and the desired physiological state of the patient to explain a given anomalous treatment or effect. Domain knowledge was also applied by the clinicians whilst examining the patient’s data to determine new facts, for example, that the patient is suffering from a myocardial infarction, and also to test/eliminate potential hypotheses. This high level reasoning demonstrated by the clinicians was generalised and used as the basis for strategies implemented in a system to suggest hypotheses for anomalous patient responses to treatment. This part of the system uses a knowledge base consisting of the OWL ontologies described previously which have been extended and instantiated for the ICU domain and a Java based program implementing the hypothesis generation strategies extracted from the domain experts’ protocols. Currently, 11 different hypothesis generation strategies have been identified and implemented, which make use of the ICU ontologies and patient data in various ways. For example, the “drug” strategy attempts to explain an anomalous effect by identifying whether another drug that the patient is receiving at the time the anomaly was observed could have the same effect (side effect or expected effect) as the observed anomalous effect (and hence this further drug explains the observed anomaly). To achieve this, the strategy queries the ICU medical ontology for all the individuals of type Drug that have sideEffects or expectedDrugEffect equal to the observed anomalous effect. The patient data that corresponds to the time that the anomaly was observed is then queried to determine if the patient was taking any of the drugs (returned by the first query) at that time. If so, a hypothesis is returned to the user for each of the matching drugs suggesting that each drug could be responsible for the anomalous effect. An initial evaluation of this hypothesis generation system has been performed using 15 test cases, each containing an individual anomaly. In total, the system produce 13 different hypotheses after running the test cases (an average of 0.9 hypotheses per test case), which were presented to and accepted by a senior ICU clinician. The framework proposed in this paper has been successfully used as part of the hypothesis generation tool discussed in this section. It is intended that further applications and subsequent evaluations of the framework will be carried out during the development of a knowledge-based system in a different medical domain, namely the renal dialysis domain.

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

124

D. Corsar et al. / Supporting the Development of Medical Ontologies

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

5. Conclusions and Future Work As previously mentioned, ontologies are widely used in the biomedical domain; several standard reference ontologies for various aspects of biomedicine already exist. These ontologies, although very general, are also very large, which can inhibit their reuse in applications. We have outlined three small ontologies which provide a framework for describing medical concepts and medical domain models. We have also outlined a fourth, alignment ontology (section 2.1) which can be used to align concepts in the developed domain models with those in other knowledge sources; further, we discussed how these ontologies can be used in the development of a new system. Below we outline two aspects of future work using these ontologies/approach. The Architecture for Clinical Hypothesis Evaluation (ACHE) [11] aims to support the performance of different studies using data from multiple sources. This is achieved by using a generic repository for time series medical data from a variety of sites, and a tool which supports transfer of data from a medical source to the repository. As part of the transfer process, users are required to define mappings between the parameter names in the incoming data and those in the ACHE repository; this could easily be updated to allow mappings to concepts stored in a medical domain ontology. A web service is provided for accessing and updating the repository; this could also be updated to use the patient data ontology for exchanging data, enabling any system using data expressed against that ontology to use data provided by ACHE. This would provide future systems with an easy method to acquire data in a supported format, regardless of the data’s original format. Knowledge-based systems often comprise of domain knowledge expressed as an ontology and the procedural (decision making) component expressed as a set of rules; in which case it is necessary to provide some linkage between the two components. The Semantic Web community have been working on formalisms for expressing rules against OWL ontologies, for example SWRL6 . Currently, tools, such as the SwrlJessTab [14] plug-in for Protégé enable the user to define and run rules over instantiated ontologies, however they can be understandably slow when dealing with very large ontologies. Although extensions to our ontologies will likely result in sizeable ontologies, they should be considerably smaller than the available standard reference medical ontologies, and so current rule-based tools should run rules in a more acceptable time. We believe that using our ontologies with existing rule plug-ins will enable developers to quickly build and test rule-based systems in a variety of medical domains. Acknowledgements • Prof. John Kinsella (University of Glasgow) for helpful discussions. • Kathryn Henderson and Jennifer McCallum (CareVue Project, Glasgow Royal Infirmary) & the staff and patients of the ICU Unit, Glasgow Royal Infirmary. • This work was an extension of the routine audit process in Glasgow Royal Infirmary’s ICU; requirements for further Ethical Committee Approval has been waved. • This work has been supported under the EPSRC’s grant number GR/N15764.

