Visit https://ebookultra.

com to download the full version and

explore more ebooks

The Semantic Web for Knowledge and Data

Management Technologies and Practices 1st
Edition Zongmin Ma

_____ Click the link below to download _____

https://ebookultra.com/download/the-semantic-web-for-
knowledge-and-data-management-technologies-and-
practices-1st-edition-zongmin-ma/

Explore and download more ebooks at ebookultra.com

Here are some suggested products you might be interested in.
Click the link to download

Advances in Fuzzy Object Oriented Databases Modeling and

Applications Zongmin Ma

https://ebookultra.com/download/advances-in-fuzzy-object-oriented-
databases-modeling-and-applications-zongmin-ma/

A Librarian s Guide to Graphs Data and the Semantic Web

1st Edition James Powell

https://ebookultra.com/download/a-librarian-s-guide-to-graphs-data-
and-the-semantic-web-1st-edition-james-powell/

Web data management S Abiteboul

https://ebookultra.com/download/web-data-management-s-abiteboul/

XML Databases and the Semantic Web 1st Edition Bhavani

Thuraisingham

https://ebookultra.com/download/xml-databases-and-the-semantic-
web-1st-edition-bhavani-thuraisingham/
Semantic Technologies for Business and Information Systems
Engineering Concepts and Applications 1st Edition Stefan
Smolnik
https://ebookultra.com/download/semantic-technologies-for-business-
and-information-systems-engineering-concepts-and-applications-1st-
edition-stefan-smolnik/

Fuzzy Logic and the Semantic Web 1st Edition Elie Sanchez
(Eds.)

https://ebookultra.com/download/fuzzy-logic-and-the-semantic-web-1st-
edition-elie-sanchez-eds/

Federated Query Processing for the Semantic Web 1st

Edition C. Buil-Aranda

https://ebookultra.com/download/federated-query-processing-for-the-
semantic-web-1st-edition-c-buil-aranda/

Knitting the Semantic Web 1st Edition Jane Greenberg

https://ebookultra.com/download/knitting-the-semantic-web-1st-edition-
jane-greenberg/

Microsoft Data Mining Integrated Business Intelligence for

e Commerce and Knowledge Management 1st Edition Barry De
Ville
https://ebookultra.com/download/microsoft-data-mining-integrated-
business-intelligence-for-e-commerce-and-knowledge-management-1st-
edition-barry-de-ville/
The Semantic Web for Knowledge and Data Management
Technologies and Practices 1st Edition Zongmin Ma
Digital Instant Download
Author(s): Zongmin Ma, Huaiqing Wang, Zongmin Ma, Huaiqing Wang
ISBN(s): 9781605660295, 1605660299
Edition: 1
File Details: PDF, 5.88 MB
Year: 2008
Language: english
The Semantic Web for
Knowledge and Data
Management:
Technologies and Practices

Zongmin Ma
Northeastern University, China

Huaiqing Wang
City University of Hong Kong, Hong Kong

InformatIon scIence reference

Hershey • New York
Director of Editorial Content: Kristin Klinger
Senior Managing Editor: Jennifer Neidig
Managing Editor: Jamie Snavely
Assistant Managing Editor: Carole Coulson
Typesetter: Sean Woznicki
Cover Design: Lisa Tosheff
Printed at: Yurchak Printing Inc.

Published in the United States of America by

Information Science Reference (an imprint of IGI Global)
701 E. Chocolate Avenue, Suite 200
Hershey PA 17033
Tel: 717-533-8845
Fax: 717-533-8661
E-mail: cust@igi-global.com
Web site: http://www.igi-global.com

and in the United Kingdom by

Information Science Reference (an imprint of IGI Global)
3 Henrietta Street
Covent Garden
London WC2E 8LU
Tel: 44 20 7240 0856
Fax: 44 20 7379 0609
Web site: http://www.eurospanbookstore.com

Copyright © 2009 by IGI Global. All rights reserved. No part of this publication may be reproduced, stored or distributed in any form or by
any means, electronic or mechanical, including photocopying, without written permission from the publisher.
Product or company names used in this set are for identification purposes only. Inclusion of the names of the products or companies does
not indicate a claim of ownership by IGI Global of the trademark or registered trademark.

Library of Congress Cataloging-in-Publication Data

The Semantic Web for knowledge and data management : technologies and practices / Zongmin Ma and Huaiqing Wang, editors.

p. cm.

Summary: “This book discusses the Semantic Web, as an extension of the current World Wide Web in which information is given well-
defined, better enabling computers and people to work in cooperation”--Provided by publisher.
Includes bibliographical references and index.

ISBN 978-1-60566-028-8 (hbk.) -- ISBN 978-1-60566-029-5 (ebook)

1. Semantic Web. I. Ma, Zongmin, 1965- II. Wang, Huaiqing.
TK5105.88815.S434 2009
025.04--dc22
2008010316

British Cataloguing in Publication Data

A Cataloguing in Publication record for this book is available from the British Library.

All work contributed to this book set is original material. The views expressed in this book are those of the authors, but not necessarily of
the publisher.

If a library purchased a print copy of this publication, please go to http://www.igi-global.com/agreement for information on activating
the library's complimentary electronic access to this publication.
Editorial Advisory Board

Alfredo Cuzzocrea Federica Mandreoli

University of Calabria, Italy University of Modena and Reggio Emilia, Italy

Danica Damljanović Wilma Penzo

The University of Sheffield, United Kingdom University of Bologna, Italy

Stathes Hadjiefthymiades Livia Predoiu

National and Kapodistrian University of Athens, University of Mannheim, Germany
Greece
Abad Shah
Lobna Karoui University of Engineering & Technology, Pakistan
Ecole Supérieure d’Electricité, France
Cláudio de Souza Baptista
Kostas Kolomvatsos Universidade Federal de Campina Grande
National and Kapodistrian University of Athens,
Greece Heiner Stuckenschmidt
University of Mannheim, Germany
Bangyong Liang
NEC Laboratories, China Jie Tang
Tsinghua University, China
Ning Li
University of Surrey, UK Edgar R. Weippl
Vienna University of Technology, Austria
Fábio Luiz Leite Júnior
University of Campina Grande, Brazil Anna V. Zhdanova
ftw. Forschungszentrum Telekommunikation Wien
Zongmin Ma Betriebs GmbH, Austria
Northeastern University, China
Table of Contents

Preface . ................................................................................................................................................. xi

Acknowledgment . .............................................................................................................................. xvi

Section I

Chapter I
Contextual Hierarchy Driven Ontology Learning .................................................................................. 1
Lobna Karoui, Ecole Supérieure d’Electricité, France

Chapter II
A Review of Fuzzy Models for the Semantic Web ............................................................................... 23
Hailong Wang, Northeastern University, China
Zongmin Ma, Northeastern University, China
Li Yan, Northeastern University, China
Jingwei Cheng, Northeastern University, China

Chapter III
Improving Storage Concepts for Semantic Models and Ontologies . ................................................... 38
Edgar R. Weippl, Vienna University of Technology, Austria
Markus D. Klemen, Vienna University of Technology, Austria
Stefan Raffeiner, Vienna University of Technology, Austria

Chapter IV
Ontologies and Intelligent Agents: A Powerful Bond . ......................................................................... 49
Kostas Kolomvatsos, National and Kapodistrian University of Athens, Greece
Stathes Hadjiefthymiades, National and Kapodistrian University of Athens, Greece

Chapter V
Probabilistic Models for the Semantic Web: A Survey . ....................................................................... 74
Livia Predoiu, University of Mannheim, Germany
Heiner Stuckenschmidt, University of Mannheim, Germany
Chapter VI
Automatic Semantic Annotation Using Machine Learning ............................................................... 106
Jie Tang, Tsinghua University, Beijing, China
Duo Zhang, University of Illinois, Urbana-Champaign, USA
Limin Yao, Tsinghua University, Beijing, China
Yi Li, Tsinghua University, Beijing, China

Section II

Chapter VII
Paving the Way to an Effective and Efficient Retrieval of Data over
Semantic Overlay Networks .............................................................................................................. 151
Federica Mandreoli, University of Modena and Reggio Emilia, Italy
Riccardo Martoglia, University of Modena and Reggio Emilia, Italy
Wilma Penzo, University of Bologna, Italy
Simona Sassatelli, University of Modena and Reggio Emilia, Italy
Giorgio Villani, University of Modena and Reggio Emilia, Italy

Chapter VIII
SWARMS: A Platform for Domain Knowledge Management and Applications .............................. 176
Jie Tang, Tsinghua University, Beijing, China
Bangyong Liang, NEC Labs, China
Juanzi Li, Tsinghua University, Beijing, China

Chapter IX
Modeling and Querying XML-Based P2P Information Systems:
A Semantics-Based Approach ............................................................................................................ 207
Alfredo Cuzzocrea, University of Calabria, Italy

Section III

Chapter X
Applying Semantic Web to E-Tourism .............................................................................................. 243
Danica Damljanović, University of Sheffield, UK
Vladan Devedžić, University of Belgrade, Serbia

Chapter XI
Semantic Web in Ubiquitous Mobile Communications ..................................................................... 266
Anna V. Zhdanova, The Telecommunications Research Center Vienna, Austria
Ning Li, University of Surrey, UK
Klaus Moessner, University of Surrey, UK
Chapter XII
Design Diagrams as Ontological Sources: Ontology Extraction and Utilization
for Software Asset Reuse ................................................................................................................... 288
Kalapriya Kannan, IBM India Research Labs, India
Biplav Srivastava, IBM India Research Labs, India

Compilation of References .............................................................................................................. 316

About the Contributors ................................................................................................................... 359

Index ................................................................................................................................................ 365

Detailed Table of Contents

Preface . ................................................................................................................................................. xi

Acknowledgment . .............................................................................................................................. xvi

Section I

Chapter I
Contextual Hierarchy Driven Ontology Learning .................................................................................. 1
Lobna Karoui, Ecole Supérieure d’Electricité, France

This chapter defines a hierarchical clustering algorithm namely Contextual Concept Discovery (CCD),
based on an incremental use of the partitioning algorithm Kmeans, guided by a structural contextual
definition and providing a web driven evaluation task to support the domain experts. The CCD algo-
rithm is implemented and tested to incrementally extract the ontological concepts. It is shown that the
context-based hierarchical algorithm gives better results than a usual context definition and an existing
clustering method.

Chapter II
A Review of Fuzzy Models for the Semantic Web ............................................................................... 23
Hailong Wang, Northeastern University, China
Zongmin Ma, Northeastern University, China
Li Yan, Northeastern University, China
Jingwei Cheng, Northeastern University, China

Based on fuzzy set theory, this chapter reviews the existing proposals for extending the theoretical
counterpart of the Semantic Web languages, description logics (DLs), and the languages themselves.
The expressive power of the fuzzy DLs formalism and its syntax and semantic, knowledge base, the
decidability of the tableaux algorithm and its computational complexity as well as the fuzzy extension
to OWL are discussed.
Chapter III
Improving Storage Concepts for Semantic Models and Ontologies .................................................... 38
Edgar R. Weippl, Vienna University of Technology, Austria
Markus D. Klemen, Vienna University of Technology, Austria
Stefan Raffeiner, Vienna University of Technology, Austria

This chapter presents an improved database schema to store ontologies. More specifically, it proposes
an intuitive and efficient way of storing arbitrary relationships, shows that their database schema is well
suited to store both RDF and Topic Maps, and explains why it is more efficient by comparing it to other
approaches.

Chapter IV
Ontologies and Intelligent Agents: A Powerful Bond .......................................................................... 49
Kostas Kolomvatsos, National and Kapodistrian University of Athens, Greece
Stathes Hadjiefthymiades, National and Kapodistrian University of Athens, Greece

This chapter provides an exhaustive description of fundamentals regarding the combination of the
Semantic Web and intelligent agent technologies. It is shown that agents build ontologies and agents
integrate ontologies in an automatic way.

Chapter V
Probabilistic Models for the Semantic Web: A Survey ........................................................................ 74
Livia Predoiu, University of Mannheim, Germany
Heiner Stuckenschmidt, University of Mannheim, Germany

Focusing on probabilistic approaches for representing uncertain information on the Semantic Web, this
chapter surveys existing proposals for extending semantic web languages or formalisms underlying
Semantic Web languages in terms of their expressive power, reasoning capabilities as well as their suit-
ability for supporting typical tasks associated with the Semantic Web.

Chapter VI
Automatic Semantic Annotation Using Machine Learning ............................................................... 106
Jie Tang, Tsinghua University, Beijing, China
Duo Zhang, University of Illinois, Urbana-Champaign, USA
Limin Yao, Tsinghua University, Beijing, China
Yi Li, Tsinghua University, Beijing, China

This chapter studies the problems of semantic annotation and introduces the state-of-the-art methods for
dealing with the problems. It also gives a brief survey of the developed systems based on the methods.
Several real-world applications of semantic annotation are introduced as well.
 Section II

Chapter VII
Paving the Way to an Effective and Efficient Retrieval of Data over
Semantic Overlay Networks .............................................................................................................. 151
Federica Mandreoli, University of Modena and Reggio Emilia, Italy
Riccardo Martoglia, University of Modena and Reggio Emilia, Italy
Wilma Penzo, University of Bologna, Italy
Simona Sassatelli, University of Modena and Reggio Emilia, Italy
Giorgio Villani, University of Modena and Reggio Emilia, Italy

This chapter presents a semantic infrastructure for routing queries effectively in a network of SONs
(Semantic Overlay Networks). The chapter defines a fully distributed indexing mechanism which sum-
marizes the semantics underlying whole subnetworks, in order to be able to locate the semantically best
directions to forward a query to. It is demonstrated through a rich set of experiments that the proposed
routing mechanism overtakes algorithms which are usually limited to the only knowledge of the peers
directly connected to the querying peer.

Chapter VIII
SWARMS: A Platform for Domain Knowledge Management and Applications .............................. 176
Jie Tang, Tsinghua University, Beijing, China
Bangyong Liang, NEC Labs, China
Juanzi Li, Tsinghua University, Beijing, China

This chapter describes the architecture and the main features of SWARMS, a platform for domain knowl-
edge management. The platform aims at providing services for (1) efficiently storing and accessing the
ontological information; (2) visualizing the networking structure in the ontological data; (3) searching
and mining the semantic data.