References [1]

Unified Medical Language System (UMLS). http://www.nlm.nih.gov/pubs/factsheets/umls.html.

6 http://www.w3.org/Submission/SWRL/

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

D. Corsar et al. / Supporting the Development of Medical Ontologies

[2] [3]

[4] [5]

[6] [7] [8]

[9] [10]

[11]

[12]

[13] [14]

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

[15]

[16] [17] [18] [19]

[20] [21] [22] [23]

125

B. Cuenca-Grau, I. Horrocks, Y. Kazakov, and U. Sattler. Just the right amoung: Extracting modules from ontologies. In Proceedings of the 16th International World Wide Conference (WWW 2007), 2007. M. d’Aquin, P. Doran, E. Motta, and V. Tamma. Towards a Parametric Ontology Modularization Framework Based on Graph Theory. In Proceedings of the Second International Workshop on Modular Ontologies (WOMO 2007), 2007. M. d’Aquin, M. Sabou, and E. Motta. Modularization: a Key for the Dynamic Selection of Relevant Knowledge Components. In Proceedings of the ISWC 2006 Workshop on Modular Ontologies, 2006. P. Doran, V. Tamma, and L. Iannone. Ontology module extraction for ontology reuse: An ontology engineering perspective. In Proceedings of the 2007 ACM CIKM International Conference on Information and Knowledge Management, 2007. T. Gruber. A Translation Approach to Portable Ontology Specification. Knowledge Acquisition, 5(2):199–220, 1993. N. Guarino. Formal Ontology in Information Systems. In N. Guarino, editor, Proceedings of FOIS’98, pages 3–15. IOS Press, June 1998. J. E. Zimmerman MD FCCM, A. A. Kramer PhD, D. S. McNair MD PhD, F. M. Malila RN MS. Acute Physiology and Chronic Health Evaluation (APACHE) IV: Hospital mortality assessment for today’s critically ill patients . Critical Care Medicine, 34:5:1297–1311, 2006. J. M.Juarez, M. Campos, J. Palma, and R. Marin. Computing context-dependent temporal diagnosis in complex domains. Expert Systems with Applications, 35:991–1010, 2008. L. Moss, D. Sleeman, M. Booth, M. Daniel, L. Donaldson, C. Gilhooly, M. Hughes, M. Sim, and J. Kinsella. Explaining Anomalous Responses to Treatment in the Intensive Care Unit. In Proceedings of 12th Conference on Artificial Intelligence in Medicine(AIME’09), 2009. L. Moss, D. Sleeman, J. Kinsella, and M. Sim. ACHE:an Architecture for Clinical Hypothesis Examination. In Proceedings of 21st IEEE Symposium on Computer-Based Medical Systems (CBMS 2008), 2008. N.F de Keizer, A.Abu-Hanna, R.Cornet, J.H.M. Zwetsloot-Schonk, C.P.Stoutenbeek. Analysis and Design of an Ontology for Intensive Care Diagnoses. Methods of Information in Medicine, 38:102–112, 1999. N. Noy and M. Musen. Specifying Ontology Views by Traversal. In Proceedings of the International Semantic Web Conference (ISWC), 2004. M. J. O’Connor, H. Knublauch, S. W. Tu, B. Grossof, M. Dean, W. E. Grosso, and M. A. Musen. Supporting Rule System Interoperability on the Semantic Web with SWRL. In Proceedings of the Fourth International Semantic Web Conference (ISWC2005), pages 974–986. Springer, November 6–10 2005. J. Patrick, Y. Wang, P. Budd, A. Rector, S. Brandt, J. Rogers, R. Herkes, A. Ryan, and B. Vazirnezhad. Developing SNOMED CT Subsets from Clinical Notes for Intensive Care Service. In Proceedings of Health Informatics New Zealand Conference (HINZ) 2008, 2008. A. L. Rector and W.A Nowlan. The GALEN project. Computer Methods and Programs in Biomedicine, 45:75–8, 1994. C. Rosse and J. L.V. Mejino Jr. A reference ontology for biomedical informatics: the Foundational Model of Anatomy. Journal of Biomedical Informatics, 36:478–500, 2003. J. Seidenberg and A. Rector. Web Ontology Segmentation: Analysis, Classification and Use. In Proceedings of the World Wide Web Conference (WWW), 2006. D. Sleeman, A. Aiken, L. Moss, J. Kinsella, and M. Sim. A system to detect inconsistencies between a domain expertŠs different perspectives on (classification) tasks. Studies in Computational Intelligence: Recent Advances in Machine Learning, In press. B. Smith and C. Rosse. The role of foundational relations in the alignment of biomedical ontologies. Medinfo, 107:444–8, 2004. K. A. Spackman and K. E.Campbell. SNOMED RT: A Reference Terminology for Health Care. Proc. of the AMIA Annual Fall Symposium, pages 640–644, 2007. Teasdale, G and Jennett B. Assessment of coma and impaired consciousness. A practical scale. Lancet, 2:81–84, 1974. A.C. Yu. Methods in biomedical ontology. Journal of Biomedical Informatics, 39(3):252–266, 2006.