Chapter IX
Modeling and Querying XML-Based P2P Information Systems:
A Semantics-Based Approach ............................................................................................................ 207
Alfredo Cuzzocrea, University of Calabria, Italy

This chapter first focuses its attention on the definition and the formalization of the XML-based P2P
Information Systems class, also deriving interesting properties on such systems. Then the chapter presents
a knowledge-representation-and-management-based framework that can efficiently process knowledge
and support advanced IR techniques in XML-based P2P Information Systems.
 Section III

Chapter X
Applying Semantic Web to E-Tourism .............................................................................................. 243
Danica Damljanović, University of Sheffield, UK
Vladan Devedžić, University of Belgrade, Serbia

This chapter presents Travel Guides - a prototype system for tourism management to illustrate how se-
mantic web technologies combined with traditional E-Tourism applications help integration of tourism
sources dispersed on the Web, and enable creating sophisticated user profiles.

Chapter XI
Semantic Web in Ubiquitous Mobile Communications ..................................................................... 266
Anna V. Zhdanova, The Telecommunications Research Center Vienna, Austria
Ning Li, University of Surrey, UK
Klaus Moessner, University of Surrey, UK

This chapter provides background information on the Semantic Web field, discusses other research fields
that bring semantics into play for reaching the ontology-enabled ubiquitous mobile communication vision,
and exemplifies the state of the art of ontology development and use in telecommunication projects.

Chapter XII
Design Diagrams as Ontological Sources: Ontology Extraction and Utilization
for Software Asset Reuse ................................................................................................................... 288
Kalapriya Kannan, IBM India Research Labs, India
Biplav Srivastava, IBM India Research Labs, India

This chapter discusses how semantic web technologies can help solution-building organizations achieve
software reuse by first learning ontologies from design diagrams of existing solutions and then using
them to create design diagrams for new solutions. The proposed technique, called OntExtract, extracts
domain ontology information (entities and their relationship(s)) from class diagrams and further refines
the extracted information using diagrams that express dynamic interactions among entities such as
sequence diagram.

Compilation of References .............................................................................................................. 316

About the Contributors ................................................................................................................... 359

Index ................................................................................................................................................ 365

 xi

Preface

The World Wide Web (WWW) has drastically changed the availability of electronically accessible
information. Currently the WWW is the biggest information repository around the world and the most
convenient means for information sharing and exchange. While the current Web provides access to an
enormous amount of information, it is currently only human-readable. It is increasingly difficult to find,
access, present and maintain the information required by a wide variety of users. In response to this problem,
the Semantic Web was proposed by Time Berners-Lee, Director of the World Wide Web Consortium. The
Semantic Web is defined as “an extension of the current World Wide Web in which information is given
well-defined, better enabling computers and people to work in cooperation”. The Semantic Web allows
the explicit representation of the semantics of the data so that it is machine interpretable. Therefore, the
Semantic Web will enable a knowledge-based web and facilities the use of agent-based technology to
better mine and filter information needs expressed by information consumers.
The Semantic Web is generally built on syntaxes which use URIs to represent data, usually in triple
based structures: that is, many triples of URI data that can be held in databases, or interchanged in the
World Wide Web using a set of particular syntaxes developed especially for the task. These syntaxes
are called Resource Description Framework (RDF) syntaxes. The layer above the syntax is the simple
datatyping model. The RDF Schema (RDFS) is designed to be a simple datatyping model for the RDF.
The Web Ontology Language (OWL) is a language as an ontology language based upon the RDF. OWL
takes the RDF Schema a step further, by giving us more in-depth properties and classes. The next step
in the architecture of the Semantic Web is trust and proof.
Being the next generation Internet technology, the Semantic Web is typically application-oriented.
With advances and in-deep applications of computer and Internet technologies in data and knowledge
intensive domains, the Semantic Web for knowledge and data management is emerging as a new dis-
cipline. The research and development of knowledge and data management in the Semantic Web are
receiving increasing attention. The requirements of large-scale deployment and interoperability of the
Semantic Web represent a major challenge to data and knowledge management, which raises a number
of issues and requirements regarding how to represent, create, manage and use both ontologies as shared
knowledge representations, but also large volumes of metadata records used to annotate Web resources
of a diverse kind. So the Semantic Web for knowledge and data management is a field which must be
investigated by academic researchers together with developers and users both from database, artificial
intelligence, and software and knowledge engineering areas.
This book focuses on the following issues of the Semantic Web: the theory aspect of the Semantic
Web and ontology, data management and processing in the Semantic Web, ontology and knowledge
management, and Semantic Web-based applications, aiming at providing a single account of technologies
and practices in the Semantic Web for knowledge and data management. The objective of the book is
to provide the state of the art information to academics, researchers and industry practitioners who are
xii

involved or interested in the study, use, design and development of advanced and emerging the Semantic
Web technologies with ultimate aim to empower individuals and organizations in building competencies
for exploiting the opportunities of the data and knowledge society. This book presents the latest research
and application results in the Semantic Web. The different chapters in the book have been contributed
by different authors and provide possible solutions for the different types of technological problems
concerning the Semantic Web for knowledge and data management.

IntroductIon

This book which consists of twelve chapters is organized into three major sections. The first section
discusses the issues of the Semantic Web and ontologies in the first six chapters. The next three chapters
covering the Semantic Web for data and knowledge management comprise the second section. The third
section containing the final three chapters focuses on the applications of the Semantic Web.
First of all, we take a look at the issues of the Semantic Web and ontologies.
Research in ontology learning had always separated between ontology building and evaluation tasks.
Moreover, it had used for example a sentence, a syntactic structure or a set of words to establish the
context of a word. However, this research avoids accounting for the structure of the document and the
relation between the contexts. Lobna Karoui combines these elements to generate an appropriate context
definition for each word. Based on the context, she proposes an unsupervised hierarchical clustering
algorithm that, in the same time, extracts and evaluates the ontological concepts. The results show that her
concept discovery approach improves the conceptual quality and the relevance of the extracted ontologi-
cal concepts, provides a support for the domain experts and facilitates the evaluation task for them.
In the Semantic Web context, information would be retrieved, processed, shared, reused and aligned
in the maximum automatic way possible. The experience with such applications in the Semantic Web
has shown that these are rarely a matter of true or false but rather procedures that require degrees of
relatedness, similarity, or ranking. Apart from the wealth of applications that are inherently imprecise,
information itself is many times imprecise or vague. In order to be able to represent and reason with
such type of information in the Semantic Web, different general approaches for extending semantic web
languages with the ability to represent imprecision and uncertainty has been explored. Hailong Wang et
al. focus their attention on fuzzy extension approaches which are based on fuzzy set theory. They review
the existing proposals for extending the theoretical counterpart of the semantic web languages, description
logics (DLs), and the languages themselves. The expressive power of the fuzzy DLs formalism and its
syntax and semantic, knowledge base, the decidability of the tableaux algorithm and its computational
complexity as well as the fuzzy extension to OWL are discussed.
Ontologies are more commonly used today but still little consideration is given of how to efficiently
store them. Edgar R. Weippl, Markus D. Klemen and Stefan Raffeiner present an improved database
schema to store ontologies. More specifically, they propose an intuitive and efficient way of storing
arbitrary relationships, show that their database schema is well suited to store both RDF and Topic
Maps, and explain why it is more efficient by comparing it to other approaches. The proposed approach
is built on reliable and efficient relational database management systems (RDBMS). It can be easily
implemented for other systems and due to its vendor independence existing data can be migrated from
one RDBMS to another relatively easy.
The emerged form of information with computer-processable meaning (semantics) as presented in the
framework of the Semantic Web (SW) facilitates machines to access it more efficiently. Information is
semantically annotated in order to ease the discovery and retrieval of knowledge. Ontologies are the basic
 xiii

element of the SW. They carry knowledge about a domain and enable interoperability between different
resources. Another technology that draws considerable attention nowadays is the technology of Intel-
ligent Agents. Intelligent agents act on behalf of a user to complete tasks and may adapt their behavior
to achieve their objectives. Kostas Kolomvatsos and Stathes Hadjiefthymiades provide an exhaustive
description of fundamentals regarding the combination of SW and intelligent agent technologies.
Recently, there has been an increasing interest in formalisms for representing uncertain information
on the Semantic Web. This interest is triggered by the observation that knowledge on the web is not
always crisp and we have to be able to deal with incomplete, inconsistent and vague information. The
treatment of this kind of information requires new approaches for knowledge representation and reason-
ing on the web as existing Semantic Web languages are based on classical logic which is known to be
inadequate for representing uncertainty in many cases. While different general approaches for extending
Semantic Web languages with the ability to represent uncertainty are explored, Livia Predoiu and Heiner
Stuckenschmidt focus their attention on probabilistic approaches. They survey existing proposals for
extending semantic web languages or formalisms underlying Semantic Web languages in terms of their
expressive power, reasoning capabilities as well as their suitability for supporting typical tasks associ-
ated with the Semantic Web.
The Semantic Web provides a common framework that allows data to be shared and reused across
applications, enterprises, and community boundaries. However, lack of annotated semantic data is a
bottleneck to make the Semantic Web vision a reality. Therefore, it is indeed necessary to automate the
process of semantic annotation. In the past few years, there was a rapid expansion of activities in the
semantic annotation area. Many methods have been proposed for automating the annotation process.
However, due to the heterogeneity and the lack of structure of the Web data, automated discovery of
the targeted or unexpected knowledge information still present many challenging research problems.
Jie Tang et al. study the problems of semantic annotation and introduce the state-of-the-art methods for
dealing with the problems. They also give a brief survey of the developed systems based on the methods.
Several real-world applications of semantic annotation are introduced as well.
The next section takes look at the data and knowledge management with the Semantic Web and
ontologies.
In a Peer-to-Peer (P2P) system, a Semantic Overlay Network (SON) models a network of peers whose
connections are influenced by the peers’ content, so that semantically related peers connect with each
other. This is very common in P2P communities, where peers share common interests, and a peer can
belong to more than one SON, depending on its own interests. Querying such a kind of systems is not
an easy task: The retrieval of relevant data can not rely on flooding approaches which forward a query
to the overall network. A way of selecting which peers are more likely to provide relevant answers is
necessary to support more efficient and effective query processing strategies. Federica Mandreoli et
al. present a semantic infrastructure for routing queries effectively in a network of SONs. Peers are
semantically rich, in that peers' content is modelled with a schema on their local data, and peers are
related each other through semantic mappings defined between their own schemas. A query is routed
through the network by means of a sequence of reformulations, according to the semantic mappings
encountered in the routing path. As reformulations may lead to semantic approximations, they define a
fully distributed indexing mechanism which summarizes the semantics underlying whole subnetworks,
in order to be able to locate the semantically best directions to forward a query to. They demonstrate
through a rich set of experiments that their routing mechanism overtakes algorithms which are usually
limited to the only knowledge of the peers directly connected to the querying peer, and that their ap-
proach is successful in a SONs scenario.
Jie Tang et al. describe the architecture and the main features of SWARMS, a platform for domain
knowledge management. The platform aims at providing services for (1) efficiently storing and accessing
xiv

the ontological information; (2) visualizing the networking structure in the ontological data; (3) searching
and mining the semantic data. One advantage of the system is that it provides a suite of components for
not only supporting efficient semantic data storage but also searching and mining the semantics. Another
advantage is that the system supports visualization in the process of search and mining, which would
greatly help a normal user to understand the knowledge inside the ontological data. SWARMS can be
easily customized to adapt to different domains. The system has been applied to several domains, such as
News, Software, and Social Network. The authors present the performance evaluations of the system.
Knowledge representation and management techniques can be efficiently used to improve data
modeling and IR functionalities of P2P Information Systems, which have recently attracted a lot of at-
tention from both industrial and academic research communities. These functionalities can be achieved
by pushing semantics in both data and queries, and exploiting the derived expressiveness to improve
file sharing primitives and lookup mechanisms made available by first-generation P2P systems. XML-
based P2P Information Systems are a more specific instance of this class of systems, where the overall
data domain is composed by very large, Internet-like distributed XML repositories from which users
extract useful knowledge by means of IR methods implemented on top of XML join queries against the
repositories. Alfredo Cuzzocrea first focuses his attention on the definition and the formalization of the
XML-based P2P Information Systems class, also deriving interesting properties on such systems, and
then he presents a knowledge-representation-and-management-based framework, enriched via semantics,
that allows us to efficiently process knowledge and support advanced IR techniques in XML-based P2P
Information Systems, thus achieving the definition of the so-called Semantically-Augmented XML-
based P2P Information Systems.
In the third section, we see the application aspects of the Semantic Web.
Traditional E-Tourism applications store data internally in a form that is not interoperable with simi-
lar systems. Hence, tourist agents spend plenty of time updating data about vacation packages in order
to provide good service to their clients. On the other hand, their clients spend plenty of time searching
for the ‘perfect’ vacation package as the data about tourist offers are not integrated and are available
from different spots on the Web. Danica Damljanović and Vladan Devedžić develop Travel Guides - a
prototype system for tourism management to illustrate how semantic web technologies combined with
traditional E-Tourism applications help integration of tourism sources dispersed on the Web, and enable
creating sophisticated user profiles. Maintaining quality user profiles enables system personalization and
adaptivity of the content shown to the user. The core of this system is in ontologies – they enable machine
readable and machine understandable representation of the data and more importantly reasoning.
The world becomes ubiquitous, and mobile communication platforms become oriented towards
integration with the web, getting benefits from the large amount of information available there, and
creation of the new types of value-added services. Semantic and ontology technologies are seen as being
able to advance the seamless integration of the mobile and the Web worlds. Anna V. Zhdanova, Ning Li
and Klaus Moessner present the overall state of the art ontology-related developments in mobile com-
munication systems, namely, the work towards construction, sharing and maintenance of ontologies for
mobile communications, reuse and application of ontologies and existing Semantic Web technologies
in the prototypes. Social, collaborative and technical challenges experienced in the project showcase
the need in alignment of ontology experts’ work across the mobile communication projects to estab-
lish the best practices in the area and drive standardization efforts. They indicate certain milestones
in integration of Semantic Web-based intelligence with Mobile Communications, such as performing
ontology construction, matching, and evolution in mobile service systems and alignment with existing
heterogeneous data models.
 xv

Ontology is a basic building block for the Semantic Web. An active line of research in semantic
web is focused on how to build and evolve ontologies using the information from different ontological
sources inherent in the domain. A large part of the IT industry uses software engineering methodolo-
gies to build software solutions that solve real-world problems. For them, instead of creating solutions
from scratch, reusing previously built software as much as possible is a business-imperative today. As
part of their projects, they use design diagrams to capture various facets of the software development
process. Kalapriya Kannan and Biplav Srivastava discuss how semantic web technologies can help so-
lution-building organizations achieve software reuse by first learning ontologies from design diagrams
of existing solutions and then using them to create design diagrams for new solutions. Their technique,
called OntExtract, extracts domain ontology information (entities and their relationship(s)) from class
diagrams and further refines the extracted information using diagrams that express dynamic interactions
among entities such as sequence diagram. A proof of concept implementations is also developed as a
Plug-in over a commercial development environment IBM’s Rational Software Architect.