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

This page intentionally left blank

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

Formal Ontologies Meet Industry R. Ferrario and A. Oltramari (Eds.) IOS Press, 2009 © 2009 The authors and IOS Press. All rights reserved.

127

Subject Index 58 66 102 10 10 58 46 46 10 34 22 34 34 66 78 46 114 102 10 46

knowledge management 90 knowledge object 90 manufacturing 22 mereology 34 modeling 58 modularization 66 NGOSS 78 ontology(ies) 22, 46, 58, 66, 78, 90, 102, 114 operator of change 10 part-whole relation 34 politics 102 Process Specification Language 22 Protégé 58 reasoning 114 technical function 34 telecommuncation 78 top level 46 veterinary public health 90

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

Algerian enterprise architecture belief complex change consistency and coherence criteria domain electromagnetic elementary change engineering ontology first-order logic functional composition functional decomposition heterogeneous information industry standards integration Intensive Care Unit (ICU) intentionality IS design ontology knowledge

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

This page intentionally left blank

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

Formal Ontologies Meet Industry R. Ferrario and A. Oltramari (Eds.) IOS Press, 2009 © 2009 The authors and IOS Press. All rights reserved.

129

Author Index 66 90 114 78 22 46 v 102 10 78 22 58 10 66 102

Kutz, O. Lechniak, M. Mhiri, M. Moss, L. Noureddine, M. Oltramari, A. Pozza, G. Sim, M. Sleeman, D. Tarricone, L. Trypuz, R. Vallone, L. Vermaas, P.E. Vetere, G. Zappatore, M.

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

Bhatt, M. Borgo, S. Corsar, D. de Francisco, D. Delaval, A. Esposito, A. Ferrario, R. Garbacz, P. Gargouri, F. Grenon, P. Grüninger, M. Hadj Tayeb, S. Hamdani, H. Hois, J. Kulicki, P.

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

66 102 10 114 58 v 90 114 114 46 102 46 34 1 46

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

This page intentionally left blank

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

This page intentionally left blank

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

This page intentionally left blank

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

This page intentionally left blank

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,

Copyright © 2009. IOS Press, Incorporated. All rights reserved.

This page intentionally left blank

Formal Ontologies Meet Industry, edited by R. Ferrario, and A. Oltramari, IOS Press, Incorporated, 2009. ProQuest Ebook Central,