Zongmin Ma and Huaiqing Wang

Editors
xvi

Acknowledgment

The editors wish to thank all of the authors for their insights and excellent contributions to this book
and would like to acknowledge the help of all involved in the collation and review process of the book,
without whose support the project could not have been satisfactorily completed. Most of the authors of
chapters included in this book also served as referees for chapters written by other authors. Thanks go
to all those who provided constructive and comprehensive reviews.
A further special note of thanks goes to all the staff at IGI Global, whose contributions throughout
the whole process from inception of the initial idea to final publication have been invaluable. Special
thanks also go to the publishing team at IGI Global. This book would not have been possible without
the ongoing professional support from IGI Global.
The idea of editing this volume stems from the initial research work that the editors did in past several
years. The assistances and facilities of Northeastern University and City University of Hong Kong, China,
are deemed important, and are highly appreciated. The research work of Zongmin Ma supported by the
Program for New Century Excellent Talents in University (NCET-05-0288).

Zongmin Ma and Huaiqing Wang

Editors
January, 2008
Section I
 

Chapter I
Contextual Hierarchy Driven
Ontology Learning
Lobna Karoui
Ecole Supérieure d’Electricité, France

AbstrAct

Research in ontology learning had always separated between ontology building and evaluation tasks.
Moreover, it had used for example a sentence, a syntactic structure or a set of words to establish the
context of a word. However, this research avoids accounting for the structure of the document and the
relation between the contexts. In our work, we combine these elements to generate an appropriate con-
text definition for each word. Based on the context, we propose an unsupervised hierarchical clustering
algorithm that, in the same time, extracts and evaluates the ontological concepts. Our results show that
our concept discovery approach improves the conceptual quality and the relevance of the extracted
ontological concepts, provides a support for the domain experts and facilitates the evaluation task for
them.

IntroductIon use of context in the ontology learning process.

Until now, most works have investigated various
Some current work in data annotation, data issues of ontology building such as methodology
integration, information retrieval, building of ontology extraction and ontology evaluation
multi-agents application, semantic web services but separately. In our research, we are interested
depends on ontologies. The development and the in defining an approach that permits, at the same
deployment of these applications are related to the time, to extract and evaluate the ontological con-
richness of the conceptualization inside the ontol- cepts by using the notion of “context”. Generally,
ogy. Moreover, giving a semantic to the web can the idea of defining the context of a word is lim-
be realized in an incremental manner by using the ited to a sentence, a syntactic structure, a set of
ontology vocabulary. In this paper, we focus on the words or a set of sentences without tacking into

Copyright © 2009, IGI Global, distributing in print or electronic forms without written permission of IGI Global is prohibited.
 Contextual Hierarchy Driven Ontology Learning

account the document’s structure and the relation permits the ontology reuse and evolution since
between the contexts. In this work, we combine the informing elements that support the experts’
these elements to establish an appropriate context interpretation are driven by the web changes
definition for each word. So, our contributions are and are stored with the experts’ comments for a
to propose firstly a new context definition that later use. We experiment the contextual cluster-
takes into account the document characteristics ing algorithm on French html document corpus
and secondly to define a new approach that not related to the tourism domain. The results show
only extracts the ontological concepts but also that our algorithm improves the relevance of the
evaluates them. This approach is based on our extracted concepts in comparison with a simple
proposed context definition that takes into ac- Kmeans. Also, our observations and discussions
count the position of a word inside the HTML with experts confirm that our evaluation process
structure and the relation between the deduced helps and assists the user.
contexts in the HTML document. For this pur- The remainder of the paper is organized as
pose, we propose an unsupervised hierarchical follows: section 2 presents the state of the art
clustering algorithm namely Contextual Concept related, sections 3 answers the question: What is
Discovery (CCD) based on an incremental use of the impact of the context on the intelligent inter-
the partitioning Kmeans algorithm, guided by a pretation, section 4 presents defines the context
structural context and producing a support for and the Contextual Concept Discovery” (CCD)
an easy evaluation task. Our context definition algorithm, section 5 experiments our algorithm
is based on the html structure and the location proposition and section 6 concludes and gives our
of each word in the documents. Each context is future directions.
deduced from the various analyses included in
the pre-processing step (Karoui et al., 2006). This
explicitly contextual representation, titled contex- relAted Work
tual hierarchy, guides the clustering algorithm to
delimit the context of each word by improving the Ontology is “a specification of a conceptualiza-
word weighting, the words pair’s similarity and tion” (Gruber, 1993). Ontology learning systems
the semantically closer cooccurent selection for as defined in (Faure, 1998; Chalendar, 2000;
each word. By performing an incremental process Meadche, 2001) have different purposes. Many
and by recursively dividing each cluster, the CCD of them extract concepts and relationships from
algorithm refines the context of each word cluster. a collection of documents related to a specific
It improves the conceptual quality of the extracted domain in order to either construct ontology or
concepts. The CCD algorithm offers the choice enhance it. For this they use a conceptual cluster-
between either an automatic execution or a user ing (Faure and Nedellec, 1998), linguistic approach
interactive one. In order to help the domain expert (Navigli, 1998; Chalendar and Grau, 2000), statis-
during the evaluation task, the last part of the CCD tic one (Maedche et al, 2000; Srinivassan, 1992).
algorithm exploits a web collection and extracts In this related work, we present an overview of
the existing contexts. This information is used the context in some domains that are related to
to compute the credibility degree associated to our research (since the notion of context is used
each word cluster in order to inform about it and in many other domains that are not close to our
facilitate the experts’ semantic interpretation. So, work such as physic, mathematical models, etc.).
the CCD algorithm extracts the domain concepts We focus on the use of context in the ontology
and proposes a quantitative and a qualitative evalu- learning. Then, we expose some ideas about the
ation for the experts. Our evaluation proposition ontology evaluation.


Contextual Hierarchy Driven Ontology Learning

context in some domains context than the training examples from which
the concepts are deduced. These features are used
In this research, our aim is to discover the concepts for a normalization task and defined manually.
that belong to a domain ontology by using the However, in Widmer (1996), the author presents
notion of “context”. The concepts that constitute a method to automatically detect the contextual
an ontology are named “ontological concepts”. attributes in an on-line learning approach. These
In order to improve the conceptual quality of determined features are used during the learning
those concepts, we use the context of each word. setting. Other researches work on the speech rec-
According to the Oxford English Dictionary ognition task by using the speaker’s accent, the
(http://www.askoxford.com/), the term context sound and the information, the electrocardiogram
has two meanings: (1) the words around the word, data and the patient’s identity to realize a heart
phrase, statement and so one used to help explain disease diagnostics, etc.
the meaning and (2) the general conditions in
which an event, action, and so on takes place. Context in Ontology Learning
In the following sections, we explain how this
notion has been used in some domains related Faure and Nédellec (1998) describe a system called
to our research. ASIUM where a cooperative conceptual cluster-
ing is applied to technical texts using syntactic
Context in Word Sense Disambiguation parser to produce an acyclic conceptual graph of
clusters. Basic clusters are formed by words that
In this field, context is used either as a micro occur with the same verb after the same preposi-
context or a topical context. As a micro context, tion. In Chalendar and Grau (2000), the authors
Yarowsky (1993) experiments different small design a system titled SVETLAN. This system
window (the larger one is 4 words). He affirms learns noun categories from texts by using Sylex
that these micro contexts are sufficient for local tool. Consequently, nouns having the same role
ambiguities. Similarly, Leacock and al (1996) with the same verb inside the same syntactic
shows that the best performance in their tests is context are homogenous and belong to the same
due to a local window of +-3 words. As topical class. Input data are semantic domain and its
context, Yarowsky (1992) has used a 100-word thematic entities learned automatically by using
window in order to extract the related words. SEGAPSITH. The difference between ASIUM
Voorhees and al (1993) define the context as and SVETLAN is that this later is independent
two sentence window. Others use either +- a of the expert intervention and the text nature.
big number of words inside the window such as TEXT-TO-ONTO (Meadche and Stabb, 2001)
Gale and al (1992) (+- 50 words) or sections of is an ontology learning environment which has
the documents (Brown et al., 1991). a bookshop of learning methods such as formal
concept analysis, association rules which permit
Context in Machine Learning to extract concepts, taxonomic relations and non
taxonomic relations, linguistic tool (analyzer)
Turney (1996) was one of the first to explicitly ac- and heuristics. WebOntEx (Han, 2000) system’s
knowledge the problem of context in learning. He goal is to extract semi automatically ontology by
gives a formal definition of contextual and context analyzing Web pages of the same specific domain.
sensitive features. His approach was motivated by Knowledge extraction is based on HTML tags (B,
resolving the learning problems related to the fact h1), lemmatization tags (verb, name) and concep-
that testing examples were governed by a different tual tags (entity, attribute) using WordNet and a


 Contextual Hierarchy Driven Ontology Learning

logic inductive programming method. However, which surround the studied object and reflects its
this methodology requires a long manual task concrete environment. It provides an active back-
during which the designer manually develops a ground for a learning activity and an appropriate
kernel ontology and extracts generic patterns from support for a semantic interpretation. In this study,
domain Web pages. OntoMiner (Davulcu, 1998) the studied object is a word, the learning activity
analyzes sets of domain specific web sites and is a clustering task, the semantic interpretation is
generates taxonomy of particular concepts and the evaluation and labeling of the word clusters
their instances. This tool uses HTML regulari- and the context definition is deduced from the html
ties within web documents in order to generate a structure (HTML tags and the relations between
hierarchical semantic structure encoded in XML. them). In our work, the contribution is not only
It explores directories of home pages in order to to adapt a clustering algorithm to the ontology
detect key domain concepts and relations between learning domain (like many other researches) but
them. OntoLearn (Navigli, 1998) is a tool based also to define a new context definition that takes
on linguistic and semantic techniques. It extracts into account the document structure and the word
domain terminology from Web documents by position in the related contexts .More details are
using a linguistic processor and a syntactic explained in our algorithm (section 4).
parser. The semantic interpretation task consists
in finding the appropriate WordNet concept for ontology evaluation
each term.
Ontology evaluation remains a real problem in the
Discussion area of the semantic web. There is no standard
methodology or approach to evaluate ontology.
For TEXT-TO-ONTO, ASIUM, SVETLAN and This is due to the fact that the ontology learning
OntoLearn systems, the linguistic context of the depends on several aspects such as the purpose
term is defined using syntactic dependencies. of the ontology building, the application (s) using
This context definition limited to a linguistic the ontology, the entities constituting the ontol-
one is not sufficient because the word context ogy, the kind of ontology (domain ontology, task
exceeds the phrase i.e. the word can exist in the ontology, etc.), etc. In this survey, we present
paragraph with other meaning. With the WebOn- research related to the evaluation of concepts
tEx and OntoMiner systems, the authors consider which constitutes the ontology.
respectively the HTML tags (separately without A concept is “A general idea derived or in-
studying the relations between them) and the home ferred from specific instances or occurrences”
page directories as contexts. We also remark that (http://www.thefreedictionary.com). Discovering
the context can be an existing ontology, domain concepts implies having a vocabulary for which
expert, some visual information, some rules, some the idea of the concept is deduced. Therefore, the
words around a target object (window), one or a ontology engineer can evaluate each or both of
set of sentences or one syntactic structure (verb the vocabulary and extracted concepts. In order
group, nominal group). For all these definitions, the to evaluate the vocabulary, Meadche and Staab
document structure is rarely taken into account. (2002) proposed an approach that aims to evaluate
Also, the relations between these contexts (for one the lexical and vocabulary level of an ontology.
application) are not considered. In our work, we They have defined a similarity measure in order
deal with these issues to improve the ontological to compare two strings: one provided from the
concept discovery (extraction and evaluation). For produced ontology and the other one from an
us, the context is a set of circumstances (situations) existing ontology. In (Brewster et al., 2004), the


Contextual Hierarchy Driven Ontology Learning

authors evaluate their lexical by using WordNet context And the IntellIgent
and precision and recall measures. Based on this InterpretAtIon
vocabulary, ontology building approaches apply-
ing clustering methods to obtain word clusters The evaluation problem is defined as the process
as potential future concepts. So, the question is that clarifies the transition from what it is ‘written’
how to evaluate these clusters. In the clustering to what it is ‘thought’ i.e. eliminates the fuzzy and
process, the quality of a cluster is generally based unpredictable character of the interpretation. In
on homogeneity or compactness. In (Vazirgiannis our research, we deal with the issue of evaluating
et al., 2003), some criteria for the statistical evalu- the concept extracted from word clusters result-
ation of unsupervised learners have been defined. ing from a clustering method. So, the expert tries
However, the ontology learning applications to give a semantic label to a group of words by
cannot rely on these standards defined for other finding the relations between them. But this task
applications. Moreover, cluster homogeneity does remains difficult for the expert. That is why we
not imply that the words in the cluster are semanti- propose to give complementary information either
cally closer or that the associated label satisfies the for the single units (words), or for all the words
domain expert. For the concepts extraction, the of a cluster considered as a group or for both of
evaluation is more challenging. In (Holsapple et them. To provide this useful information, we
Joshi, 2005), the authors proposed an evaluation use the contexts of the words of each cluster. So,
method based on a collaborative manual ontology what is the impact of the context on the experts’
engineers in order to maintain the suggestions evaluation task?
resulting from the maximum number of experts. In this analysis, we explain how the context
Navigli and al (2004) proposed a qualitative evalu- affects the intelligent interpretation in at least
ation by multiple domain experts that answer to a five different ways.
questionnaire in which they evaluate the quality First, recognizing the context allows the user to
of the discovered concepts. In (Holsapple et Joshi, make predictions about the possible meanings of
2005; Navigli and al., 2004), the concept evalu- some elements. Contextual information provides
ation is based on a human intervention which is top-down predictions that permits recognition of
a painful task. In other research (Meadche and revealing elements based on partial knowledge
Staab, 2002; Brewster et al., 2004; Vazirgiannis et about them and it can serve as source of hypotheses
al., 2003), the evaluation and the labeling process about the meanings of words not yet known.
is based on statistic measures, a thesaurus, an Second, contexts modulate interpretation by
existing ontology or an application. Generally, we setting some parameters or suggestions to orient
remark that thesaurus doesn’t cover all specific the reasoning. It limits the words’ ambiguities
aspects of a domain. Also, the evaluation based in function of other related contexts which are
on an existing ontology is not sufficient because, more general or more specific. For example, when
in some fields, ontology does not exist or does not a person hears me saying (Brézillon, 1993): “I
contain all the concepts founded in the produced heard a lion roar in my office this morning’”, he
ontology. Moreover, it is not evident to find an can suppose that I am speaking about my boss,
application that uses the produced ontology. so it is necessary to introduce explicitly some
As a conclusion, we can say that the evalua- piece of knowledge such as adding the following
tion is a difficult task. That is why we deal with sentence: “I work in a university near a zoo that
this issue. In the following section, we analyze can be seen from the window of my office.” We
the impact of the context on the interpretation in note that within these two sentences, the listener
order to facilitate the experts’ evaluation. knows that I speak about a real lion and not my


 Contextual Hierarchy Driven Ontology Learning

boss. So the context (first sentence) allows a first not said explicitly. So, we can consider accepting
interpretation and the second one adjusts it in animals and having parking as property to the
order to achieve the right one. concept “hotel”.
Third, context helps user focus his attention. We are conscious now that the context has a
The user have often many interpretations, con- real impact on the semantic interpretation. These
sequently he needs some way for determining ideas are applied in our algorithm (section 4.2).
which of his interpretation is the most appropri-
ate one, which differs depending on the context.
For example, having the sentence “in the east the contextuAl concept
region of USA, The means of accommodation dIscovery AlgorIthm
are comfortable” allows the user to have different
interpretations about the word “accommodation” Our algorithm proposition is constituted of three
so it could be a medical accommodation for doc- parts which are: a structural context definition, a
tors or a social accommodation for helping poor clustering process to group the semantic words
people or a tourism accommodation for tourist. together and a web driven evaluation task based
It could also be just a hotel as accommodation or on various context types with a quantitative cri-
hotel and camping or other kind of accommoda- terion (credibility degree).
tion. So this context is insufficient to guide the
user’s interpretation. But by giving supplementary The Context Definition
information like “the east region of USA is a tour-
istic one and the means of accommodations are In this section, we focus on the selection of the
comfortable.”, the user knows that it is a tourism semantically closer cooccurrents and the word
accommodation so we can have as sub concepts weighting process. Generally, if two words ap-
hotels, camping and so one. pear together in a phrase or a document, we at-
Fourth, context influences the user’s choice of tribute them either the value 1 (i.e. present in this
the appropriate meaning of the word. For example, context) or another value between 0 and 1. In the
in French language the word “pêche” which could two cases, the context (phrase, document, etc.)
be translated in English as “peach” or “fishing” could be appropriate to some words but not all
has two meanings, so there is an ambiguity. But, the words of a corpus. For example, we consider
by having the word “sea” in the same context or in the three phrases: “USA gives a lot of lodging
related contexts, the user chooses the meaning of possibilities. In the north east region, it offers
fishing rather than peach because the word “sea” hotels and residences. Also, it gives various leisure
which is an element of the context had influenced activities.” If we fix the context to a sentence, we
the user’s choice. find that ‘lodging’ and ‘hotel’, which belong to the
Fifth, context determines how the user can same concept, have the value 0 because they don’t
discover implicit meanings. For example, let us belong to the same sentence. If we fix the context
take these two sentences: “Mrs Smith reserves a to a window whose size is 14 words, we find that
hotel in Florida and decides to go by car. Since, ‘residence’ and ‘leisure’ have a value while they do
he can not be separated from his dog; he decides not belong to the same concept (residence belong
to take it with him instead of leaving it to his to the concept ‘lodging’ and ‘leisure’ is another
neighbour.” Within this context, the existing in- concept). So, ‘leisure’ should not be a cooccurrent
formation allow the user to discover that the hotel to ‘residence’. For these reasons, we believe that
that Mr. smith chooses accepts animals and has the context should be refined in order to take more
a parking in spite of these two information are into account the location of the word.


Contextual Hierarchy Driven Ontology Learning

Actually our challenge is to answer correctly rence between these two words in a unit related
the question: how can we give a weighting that to the context. In our study, the concept of context
effectively illustrates the importance of a term is variable, so we do not fix the context during
in its domain and characterizes its relations with the process but we define a new concept which is
other words? a contextual hierarchy represented explicitly. By
Let us go back to our corpus and the various taking this structure into account, we can define
analyses performed on the HTML documents. We a link between terms, if they appear in the same
note that there exist relations between the exist- block-level tag (<p>, <td>, etc.). In this case, the
ing HTML elements. For instance: <h1>  <p> link is a neighbourhood co-occurrence (N.C)
(heading  paragraph). We also note that key tags and the context is limited to the tag. However, if
(defined tags like <keywords>, <glossary>, etc.) they appear in different tags that are related by a
and <title> are related to other existing HTML physical or a logical link defined in the contextual
elements. For example: <TITLE_URL> (header hierarchy, we define the concept of link co-oc-
of a hyperlink)  <H1> (headings of a part of currence (see Figure 1). In this second case, the
document). The first group of links is physical context is the association of the two related tags.
noted P.L because it depends on the structure The neighbourhood co-occurrence permits find-
of the HTML document. But the second group ing the word’s cooccurrents in a unique context
shows a logical link (L.L) that is not always visible while a link cooccurrence depends on the word
(elements are not necessarily consecutives). In position in a context and the relation between
order to represent links between tags, we define this context and the other existent ones. So, it is
two new concepts: contextual hierarchy (C.H) a generic context that will be instantiated accord-
based on HTML elements and link co-occurrence ing to the location of the term in HTML tags (for
(L.C). A contextual hierarchy is a tag hierarchy. example the context can take various values since
It illustrates possible relations existing within it could be in one case the tag <B> and in another
HTML documents and between them. case the association of <H1> and <TITLE>).
When we find two terms in the same context The structural context definition represents the
unit (paragraph, text), we speak about co-occur- term’s adaptability in its corpus. The associated

Figure 1. Examples of contexts use and their deduced cooccurrents


 Contextual Hierarchy Driven Ontology Learning

contextual model respects the word location in for the evaluation and labeling task. In order to
order to take a multitude of term situations into explain this idea, we take the sentence: the pos-
account when computing its weight. The term sible accommodations in the east region of USA
weighting is computed by applying the Equiva- are hotels and residences. Within this example,
lence Index defined by Michelet (1988) (Ci: term when we limit the context to the association of
occurrence; Cij: terms cooccurrences): ‘hotels’ and ‘residences’ by the conjunction ‘and’,
we deduce that ‘hotels’ and ‘residences’ belongs
Eij = C2ij / (Ci x Cj) (1) to the same concept. However, when we limit the
context to the entire sentence, we can say that the
the Fundamental directives of the associated concept to these two words is ‘accom-
evaluation task modation’. So, thanks to the contextualization
task, we can deduce either the meaning of each
Our idea is as follows: looking through the Web word, or the semantic association between some
in order to understand the meaning of each word words or the concept associated to some words.
or two words together and so on could be a solu- Taking into account a static context i.e. only one
tion but why? such as a sentence for all the word clusters is not
This task is a contextualization operation. sufficient since in some case the sentence does
Brézillon (1999) says: Context is what gives not contain all the words of a cluster. That’s why,
meaning to data and contextualization is the our evaluation is not restricted to a unique context
process of interpreting data, transforming data on the contrary it depends on various granularity
to information. levels which are applied and considered con-
In order to extract concepts, terms are selected secutively. The several contexts defined from
from their context in order to group them, but the domain web documents are provided by two
they are presented to the knowledge engineer or sources. The first one is a linguistic analysis that
the expert domain without any context after a permits to give us the various nominal groups
decontextualization process that’s why the evalu- and verbal groups. It also procures the various
ation step is always difficult. McCarthy (1993) word associations by a preposition (of, on, etc.)
says that decontextualization is to abstract a piece or a co-coordinating conjunction (and, or, etc.).
of knowledge from contexts into a more general The second source is a documentary analysis that
context that cover the initial context. In our case, permits to give us the various sections of phrases
for each cluster the general context is the domain (part of a phrase finished by a punctuation like ‘;’
(tourism). But this information is not sufficient to or ‘,’), the sentences, the paragraphs and the docu-
evaluate a cluster and to give it a semantic tag. ments. So, we have two types of contexts which
A solution for ensuring an easy analysis re- are a linguistic context and a documentary context.
quires a big collection of web documents related By using the first one, we obtain the close words
to the same studied domain and extracting several of our target terms. By using the second one, the
contexts from them. context is more generalized than the linguistic
The context granularities degrees: Our hypoth- one and the information deduced will be either
esis is having words in the same context imply that complementary information or completely new
they share common information which permit to information for the words of a cluster.
attribute an appropriate concept for these terms. In Our context definition is dynamic since it
this case a context is an appropriate support for a depends on the presence of the target words in
semantic interpretation i.e. it limits the associated one context. For example, with a cluster with four
knowledge of each word and gives a background words and by using two nominal groups, we find


Contextual Hierarchy Driven Ontology Learning

that these words are associated and we can give Algorithmic principles
them a concept so we are not obliged to look
for their documentary context. But, when the 8 In this section, we present an unsupervised hier-
words on a cluster do not belong to one of the four archical clustering algorithm namely Contextual
results of the linguistic context, we are obliged Concept Discovery (CCD) to extract ontological
to look deeply into the documentary context. So, concepts from HTML documents. It is based on
the expert evaluation thanks to our contextualiza- an incremental use of the partitioning algorithm
tion process is easier than the existing ones and Kmeans (McQueen, 1967) and is driven by the
it is done by respecting this order for each word structural contexts in which words occur. We
clusters: (1) Linguistic context (Nominal groups chose Kmeans because it is an unsupervised al-
based context, Verbal groups based context, gorithm able to classify a huge volume of objects
Prepositional groups based context, Conjunctional within a short execution time. The algorithm
groups based context); (2) Documentary context description is given in Figure 2. The different
(Sections of phrase based context, sentences based notations used in the CCD algorithm are given
context, paragraphs based context, documents in Table 1.
based context). The clustering algorithm proceeds in an in-
In order to present a context analysis example, cremental manner. It computes the occurrences
in the following sentence the possible accommo- of each word and selects their semantically closer
dations are hotels and residences, we find that are cooccurrents according to the context definition.
hotels and residences is the verbal group and the Then, it divides the clusters obtained at each
possible accommodations is the nominal group. step in order to refine the context of each group
So this sentence which is a context contains two of words. So, the algorithm refines at the same
contexts. time the context of a word and the context of
each cluster.

Table 1. Notations used by the CCD algorithm

F The file of the dataset

WC The set of word clusters WC ={C1, C2, C3, …, CT} with T: the
total number of clusters
K The number of clusters
P The highest number of words per cluster accepted by the user
M The possible number of words belonging to one cluster as a
complementary information which is defined and accepted by the
user
S The similarity measure
Di The word distribution into the various clusters Ci with Di ={C1,
C2, C3, …, Ci}
Ci A word cluster belonging to Di As Ci Є Di
Word-Number The number of words in the cluster Ci
(Ci)
Wi Each word belonging to Ci So Ci = {W 1, W 2 , …, W i}
Cc-c The cluster that has the closest centroid to the target word W i


 Contextual Hierarchy Driven Ontology Learning

Figure 2. The Contextual Concept Discovery Algorithm

0: Algorithm Contextual-Concepts-Discovery (In: F, K, P, M, S, Out: WC)

1: Apply our Context definition and compute the occurrences of the population {/*Step
0*/}
2: Di  Ф {/* Step 1*/}
3: Choose sporadically the K first centers
4: Assign each word to the cluster that has the closest centroid
5: Recalculate the positions of the centroids
6: if (the positions of the centroids did not change) then
7: go to the step 10
8: else
9: go to the step 4.
10: Di  Di U {C1, C2, C3, …, Ck}
11: For all Ci Є Di do {/* Step 2*/}
12: if (Word-Number (Ci) ≤ P) then
13: WC  WC U {Ci}
14: Di  Di \ {Ci}
15: else
16: Di  Di \ {Ci}
17: Unbolt the words Wi belonging to the cluster Ci
18: Compute the value of K
19: Go to the steps 3, 4, 5 and 6
20: For all Ci Є WC do {/* Step 3*/}
21: if (Word-Number (Ci) = 1 and Wi Є Ci) then
22: Calculate the position of Wi to the existing centroids of the clusters Ci Є WC
23: if (Word-Number (Cc-c) > P+M) then
24: Choose another cluster Cc-c that has the following closest centroid
25: Go to the step 20
26: else
27: WC  WC \ {Cc-c, Ci}
28: Assign Wi to the cluster Cc-c
29: WC  WC U {Cc-c}
30: Form a collection of web documents related to the domain {/* Step 4*/}
31: Process the corpus
32: Analyze the corpus
33: Extract the various contexts
34: For each Context type Do
35: For all Ci Є WC Do
36: Find the cooccurrents of the cluster Ci in each sample of the chosen context
37: Compute the Credibility Degree of the cluster Ci
38: Go to step 35
39: Go to step 34
40: Return (WC) + the different credibility degrees associated to each context and each
cluster Ci of WC
41: End

Also, it offers to the user the possibility to the accepted margin M representing an additional
choose either a complete automatic execution or number of words in a resulting cluster accepted
an interactive one. If he/she decides the first ex- by the user and the similarity measure S. If he
ecution manner, he/she should either define some prefers to evaluate the intermediate word clusters,
parameters or choose the default ones resulting he should choose the interactive execution. In
from our empirical experiments. These parameters this case, the algorithm allows him to analyze
are: the highest number of words per cluster P, the word cluster at the end of each intermediate

0
Contextual Hierarchy Driven Ontology Learning

clustering in order to define the value of k’ and is not defined by the user but computed by
to decide whether he prefers to continue within the system. We implement a proportional
an interactive execution or to run an automatic function that automatically defines the value
one for the rest of the clustering process. In the of K’. Based on several empirical experi-
interactive execution, the process takes longer ments and the domain experts’ knowledge,
than the automatic one but it offers an opportunity we define a proportional link needed for our
to the user to intervene in order to obtain better function. By using this information and solv-
hierarchical word clusters. ing the following equation, we can compute
As input to this algorithm, the user should the value of K’. The useful equation, which
first choose the input dataset. Secondly, he should is applied in any domain, is:
define the number of clusters K, and chooses
whether he prefers an automatic execution or an K’ = a *ln (Word-Number (Ci)* b)
interactive one. An automatic execution of our (2)
algorithm is defined in four steps:
If the number of words in a cluster (Word-
• Step 0: Applying the context definition. In Number (Ci)) is less than (1/b), the value of
this step, the algorithm takes into account K’ will not be computed but defined as 2.
the data file in which we find the word and When the number of words per cluster Ci
their outbuilding to the html tags. Based on is less or equal to P, we include the cluster
this input, it applies the context definition Ci to the set WC.
and returns a file containing the candidates • Step 3: Affecting the single word to the
(words), their cooccurrents (attributes) and formed clusters. When applying the division
their associated weightings. process, we can obtain clusters with only one
• Step 1: Executing the first Kmeans. The word. Our idea is to automatically associate
goal of this step is to execute the kmeans each word alone in a cluster resulting from
algorithm in order to obtain the first word step 2. Another problem appears when the
distribution Di. We obtain k word clus- algorithm affects too many words to the
ters. same cluster (by respecting the similarity).
In this case and for a little number of clusters,
These steps 0 and 1 concern the application we can obtain clusters with a great number
of the structural context which is deduced from of words. If a word is assigned to a cluster
the various analyses, computed from the dataset already containing P+M words, the CCD
F and stored in a matrix and the execution of algorithm will choose the cluster which is
the partitioning algorithm to obtain the first k the closest centroid to the target word.
clusters.
Choosing an interactive process implies apply-
• Step 2: Performing the hierarchical cluster- ing the same steps but with the user intervention
ing. This step finds clusters respecting the firstly after step 1 and during step 2 and 3. The
criterion defined by the user: the P value. For clustering method adopts a vector-space model
each intermediate execution of kmeans, we and represents a term as a vector containing at-
should define the value K’ which represents tributes that belong to the corpus and are stored
the number of clusters related to the words in a matrix. In step 2, if a cluster is divided, the
of a cluster that could be divided. In an auto- algorithm allows to refine the context of each
matic execution of the algorithm, this value cluster by taking into account only the associated


 Contextual Hierarchy Driven Ontology Learning

attributes of its belonging words. By applying this Those documents (French language) contain
method, the similarity computed better represents some HTML structure. We collect them by
the association degree between each two words. By using the HTTrack Website Copier. Since
applying the step 3, it avoids the cases of having we aim to perform an incrementally concept
only one word in a cluster and those containing extraction process, we start the process
the majority of the single words. with key and title tags to give an outline of
the domain’s semantic information. These
• Step 4: Evaluating the word clusters. To terms are those chosen by the site designer
obtain a domain web collection of French as keywords, document titles, glossaries, etc.
documents, we use a cleaner (for example Our dataset is composed of 872 words. So,
HTTrack Website Copier). Then, we treat our general context is the first two levels of
them thanks to the pre-processing step of the contextual model (key tags + title tags+
our system. This last cleans and structures headings tags). In the following research, this
the collected web pages. Then, we perform method will progressively integrate all the
various analyses (Karoui et al., 2006). Now, html elements and the terms of the corpus
the problem is that the expert is incapable, in order to obtain more domain concepts.
even when he is given all the results of the • Experimental protocol. We perform two
analysis, to find the possible association of experiments. Firstly, we compare between
the targets words, especially when working two contextual definitions which are a simple
on a big corpus. In order to facilitate this window and our context definition. Our new
process, we define a semantic index which context definition improves the words’ pairs
represents the credibility of the target words’ weighting and the words’ pair’s similarities.
association in relation with the different con- These results are presented in (Karoui et al.,
texts. This index is named credibility degree. 2006) in which we explain clearly the advan-
It is computed for each word cluster and tage of our context. Secondly, we keep our
for each context definition in an automated context definition and we apply it with two
way. More details about this criterion and algorithms which are Kmeans and CCD. We
its importance are provided in sub section evaluate the CCD algorithm results by com-
titled the impact of the credibility degree. paring them to those obtained by the Kmeans
one. We chose the Euclidian distance as a
In the following section, we experiment with similarity measure. The Kmeans algorithm
the Contextual Concept Discovery algorithm. distributes them in 156 clusters. The CCD
algorithm is experimented with various
values for each parameter. We present an
experImentAl evAluAtIon oF automatic execution while respectively ap-
the contextuAl concept plying to k, P and M the values of 20, 10
dIscovery AlgorIthm (ccd) and 22 (more significant results). We obtain
162 clusters.
Our objective is to extract and evaluate the on- • Evaluation protocol. Our evaluation proc-
tological concepts. ess is applied within a close interaction with
the domain experts. Firstly, we execute the
• Source of information. In our experiments, Kmeans algorithm and the three first steps
we use HTML documents (600 pages) col- of the CCD algorithm. The obtained results
lected from the Canada tourism web sites. are presented to two domain experts. Indi-


Contextual Hierarchy Driven Ontology Learning

vidually, each of them evaluates and labels Semantic Interpretation: The domain expert
manually the word clusters (318 clusters). notes that there are three types of word clusters
Then, they work together in order to discuss which are advisable clusters, improper clusters
about the results, their label propositions and and unknown clusters. Advisable clusters are
to give us only one evaluation and labelling those for which the expert is able to associate a
result about which they agree. After this label and in which words belonging to the same
first evaluation step, we have discussed with group are close semantically. Improper clusters
them, analysed their results and defined four are either those with an amount of words having
criteria which are the word distribution, no relation with the principle extracted concept,
the semantic interpretation, the extracted or those containing more than one concept. Un-
concepts and the generality degree of the known clusters are clusters where words do not
extracted concepts. Secondly, in order to have any semantically relation and the expert
evaluate the impact of the evaluation step could not find any semantic interpretation. Thanks
of our algorithm, we complete the execu- to the P and the M values, in each word cluster,
tion of the CCD algorithm (Step4). Then, the percentage of noisy elements decreases a lot.
we present the results to two other domain As a consequence, the percentage of unknown
experts. They maintain the same evaluation and improper clusters is reduced (respectively
protocol which is a manually one and they 20. 51% and 26. 28% for the kmeans algorithm
provide their results, remarks and conclu- versus only 14. 81% and 16. 66% for the CCD
sions. For us, we have remarked that the algorithm). Moreover, we obtain 68.52% advisable
second evaluation is easier and faster than clusters with the CCD algorithm which is more
the first one. The conclusions of the experts important than only 53.2% with the Kmeans one
confirm our remarks. (see Table 2).
Extracted Concepts: We take into account only
comparison between kmeans and the advisable clusters in the two cases and we
the ccd Algorithm compute the precision. In our study, Precision is
the ratio of relevant words having between them
We start by presenting the four criteria with the a great semantic similarity with the total words
first evaluation step. Then, we present the impact for a given cluster. By applying this criterion, we
of our parameters (P and M) in the results. obtain respectively 86.18% and 86.61% with the
Word Distribution: With the Kmeans algo- kmeans algorithm and the CCD one.
rithm, we have 13% of our initial words that are Generality Degree of the Extracted Concepts:
grouped together. While with the CCD algorithm, Another element which affects the concept’s qual-
we obtain only 3.66% of our initial set of words ity is the level of generality for a concept. In order
in the same cluster. to evaluate the generality degree of the concepts,
we focus only on concepts extracted from the

Table 2. The detail of the extracted concepts

Advisable Improper Unknown
clusters clusters clusters
Kmeans Algorithm 53. 2 % 26. 28 % 20. 51 %
CCD Algorithm 68.52% 16.66% 14.81%


 Contextual Hierarchy Driven Ontology Learning

advisable clusters. We based our manual evalu- The impact of the Credibility Degree: Let us
ation on the OMT thesaurus (1999). It contains take some resulting word cluster in order to explain
generic and open terms presenting general key concretely the credibility degree’s importance:
domain concepts. In our experiments, we obtain Our Credibility Degree Computation criterion
respectively with the kmeans algorithm and the is executed on a set of word clusters in order to
CCD one 78.31% and 85.58% general concepts. compute their credibility degrees. For instance,
Also, we remark that the clusters obtained with with the example 1 (Table 3) and according to one
the CCD algorithm are more enhanced than with context definition (nominal groups or sentence),
the Kmeans algorithm. For example, we find with the algorithm finds all the possible combination
the Kmeans algorithm the cluster C1 :{Event, fes- in the context i.e. tries to find the four words
tival, music} while with the CCD one we obtain (academy, golf, golfer, club), then the association
C2:{Event, festival, music, party}. of three words and so on. For each found associa-
The impact of the P parameter and the step tion, it presents the associated words and gives a
2 of the CCD Algorithm: A first execution of the degree representing how many times this type of
kmeans algorithm with 20 clusters (step1) gives a association is found. For example, with the same
cluster having 68.69% of the initial set of words. example, it finds two possible associations with
By defining the P parameter, we decide to divide three words which are {academy, golf, golfer} and
the word clusters and consequently to perform {golf, golfer, club} so the credibility degree is 32
an intermediate clustering based only on the i.e. two associations of three words.
common attributes of the P words. So, we refine Our criterion has several functionalities which
the context of this cluster and we obtain better are:
results thanks to new similarities between the P
words. For example, with the kmeans algorithm, • Finding the associations between some
we found the word civilization with a big set of words in order to facilitate the labelling step.
words without any relation between them, but With Example 5, our criterion finds only the
with the CCD algorithm, we found it in a cluster association that permits the user to give a
where there are words semantically similar like label by him self like ‘excursion on foot’.
archaeology, ethnology, etc. • Finding at the same time the available as-
The impact of the step 3 of the CCD Algorithm: sociation in the context and the concept
The CCD algorithm allows assigning these single (Example 2, the concept is ‘civilization’).
words to the existing clusters. We remarked that • Detecting the noisy elements in a cluster
some words are assigned to the appropriate cluster. and either delete them or move them to
For example before and after the step 3, we obtain
respectively the clusters {Academy, club, golfer}
and {Academy, club, golfer, golf}. Table 3. Examples of term clusters
We now present the importance of the cred-
ibility degree criterion in the evaluation task, Examples Word Clusters

especially to facilitate the semantic interpretations Example 1 academy, golf, golfer, club

of the results by the domain experts. Example 2 Civilization, archeology, ethnology, people
Example 3 Park, national, cliff, rock

the evaluation step Example 4 Cult, church, evangelization, memory, reli-

gious, sanctuary
Example 5 excursion, foot, person
In this section, we expose the contribution of our
evaluation step included in the CCD algorithm. Example 6 Hiker, gorges


Contextual Hierarchy Driven Ontology Learning

another cluster. For instance, in Example 5, ones based on the word associations deduced
the word ‘hiker’ is found inside the several from the various contexts and the quantitative
association returned by our algorithm and ones resulting from the computed credibility de-
corresponding to ‘excursion’ and ‘foot’ but gree index. The qualitative evaluation provides a
the problem is that this word belongs to the semantic support for an easy user interpretation.
Example 6. Our evaluation process decides Moreover, our proposition, based on a large col-
to remove it from example 6 and include it lection of domain web documents, several context
into example 5 because no association is definitions and different granularity degrees,
found related to the other words of cluster 6. permits to an ordinary user to help the expert by
If the words exist in two associations related manipulating the word clusters and giving him
to two clusters, the CDD algorithm presents semantic tags as suggestions. Consequently, the
the word using the red colour to announce expert should decide on the appropriateness of
a possible ambiguous situation. these labels as well as clusters homogeneities
• Enhancing a cluster by other words from which are not labeled. Moreover, our algorithm
the associations. For Example 3, we can assures ontology reuse and evolution since the
enrich the group by the word nature and elements on which the expert’s interpretations
inheritance and find the concept which is are based (the provided word associations) de-
‘natural inheritance’. pend on the web changes. For example, when
the web documents change, the various extracted
Since our evaluation task is based on various contexts change too and the results of the map-
context definitions, if the user does not find a ping operation between the words belonging to
connection between some words by using the the clusters and the contexts are updated. This
linguistic contexts, he can analyse the provided resulting information about the word clusters is
association by the documentary contexts in which presented to the experts in order to help them
the probability to find more relation is bigger than during the evaluation task. Then, they are stored
with the first context type (Example 4). with the experts’ comments in order to be reused
Thanks to the credibility degrees computed by another expert either during the same period or
for each cluster and for each context, the user later (after some months or years depending on the
obtains an amount of information useful and in frequency of updates). Also, thanks to our context
some cases sufficient to manipulate (delete word, based hierarchical clustering, the CDD algorithm
remove word, etc.), evaluate and label the cluster. follows the document changes whether it is at the
For example, for a same cluster, if he finds the level of the structure or the content.
three credibility degrees (51, 43, 38, 215) which are Our results are presented to the user in a HTML
a quantitative indications, he starts by analysing format (Figure 3).
the association with 5 words. If it is not sufficient,
he analyses the three associations of four words discussion
(43) and so on. If the information returned by our
algorithm to this cluster and for one context is not In a previous work, we have compared different
enough, he can look to the other credibility de- clustering algorithms (kohonen, EM, etc.) to
grees provided by the other contexts by respecting kmeans and we have remarked that all of them
the previous order (linguistic context type then give as results big clusters. Those clusters con-
documentary context one). tain a lot of noisy words and consequently their
So, our web driven concept evaluation method evaluation by the expert is very difficult. That is
provides two revealing aspects: the qualitative why we have decided to define a new clustering


 Contextual Hierarchy Driven Ontology Learning

Figure 3. An extract from the resulting file related to a group nominal context

algorithm in order to resolve this problem and provide information. In this work, we can not use
compare it, as first experiments, to the kmeans one. the precision and the recall since we used a col-
In this section, we have shown that the clustering lected domain pages which have not pre-clusters.
method guided by the structural context definition But our architecture [18] provide a pre-treatment
gives better results than the simple kmeans (with component treating the HTML documents, so the
a usual context definition). The generic context hierarchical clustering method named CCD can
is based on a contextual hierarchy constituted by be applied either in a benchmark related to one
the HTML links. In order to keep our method specific domain (in this case we can apply the
functional, we should have a minimum structure precision and the recall measure since we have
in the documents. However, the structure analysis classified terms) or a domain collection either
adapts the context definition by taking into ac- collected by the researcher or given by a third
count most of the used HTML tags. Moreover, party (institution, industry, etc.).
the absence of some HTML tags does not affect
the functionality of our method. For example, if
we do not have the sub title tag <H1>, the method conclusIon
can be applied since it is incremental and searches
the word’s cooccurrents in other lower levels of It is important to discover and organize the avail-
the C.H (in the tags <p>, <td>, etc.). In addition, able knowledge about an application domain in
when the tag <title>, which permits to initiate the a coherent manner. In this paper, we defined a
clustering process and gives future concepts, is hierarchical clustering algorithm namely Con-
not present, we have other HTML key tags which textual Concept Discovery (CCD), based on an


Contextual Hierarchy Driven Ontology Learning

incremental use of the partitioning algorithm are going to pursue the reflection to develop our
Kmeans, guided by a structural contextual defini- evaluation method that will be based on some
tion and providing a web driven evaluation task to synthetic criteria computed by considering the
support the domain experts. The CCD algorithm notion of ‘contextualization’.
proceeds in an incremental manner to refine the
context and to improve the conceptual quality
of each word cluster. The last functionality of Future reseArch dIrectIons
our algorithm is a new evaluation method based
on a large web collection and different contexts In this chapter, we defined a hierarchical clustering
extracted from it. We implement and experiment algorithm namely Contextual Concept Discov-
the CCD algorithm to incrementally extract the ery (CCD). It is guided by a structural context
ontological concepts. We show that the context- definition and provides a web driven evaluation.
based hierarchical algorithm gives better results It proceeds in an incremental manner to refine
than a usual context definition and an existing the context and improve the conceptual quality
clustering method. Moreover, our evaluation of each word cluster. The last functionality of
process permits to help either an ordinary user our algorithm is a new evaluation method based
(knowledge engineer) or the expert to take the on a large web collection and different contexts
right decision about the semantic homogeneity of extracted from it. We implement and experiment
a cluster. So, we have defined an algorithm that the CCD algorithm to incrementally extract the
at the same time extracts the domain concepts, ontological concepts. We show that the context-
evaluates them and considers their contexts dur- based hierarchical algorithm gives better results
ing the two activities. In our research, the CCD than a usual context definition and an existing
algorithm is defined for the ontology building. clustering method. Moreover, our evaluation
But, it is important in many other applications like process helps either an ordinary user (knowledge
information retrieval, annotation, multi-agent sys- engineer) or the expert to take the right decision
tems, etc. For the information retrieval research, about the semantic homogeneity of a cluster. So,
first, discovering knowledge inside the documents we have defined an algorithm that at the same
is very important. By using our algorithm, we can time extracts the domain concepts, evaluates
extract their concepts and evaluate them in order them and considers their contexts during the two
to understand their content. Second, we should activities. In our research, the CCD algorithm is
refine the user request. The CCD algorithm dis- used for the ontology building. But, it is impor-
covers the domain concepts that can be used to tant in many other applications like information
replace the keywords from the requests. For the retrieval, annotation, multi-agent systems, etc.
annotation, we need also discovering concepts For the information retrieval research, we need
from web documents by using our algorithm. to refine the user request. The CCD algorithm
As a perspective, we will define and experiment discovers the domain concepts that can be used
(with the CCD algorithm) a linguistic context to replace the keywords from the requests. For
and a documentary context combined with the the annotation, we also need to locate concepts in
structural one and apply them to words belonging web documents by using our algorithm. Recently,
to the HTML tags in order to enhance the set of we defined an evaluation method based on some
extracted concepts, obtain more specific concepts synthetic criteria computed by considering the
and take into account other types of documents notion of ‘contextualization’. In addition, we
such as textual ones. We will develop ideas related are working on relation extraction and concept
to the instance and relation discovery. Also, we hierarchy evaluation. We are developing an ap-


 Contextual Hierarchy Driven Ontology Learning

plication that visualizes the concept hierarchy and ings of the ECAI 2000 Workshop on Ontology
evaluates it either automatically or manually. We Learning.
now plan to define and experiment (with the CCD
Davulcu, H., Vadrevu, S. & Nagarajan, S.
algorithm) linguistic, documentary and structural
(1998). OntoMiner: Boostrapping ontologies
contexts and apply them to other HTML docu-
from overlapping domain specific web sites. In
ments. We also will take into account other types
AAAI’98/IAAI’98 Proceedings of the 15th Na-
of document such as textual ones. Later, we hope
tional Conference on Artificial Intelligence and
to link this work together - concept extraction and
the 10th Conference on Innovative Applications
evaluation, hierarchy evaluation and visualization
of Artificial Intelligence.
and relation extraction - as a unique application
useful for the semantic web community. The Faure, D., Nedellec, C. & Rouveirol, C. (1998).
background of this research is related to the notion Acquisition of semantic knowledge uing machine
of ‘context’, the ontology learning (that contains learning methods: the system ASIUM. Technical
the concept extraction) and the ontology evalua- report number ICS-TR-88-16, inference and learn-
tion. In the related work, I present some research ing group, University of Paris-sud.
related to these three subjects. In the readings
Gale, W.A., Church, K.W., & Yarowsky, D.
section, I add more references related to these
(1992a). Using bilingual materials to develop
topics and others such as relation extraction, the
word sense disambiguation methods. Proceedings
use of context in other domains, some measures
of the International Conference on Theoretical
to extract concepts, etc.
and Methodological Issues in Machine Transla-
tion, 101-112.
reFerences Gruber, T.R. (1993). Towards Principles for the
Design of Ontologies used for Knowledge sharing.
Brewster, C., Alani, H., Dasmahapatra, S. & In N. Guarino & R. Poli (Eds), Formal Ontology
Wilks, Y. (2004). Data Driven Ontology Evalu- in Conceptual Analysis and Knowledge Repre-
ation. Proceedings of Int. Conf. on Language sentation. Deventer, the Netherlands: Kluwer
Resources and Evaluation, Lisbon. Academic Publishers.
Brézillon, P. (1999). Context in problem solving: Han, H. & Elmasri, R (2000). Architecture of
A survey. The Knowledge Engineering Review, WebOntEx: A system for automatic extraction of
Volume: 14, Issue: 1, Pages: 1-34. ontologies from the Web. WCM 2000.
Brézillon, P. (1993). Context in Artificial Intel- Karoui, L., Bennacer, N. & Aufaure, M-A. (2006).
ligence: II. Key elements of contexts. Computer Extraction de concepts guidée par le contexte.
and AI. XIIIème Rencontres de la Société Francophone
de Classification SFC’06, pp 119-123.
Brown, S., Peter, F., Della, P., Vincent, J. & Mercer
Robert, L. (1991). Word sense disambiguation us- Kiyota, Y. & Kurohashi, S. (2001). Automatic
ing statistical methods. Proceedings of the 29th summarization of Japanese sentences and its ap-
Annual Meeting of Association for Computational plication to a WWW KWIC index. Proceedings
Linguistics, Berkeley, California. of the 2001 Symposium on applications and the
internet, page 120.
Chalendar, G. & Grau, B. (2000). SVETLAN A
system to classify nouns in context. Proceed- MacQueen, J. (1967). Some methods for clas-
sification and analysis of multivariate observa-


Contextual Hierarchy Driven Ontology Learning

tions. Proceedings of 5th Berkeley Symposium Vazirgiannis, M., Halkidi, M. & Gunopoulos, D.
on Mathematics, Statistics and Probability, (2003). Uncertaintly handling and quality assess-
1:281- 298, 1967. men in data mining. Springer.
Meadche, A. & Staab S. (2001). Ontology learning Voorhes, E-M. (1993). Using WordNet to disam-
for the semantic Web. IEEE Journal on Intelligent biguate word senses for text retrieval. ACM SIGIR
Systems, Vol. 16, No. 2, 72-79. Conference on Research and Development in
Information Retrieval, Pittsburgh, Pennsylvania,
Meadche, A & Staab, S. (2002). Measuring si-
171-180.
milarity between ontologies. Proc. CIKM 2002.
LNAI vol.2473. Widmer, G. (1996). Recognition and exploitation
of contextual clues via incremental meta-learn-
Michelet, B. (1988). L’analyse des associations.
ing. Oesterreichisches Forshungsinstitut fuer
Doctoral thesis, Université de Paris VII, UFR de
artificial intelligence, Wien, TR-96-01.
Chimie, Paris.
Yarowsky, D. (1993). One sense per collocation.
Navigli, R., Velardi, P., Cucchiarelli, A. & Neri, F.
Proceedings of ARPA Human Language Technol-
(2004). Quantitative and qualitative evaluation of
ogy Workshop, Princeton, New Jersey, 266-271.
the ontolearn ontology learning system. In Proc.
Of ECAI-2004 Workshop on Ontology Learning Yarowsky, D. (1992). Word sense disambiguation
and Population, Valencia, Spain. using statistical models of Roget’s categories
trained on large corpora. Proceedings of the 14th
Navigli, R. & Velardi, P. (1998). Learning domain
International Conference on Computational Lin-
ontologies from document warehousees and dedi-
guistics, COLING’92, Nantes, France, 454-460.
cated web sites. In AAAI’98/IAAI’98 Proceedings
of the 15th National Conference on Artificial In-
telligence and the 10th Conference on Innovative
Applications of Artificial Intelligence. AddItIonAl reAdIng
Paslaru Bontas, E. (2005). Using context in-
Bastien, C. (1992). The shift between logic and
formation to improve ontology reuse. Doctoral
knowledge. Le courier du CNRS, Numéro Spécial
Workshop at the 17th Conference on Advanced
“sciences cognitives”, No 79, pp.38.
Information Systems Engineering CAiSE’05.
Barwise, J. & Seligman, J. (1992). The rights
Thesauus on tourism and leisure activities (1999).
and wrongs of natural regularity. In J. Tomberlin
First edition copyright Secretariat of state for tour-
(ed.), Philosophical Perspectives 8: Logic and
ism of France and World Tourism Organisation.
language. Atascadero, CA: Ridgeview.
Padmini, S. (1992). Thesaurus construction. In
Berland, M. & Charniak, E. (1999). Finding parts
William B. Frakes & Ricardo Baeza-yates (Eds.),
in very large corpora. In Proc. of the 37th Annual
Information retrieval: data structures and algo-
Meeting of the Association for Computational
rithms, 161-218. New York: Prentice-Hall.
Linguistics.
Turney, P. (1996). The management of context-
Bruandet, M.F. (1990). Domain Knowledge Ac-
sensitive features: A review of strategies. Proceed-
quisition for an Intelligent Information Retrieval
ings of the ICML-96 Workshop on Learning in
System : Strategies and Tools. In Expert Systems
Context-Sensitive Domains, pp. 53-69.
Applications EXPERSYS - 90, Grenoble, pp231-
235.


 Contextual Hierarchy Driven Ontology Learning

Buitelaar, P., Olejnik, D. & Sintek, M. (2004). A discovery of part-whole relations. In Proc. of the
protege plug-in for ontology extraction from text HLT- NAACL.
based on linguistic analysis. ESWS.
Goh, C.H., Madnick, S.E. & Siegel, M.D (1994).
Buyukkokten, O., Garcia-Molina, H. & Paepcke, Context interchange: overcoming the challenge
A. (2001). Accordion summarization for end-game of large-scale interoperable database systems in
browsing on PDAs and Cellular Phones. In Proc. a dynamic environment. In proceedings of the
Of Conference on Human Factors in Computing third international conference on information and
Systems (CHI01). knowledge management, Gaithersburg (USA),
pp. 337-346.
Cai, D., Yu, S., Wen, J. & Ma, W. (2004). Block-
based web search. Proceedings of the 27 Annual Grefenstette, G. (1992). Use of syntatic context to
International ACM SIGIR Conference on Re- produce terms association list for text retrieval. In
search and Development in Information Retrieval, Conference in Researche and Developement in
pages 456-463. Information Retrieval (SIGIR’92), Copenhagen,
Denmarke, pages 89-97.
Cilibrasi, R. & Vitanyi, P. (2006). Automatic
Extraction of Meaning from the Web. The IEEE Guarino, N. (1998). Formal Ontologies and In-
International Symposium on Information Theory, formation Systems. FOIS’98. Trento, Italy: IOS
Seattle, WA. Press.
Cimiano, P. & Volker, J. (2005). Towards large- Haddad, M-H. (2002). Extraction et impact des
scale, open-domain and ontology-based named connaissances sur les performances des systèmes
entity classification. RANLP. de recherche d’information. Doctoral thesis,
université joseph Fourier.
Claveau, V., Sebbillot, P., Fabre, C.& Bouillon,
P. (2003). Learning Semantic Lexicons from a Hearst, M-A. (1992). Automatic acquisition of
Part-of-Speech and Semantically Tagged Corpus hyponyms from large text corpora. In proceed-
using Inductive Logic Programming. Journal ings of the fourteenth international conference
of Machine Learning Research, special issue on computational linguistics, Nantes, France,
on ILP. pages 539-545.
Compton, P. & Jansen R. (1988). Knowledge in Holsapple, C. & Joshi, K.D. (2005). A collaborative
context: a strategy for expert system maintenance. approach to ontology design. Communications of
In J.Siekmann (Ed), Lecture Notes in Artificial ACM, Vol. 45(2) pp. 42-47.
Intelligence, Subseries in computer sciences,
Iria, J. & Ciravegna, F. (2005). Relation Extrac-
Vol. 406.
tion for Mining the Semantic Web. Dagstuhl
Gamallo, P., Gonzalez, M., Agustini, A., Lopes, Seminar on Machine Learning for the Semantic
G. & de Lima, V. S. (2002). Mapping syntactic Web, Dagstuhl, Germany.
dependencies onto semantic relations. ECAI
Karoui, L. & El Khadi, N. (2007). Relation Extrac-
Workshop on Machine Learning and Natural
tion and Validation Algorithm. 4th International
Language Processing for Ontology Engineering,
Conference on Distributed Computing and Inter-
Lyon, France.
net Technology, LNCS volume 3347.
Girju, R., Badulescu, A. & Moldovan, D. (2003).
Karoui, L. (2006). Intelligent Ontology Learning
Learning semantic constraints for the automatic
based on Context: Answering Crucial Questions.

0
Contextual Hierarchy Driven Ontology Learning

The IEEE International Conference on Compu- Meadche, A. & Staab, S. (2002). Measuring
tational Intelligence for Modelling, Control and similarity between ontologies. Proc. CIKM 2002.
Automation - CIMCA06, Sydney. LNAI vol.2473.
Karoui, L. & Aufaure, M-A (2007). Revealing Maedche, A. & Staab, S. (2000). Discovering
Criteria for the Ontology Evaluation Task. Special conceptual relations from text. ECAI.
Issue of Journal of Internet Technology (JIT) on
Mittal, V.O & Paris, C.L. (1995). Use of context
Ontology Technology and Its Applications.
in explanations systems. International Journal
Karoui, L., Aufaure, M-A. & Bennacer, N. (2007). of Expert Systems with Applications, Vol.8 No.
Analyses and Fundamental ideas for a Relation 4, pp. 491-504.
Extraction Approach. The IEEE proceedings
Morin, E. (1999). Using Lexico-Syntactic Pat-
of the Workshop on Data Mining and Business
terns to Extract Semantic Relations between
Intelligence in conjunction with the IEEE 23rd
Terms from Technical Corpus. In Proceedings,
International Conference on Data Engineering
5th International Congress on Terminology and
(ICDE’07), Turkey.
Knowledge Engineering (TKE), 268–278. Ter-
Loh, S., Wives, L. & Oliveira, J.P. (2000). Con- mNet, Innsbruck, Austria.
cept-based knowledge discovery in texts extracted
Nakache, J.P. & Confais, J. (2005). Approche
from the web. In SIGKDD Explorations. Volume
pragmatique de la classification : arbres hié-
2, Issue 1, Page 29.
rarchiques, partitionnements. Editions Technip,
Magnini, B., Negri, M., Pianta, E., Romano, Paris.
L., Speranza, M. & Sprugnoli, R. (2005). From
Nazarenko, A. (1994). Compréhension du langage
Text to Knowledge for the Semantic Web: the
naturel : le problème de la causalité. Doctoral
ONTOTEXT Project. SWAP 2005, Semantic Web
thesis.
Applications and Perspectives, Trento.
Reinberger, M.L., Spyns, P. & Pretorius, A.J.
McCarthy, J. (1993). Notes on formalization
(2004). Automatic initiation of an ontology. On
context. Proceedings of the 13th IJCAI, Vol. 1,
the Move to Meaningful Internet Systems 2004:
pp. 555-560.
CoopIS, DOA, and ODBASE, LNCS 3290, Napa,
McDonald, R., Pereira, F., Kulick, S., Winters, S., Cyprus, pp. 600-617.
Jin, Y. & White, P. (2005). Simple Algorithms for
Sabou, M. (2004). Extracting ontologies from
Complex Relation Extraction with Applications
software documentation: a semi-automatic
to Biomedical IE. 43rd Annual Meeting of the
method and its evaluation. In Proceedings of the
Association for Computational Linguistics (ACL-
ECAI-2004 Workshop on Ontology Learning and
2005), Ann Arbour, Michigan, pp. 491-498.
Population (ECAI-OLP).
Maedche, A. (2002). Ontology Learning for the
Schutz, A. & Buitelaar, P. (2005). RelExt: A Tool
Semantic Web. Kluwer Academic Publishers,
for Relation Extraction from Text in Ontology
Norwell, MA.
Extension. 4th International Semantic Web Con-
Meadche, A. & Staab, S. (2001). Ontology learn- ference (ISWC-2005), Galway, pp. 593-606.
ing for the semantic Web. The IEEE Journal on
Schmid, H. (1994). Probabilistic Part-of-Speech
Intelligent Systems, Vol. 16, No. 2, pp. 72-79.
Tagging Using Decision Trees. IMS-CL, Institut
Für maschinelle Sprachverarbeitung, Universität
Stuttgart, Germany.


 Contextual Hierarchy Driven Ontology Learning

Soderland, S. (1999). Learning information ex- Van. Rijsbergen, C.J. (1979). Information Re-
traction rules for semi-structured and free text. trieval. London: Butterworths.
Machine Learning.
Robison, H.R. (1970). Computer detectable
Soo-Guan Khoo, C. (1995). Automatic identifi- semantic structures. Information Storage and
cation of causale relations in text and their use Retrieval, 6, 273-288.
for improving precision in information retrieval.
Walther, E., Eriksson, H. & Musen, M.A. (1992).
PhD thesis.
Plug-and-play: Construction of task-specific
Sciore, E., Siegel, M. & Rosenthal, A. (1992). expert-system shells using sharable context on-
Context interchange using meta-attributes. Pro- tologies. Proceedings of the AAAI Workshop on
ceedings of the 1st International Conference in knowledge Representation Aspects of Knowledge
Information and knowledge Management, pp. Acquisition, San Jose, CA, pp. 191-198.
377-386.
Yangarber, R., Grishman, R. & Tapanainen, P.
Smith, B. (2003). Ontology. In L. Floridi (ed.), (2000). Unsupervised Discovery of Scenario-
Blackwell Guide to the Philosophy of Comput- Level Patterns for Information Extraction. 6th
ing and Information. Oxford: Blackwell, pp. ANLP, Seattle, pp. 282-289.
155–166.
Stevenson, M. & Greenwood, M. (2005). A Se-
Spyns, P. (2005). Evalexon: Assessing triples mantic Approach to IE Pattern Induction. 43rd
mined from texts. Technical report 09, STAR Lab, Meeting of the Association for Computational
Brussels, Belgium. Linguistics (ACL-05), Ann Arbour, Michigan,
p. 379-386.


 

Chapter II
A Review of Fuzzy Models for
the Semantic Web
Hailong Wang
Northeastern University, China

Zongmin Ma
Northeastern University, China

Li Yan
Northeastern University, China

Jingwei Cheng
Northeastern University, China

AbstrAct

In the Semantic Web context, information would be retrieved, processed, shared, reused and aligned
in the maximum automatic way possible. Our experience with such applications in the Semantic Web
has shown that these are rarely a matter of true or false but rather procedures that require degrees of
relatedness, similarity, or ranking. Apart from the wealth of applications that are inherently imprecise,
information itself is many times imprecise or vague. In order to be able to represent and reason with
such type of information in the Semantic Web, different general approaches for extending semantic web
languages with the ability to represent imprecision and uncertainty has been explored. In this chapter,
we focus our attention on fuzzy extension approaches which are based on fuzzy set theory. We review
the existing proposals for extending the theoretical counterpart of the semantic web languages, descrip-
tion logics (DLs), and the languages themselves. The following statements will include the expressive
power of the fuzzy DLs formalism and its syntax and semantic, knowledge base, the decidability of the
tableaux algorithm and its computational complexity etc. Also the fuzzy extension to OWL is discussed
in this chapter.

Copyright © 2009, IGI Global, distributing in print or electronic forms without written permission of IGI Global is prohibited.
 A Review of Fuzzy Models for the Semantic Web

IntroductIon in another ontology. To overcome the difficulty

arising from the crisp logics, existing ontology
The Semantic Web is an extension of the cur- languages need to be extended to be able to capture
rent web in which the web information can be uncertainty knowledge about the concepts, prop-
given well-defined semantic meaning, and thus erties and instances in the domain and to support
enabling better cooperation between computers reasoning with partial, imprecise information.
and people. From this point of view, we should Along this direction, researchers in the past have
find some methods which can describe the se- attempted to apply different formalisms such as
mantic meaning of the web. Fortunately, “ontol- Fuzzy logic (Zadeh, 1965), Rough set theory and
ogy” can do this. The core of the Semantic Web Bayesian probability as well as ad hoc heuristics
is “ontology” which refers to a set of vocabulary into ontology definition and reasoning.
to describe the conceptualization of a particular In this chapter, we review existing proposals
domain. Over the past few years, several ontol- to extend semantic web languages with the capa-
ogy definition languages for the Semantic Web bility to handle uncertain information to better
have emerged, including RDF(S), OIL, DAML, deal with the situations mentioned above. There
DAML+OIL, and OWL. Among them, OWL is are many ways of representing and dealing with
the newly released standard recommended by uncertainty. In this chapter, we restrict our at-
W3C. As the Semantic Web expects, OWL has tention to approaches that use fuzzy methods for
the reasoning nature because description logics representing uncertain information. In particular,
(DLs) (Baader, 2003) are essentially the theoretical we will not cover recent proposals for probabilistic
counterpart of OWL and play a crucial role in this extensions of semantic web languages. We will
context. DLs provide a logical reconstruction of also not discuss non-monotonic and non-standard
object-centric and frame-based knowledge rep- logics for representing uncertainty. As described
resentation languages. It is a subset of first-order above, existing Semantic Web languages are
logic that provides sound and decidable reasoning mainly based on logic and do not support repre-
support (Baader, 2003). senting imprecise and uncertain information. In
It is clear that DLs play a key role in the Se- this chapter, we therefore review a number of pro-
mantic Web. As with traditional crisp logic, any posals for extending logical languages with fuzzy
sentence in OWL, being asserted facts, domain extensions in more details. We focused on:
knowledge, or reasoning results, must be either
true or false and nothing in between. However, 1. Approaches that extend description logics
most real world domains contain uncertainty which play as the theoretical counterpart of
knowledge and incomplete or imprecise informa- the semantic web languages.
tion that is true only to a certain degree. Ontologies 2. Approaches that directly extend semantic
defined by these languages thus cannot quantify web languages, in particular OWL.
the degree of the overlap or inclusion between
two concepts, and cannot support reasoning in In the first category, we cover fuzzy exten-
which only partial information about a concept sions of description logics which are commonly
or individual in the domain can be obtained. accepted as being the formal basis of OWL. Even
Uncertainty becomes more prevalent when more though most approaches only cover logics that are
than on ontologies are involved where it is often much weaker than OWL, the methods proposed
the case that a concept defined in on ontology can can directly be applied to the corresponding subset
only find partial matches to one or more concepts of OWL without changes because the description


A Review of Fuzzy Models for the Semantic Web

logics play as the theory counterpart of the OWL. prelImInArIes And

When talking about the different approaches, bAckground
we will survey them according to the expressive
power from weaker to stronger. And in the fol- description logics
lowing survey, we should discuss the following
issues of the different approaches: In the last decade a substantial amount of work
has been carried out in the context of Description
• Expressiveness of the logical language Logics. DLs are a logical reconstruction of the
• The syntax and semantic of the fuzzy exten- so-called frame-based knowledge representation
sion to description logics languages, with the aim of providing a simple well-
• The components of the knowledge base established Tarski-style declarative semantics to
• Tableaux algorithm for the description log- capture the meaning of the most popular features
ics of structured representation of knowledge. Nowa-
• The decidability and complexity of the days, DLs have gained even more popularity due
tableaux algorithm to their application in the context of the Semantic
Web (Berners-Lee, 2001).
Indeed, the balance of expressive power The recent research about description logics
and the computability of the fuzzy extension of can be divided into three categories:
description logics is a hot topic of the research.
Generally speaking, the more expressive of the • Introducing the theoretical foundations of
description logic, the higher computational com- description logics, addressing some of the
plexity of it, so we should consider the balance most recent developments in theoretical
of the two factors in a real application. At last, in research in the area;
the latter category, we also review a number of • Focusing on the implementation of knowl-
proposals for extending the ontology description edge representation systems based on
language OWL. Descriptions Logics, describing the basic
The chapter is structured as follows. We first functionality of a DL system, surveying
present some background information on semantic the most influential knowledge representa-
web languages and related formalisms that are the tion systems based on descriptions, and
basis for the logical languages used in the different addressing specialized implementation
approaches discussed later in the chapter. We also techniques;
provide a brief introduction to fuzzy set theory • Addressing the use of description logics and
which the fuzzy description logics are based on. of DL-based systems in the design of several
In the mainly part of this chapter, we survey the applications of practical interest.
different approaches to extend the description
logics to represent the imprecise and uncertainty In the following statements, we mainly focus
information according their expressive power. on the first category, especially the theoretical
We also discuss proposals for fuzzy languages formalism of the description logics, with respect
OWL for the semantic web. Finally, we conclude to the balance between its expressive power and
with a critical review of the state of the art and an its computational complexity. Indeed, subsequent
analysis of directions for future research. results on the tradeoff between the expressiveness
of a DL language and the complexity of reasoning
with it, and more generally, the identification of
the sources of complexity in DL systems, showed


 A Review of Fuzzy Models for the Semantic Web

that a careful selection of language constructs was language) has been introduced as a minimal
needed and that the reasoning services provided language that is of practical interest. The other
by the system are deeply influenced by the set languages of this family are extensions of AL.
of constructs provided to the user. We can thus
characterize three different approaches to the The Basic Description Logic AL
implementation of reasoning services. The first
can be referred to as limited + complete, and Concept descriptions in AL are formed according
includes systems that are designed by restricting to the following syntax rule:
the set of constructs in such a way that subsump-
tion would be computed efficiently, possibly in C, D ::= T| (universal concept)
polynomial time. The second approach can be ⊥| (bottom concept)
denoted as expressive + incomplete, since the A| (atomic concept)
idea is to provide both an expressive language CD| (intersection)
and efficient reasoning. The drawback is, how- ∀R . C (value restriction)
ever, that reasoning algorithms turn out to be ∃R.⊥ | (limited existential quantification)
incomplete in these systems. After some of the
sources of incompleteness were discovered, often In AL, negation can only be applied to atomic
by identifying the constructs—or, more precisely, concepts, and only the top concept is allowed in
combinations of constructs—that would require the scope of an existential quantification over
an exponential algorithm to preserve the com- a role. In order to define a formal semantics of
pleteness of reasoning, systems with complete AL-concepts, we consider interpretations I that
reasoning algorithms were designed. Systems of consist of a non-empty set ∆I (the domain of the
this sort are therefore characterized as expressive interpretation) and an interpretation function •I,
+ complete; they were not as efficient as those which assigns to every atomic concept A a set AI
following the other approaches, but they provided ⊆∆I and to every atomic role R a binary relation R I
a test bed for the implementation of reasoning ⊆∆I×∆I . The interpretation function is extended
techniques developed in the theoretical inves- to concept descriptions by the following induc-
tigations, and they played an important role in tive definitions:
stimulating comparison and benchmarking with
other systems. TI = ∆I
Now, we survey the languages of the descrip- ⊥I = Φ
tion logics according to their expressive power (¬A)I = ∆I \ AI
with the beginning of AL. Elementary descrip- (C  D)I= CI ∩ DI
tions are atomic concepts and atomic roles (also (∀R .C)I = {a ∈ ∆I | ∀b. (a, b) ∈ R I → b ∈ C I}
called concept names and role names). Complex (∃R .□)I = {a ∈ ∆I | ∃b. (a, b) ∈ R I}
descriptions can be built from them inductively
with concept constructors and role constructors. We say that two concepts C, D are equivalent,
In abstract notation, we use the letters A and B and write C ≡ D, if CI =DI for all interpretations
for atomic concepts, the letter R for atomic roles, I.
and the letters C and D for concept descriptions.
Description languages are distinguished by the The Family of AL-Languages
constructors they provide. In the sequel we shall
discuss various languages from the family of We obtain more expressive languages if we add
AL-languages. The language AL (= attributive further constructors to AL. The union of concepts


Another Random Scribd Document
with Unrelated Content
which considered tenantry and wage-questions; and it is becoming
more common day by day. Within sixteen years twenty-one States
and the United States have passed more or less effective measures
looking to its use in labor disputes. Political coercion is also a matter
that has won a large share of legislative attention; twenty-nine
States and two Territories have enacted laws regarding it. There is,
however, an important distinction to be made. In an ordinary conflict
of political issues, when the magnates and their retainers are to be
found in both parties, it is obvious confusion and the unsettling of
political conditions for the employers to dictate how their workmen
shall vote. But when political issues suggest a class conflict, as in
1896, some of the provisions of these laws are by common consent
waived. The humble toiler may vote as he likes on the immaterial
questions of ordinary campaigns; but on questions having to do with
the salvation of society and the preservation of the hallowed code of
“business,” instruction and even gentle pressure become the solemn
duty of his social betters. There are fewer laws, it may be observed,
regarding another kind of coercion. Discharges on account of
membership in a labor union are forbidden in but fifteen states; and
in two of these (Illinois and Missouri) such provisions have been
found, after much painstaking study, to be unconstitutional. The
discovery is considered a most happy one; and according to the
injunction of the Federal Constitution, that “full faith and credit shall
be given in each State to the public acts, records, and judicial
proceedings of every other State,” the ruling will no doubt be found
applicable in a number of the other commonwealths.

IV
Our lawmakers are not to be blamed for decisions of
unconstitutionality. Rather, they are to be congratulated. For the
recent tendency of the judges to determine for themselves what
shall be enacted into law has developed new refuges for the
lawmakers. We have now Solon, the legislator, and Rhadamanthus,
the judge, in new rôles—the rôles of the good and bad partner of
Dickens’s novel. To the humble voter, when the pressure from below
conflicts with the pressure from above, Solon is now able to stand as
the supporter of popular measures, and to throw upon the less
responsible Rhadamanthus the onus of declaring them bad law. The
fury of the magnate at Solon’s demagogy is mitigated, if not
extinguished, when he considers the difficulties of the lawmaker’s
position, and especially by the further consideration that
Rhadamanthus has the final word to say. Solon has other refuges, it
is true; and sometimes these must be availed of, for it is not always
certain that a projected popular measure can be declared
unconstitutional. For several years it had been considered possible,
for instance, that an employers’ liability act, if passed in New York,
would stand the test of the courts. It became the custom, therefore,
when an adequate measure on this subject was introduced, for the
adverse interests to introduce a conflicting bill. The ingenious
lawmaker thereupon regretfully found a divided public sentiment,
and as a consequence no bill was passed. There are no reasons at
hand for accounting for the fact that at the last session of the Albany
legislature such a measure was actually enacted.

V
How far our legislators are enabled to withstand public sentiment, no
matter how strongly based in reason and how definite in objective,
may be instanced in the attitude of Congress regarding the Safety-
appliance act of 1893. Agitation for this measure had grown to such
an extent that action could no longer be delayed. But though action
on the bill could not be delayed, the terms of fulfilment of the bill
could be postponed to a comparatively remote period. The number
of railway employees killed in the year ended June 30, 1893, was
2727, a number exceeding the Union death roll in every battle of the
Civil War except Gettysburg, and within 243 of that record. In the
same year the number of wounded (31,729) was more than three
times as great as the number of Union wounded at either Antietam
or Chancellorsville, and more than double that at Gettysburg. Yet
despite this tremendous carnage, the legislators, wavering between
the public demands and the demands of the magnates, though they
passed the bill, generously granted five years for its complete
observance, and then gave the Interstate Commerce Commission
the power to grant further delays—in effect giving seven years for its
fulfilment. In those seven years 13,906 employees were killed—a
loss exceeding the Union death roll at Gettysburg, Spottsylvania, the
Wilderness, Antietam, Chancellorsville and Chickamauga combined—
and approximately 220,000 were wounded, or more than three times
the number of Union wounded in those six battles. That a great part
of this casualty record was avoidable is evidenced in the August
report of the Interstate Commerce Commission, which shows that
the number of employees killed in coupling accidents in the year
ended June 30, 1901, declined from 282 to 198, and the number
injured from 5229 to 2768. It was in 1893 that this generous latitude
was granted the magnates. Were the occasion to arise now, it is
probable that the term of grace would number fourteen years
instead of seven.
 CHAPTER VI
Our Interpreters of Law

The attitude of the judiciary in matters involving class antagonisms is

a subject upon which only the most restrained language is tolerable.
Even general inferences which suggest such a thing as judicial bias
must be avoided. Faith in the rectitude and wisdom of our judges is
a virtue sedulously preached,—perhaps most insistently by those
who do most toward their corruption,—and though the virtue as we
know it is rather vocal than immanent, it is sufficiently deep-seated
to be intolerant of spoken heresy. Were it openly questioned by any
considerable body of citizens, the foolhardy persons would soon
bring down upon themselves the rallying onslaught of those
heterogeneous elements which Karl Marx somewhat extravagantly
pictured, “landlords and capitalists, stock-exchange wolves and
shopkeepers, protectionists and free-traders, government and
opposition, priests and freethinkers, young street-walkers and old
nuns—under the common cry for the salvation of property, religion,
the family, and society.” Such heretics might have all the certainty of
Paul, “that the law is good, if a man use it lawfully,” and yet it would
be a parlous thing to be openly sceptical of the assumption that it is
always lawfully used.
But at least one may, without attainder of anarchy, assemble and
classify certain instances, and point out their coincidences and their
contrarieties. There is, for example, a notable sameness in kind of
the laws which are declared unconstitutional. There is, to utter it
mildly, a vast preponderance in the number of injunctions against
striking, boycotting, and agitating over the number against locking-
out, blacklisting, and the employment of armed mercenaries. There
is a practical, though not an entire, unanimity against the awarding
of damages to injured employees, whether the decision be based on
common or statute law; and, finally, there is a considerable diversity
between the decisions usually rendered by judges elected for short
terms, and therefore directly responsible to the people, and those
rendered by the less responsible judges, elected for long terms or
appointed.

I
The legislative aspects of employers’ liability have already been
considered. Certain judicial aspects of the matter need also to be
touched upon. The question is one of grave social import. The
worker no longer owns his tools, but must use the machinery
provided for him. A certain element of danger inheres in the
operation of probably all machinery; but when old, defective, or with
its dangerous parts unguarded, injuries to its operatives are well-
nigh certain. Yet for such injuries, with their awful consequences to
the operative and his dependent ones, there is generally no redress,
except in a few States where statutes have fixed the matter of
liability in set terms which leave no room for judicial discretion.
Under the common law the workman is held to assume the risk
attending his employment. He is a free agent—so the legal fiction
runs—and if afraid of injury need not work. Common law also
presupposes the providing of a “reasonably safe” place and
“reasonably safe” machinery by the employer. It would be difficult to
determine, however, from the mass of decisions under the common
law, what is meant by “reasonably safe.” A Colorado lower court
gave damages to the mother of a miner killed by falling rock while
removing débris from one of the mines of the Moon-Anchor
Consolidated Gold Mines, Limited. The case came finally to the
United States Circuit Court of Appeals for the Eighth District, and the
judgment was reversed, Judges Sanborn and Adams concurring and
Judge Thayer dissenting. The work was admittedly hazardous; in the
opinion of Judge Thayer “the place was needlessly made unsafe by
the master’s negligence.” The concurring judges, however, decided
that the company’s negligence was not responsible, and that “the
deceased of his own free will determined to cope with these risks
and hazards.... In this, his own voluntary conduct, is found the
intervening, proximate, and responsible cause of his injury.” (111
Federal Reporter, 298.)
Even when the employer assures the workman of the safety of a
machine, the risk is still, according to many decisions, the
workman’s. The Circuit Court of Shiawassee County, Michigan,
refused to award damages to a workman for injuries sustained from
a defective machine which he was operating for his employer. The
case went to the Supreme Court on a writ of error, and on December
15, 1900, that court affirmed the previous judgment. It had been
shown that the plaintiff warned his employer of the danger of the
machine, and that the employer gave assurances to the contrary.
Nevertheless, in the words of Judge Moon (Moore?), “one cannot
continue to operate a machine which he knows is dangerous simply
upon the assurance of his employer that it is not, if he has just as
much knowledge of the danger arising from the operation of the
machine as the principal has [without assuming the risk].” (82 N. W.
Reporter, 1797.)
The decision, read by Judge McLennan, in the recent case of Rice vs.
the Eureka Paper Company (76 App. Div. 336) before the Fourth
Appellate Division of New York State, would seem to indicate that
the burden of risk is not to be shifted from the workman even when
his employer acknowledges a defect in machinery and promises to
remedy it. There is some doubt, however, if such a decision, though
valid in many States, will stand in the State where it was given; for
the Court of Appeals has several times decided that liability follows
from an acknowledgment of defective machinery. On the other hand,
this highest court of New York State has won the distinction of
carrying the doctrine of assumption of risk to an extreme degree.
The case of Gabrielson vs. Waydell (135 N. Y. 1) involved the
question of the liability of the owners of a maritime vessel for
injuries suffered by a sailor in their employ. The captain of the vessel
had committed a confessedly unprovoked and particularly brutal
assault upon the sailor, who had subsequently sued the owner for
damages. The court decided that the sailor had no redress; that “the
misconduct of the captain was a risk assumed by the seaman, for
the consequences of which the owners are not responsible.”
A fact more curious yet to the unlegal mind is the judicial contention,
instanced in the previous chapter, that statutory provisions for the
safeguarding of machinery may be waived by the workman.
Evidently his burden of risk, like the Hindu’s caste, is born with him,
and cannot be laid aside or escaped. The case of the E. S. Higgins
Carpet Company vs. O’Keefe (79 Federal Reporter, 900) is an
illustration. Damages for an injury received from an unguarded
machine had been given a fifteen-year-old boy in the United States
Circuit Court for the Southern District of New York. The United
States Circuit Court of Appeals for the Second Circuit, however,
reversed the judgment. The plaintiff was a minor, but this fact was
held to have no bearing. “We think the circumstance that he was a
minor of no importance,” read the decision of Judge Wallace. “The
rules which govern actions for negligence in the case of children of
tender years do not apply to minors who have attained years of
discretion.” The New York factory act required guards for this
particular kind of machine. But that, also, was immaterial. “The
provisions of the statute ... requiring cogs to be properly guarded,
have no application to the case, except as regards the question of
the negligence of the defendant. As construed by the highest courts
of the State, the statute does not impose any liability upon an
employer for injuries received by a minor in his service in
consequence of the fault of the employee, or arising from the
obvious risks of the service he has undertaken to perform.” To clinch
the matter, Judge Wallace cited the then recent case of Graves vs.
Brewer before the Fourth Appellate Division of New York State,
wherein the court held that “the liability of the employer was not
changed by reason of the factory act requiring cog-wheels to be
covered, because such protection could be waived and was waived
by a person accepting employment upon the machine with the cogs
in an unguarded condition, as the danger was apparent, and one of
the obvious risks of the employment.” The case of Knisley vs. Pratt
(148 N. Y. 372) before the New York Court of Appeals was decided in
the same way, and also the case of White vs. Witteman Lithographic
Company. In the latter case the plaintiff was a child of fourteen.
Such decisions are common in more States than one. Another case
which may prove of some interest to the lay mind is that of Gillen vs.
the Patten and Sherman Railroad Company (44 Atlantic Reporter,
361). The plaintiff, while uncoupling cars, had his foot crushed in an
unfilled frog, and had been awarded damages. A motion for a new
trial was argued before the Supreme Judicial Court of Maine, and
was granted. The decision, delivered by Judge Lucilius A. Emery,
acknowledged the existence of a statute (chapter 216 of 1889)
requiring the filling or blocking of guard rails or frogs on all railways
before January 1, 1890. It held, however, that such filling and
blocking was not immediately mandatory upon a railroad constructed
after that date. “Such company is entitled to a reasonable time for
compliance with that statute.” It was at a crossing of such a railway
that the trainman lost his foot. He had no right to assume that the
rails were blocked, merely because a statute said they should be.
The brakeman, therefore, assumed the risk, and he also furnished
contributory negligence, since “to move about over frogs and
switches while coupling and uncoupling cars, even in moving trains,
without taking any thought of the frogs and guard rails, or as to
where he may be stepping, is negligence on his part contributing to
the catching his foot in them.”
When the doctrine of assumption of risk is inapplicable, when
personal negligence cannot be shown, and when there has been no
waiving of statutory provisions by the workman, there is yet, in
judicial eyes, one last resort for the defendant company—the
common-law plea of negligence on the part of a fellow-workman.
There is some diversity of opinion among eminent judges as to who
are strictly fellow-servants. “The courts of the majority of the States
hold, however,” writes Mr. Stephen D. Fessenden, in the Bulletin of
the Department of Labor for November, 1900, “that the mere
difference in grades of employment, or in authority, with respect to
each other, does not remove them from the class of fellow-servants
as regards the liability of the employer for injuries to the one caused
by the negligence of the other.” Thus it has happened that a
workman acting in the capacity of agent for his employer, and
ordering other workmen to do tasks at which injuries have resulted,
has been held to be a fellow-servant—a judgment relieving his
employer of liability. To the lay mind it would seem that workmen in
different departments could hardly be classed as fellow-servants;
and the United States Supreme Court has rendered a decision which
makes possible, under certain circumstances, such a discrimination.
Since then, however, the Federal courts have suffered a reaction on
the question, and current decisions tend the other way.
A case before a State tribunal—the Supreme Court of Georgia (35
Southeastern Reporter, 365)—illustrates the possibilities which lie in
this doctrine. A lineman, while repairing a wire for the Brush Electric
Light and Power Company, at Savannah, Ga., was killed through the
act of the engineer in turning on the current. The city court of
Savannah gave damages to his widow. The case was taken to the
State Supreme Court, and decision rendered March 3, 1900. The
counsel for the plaintiff contended that the fellow-servant doctrine
could not apply, on account of the lineman and engineer working in
different departments, “so that there was no opportunity for the
exertion of a mutual influence upon each other’s carefulness.” The
court, however, reversed the verdict.
The disparity of opinion between inferior judges and superior judges
in cases of this kind is remarkable. The monthly Bulletins of the
Department of Labor give a fairly excellent summary of court
decisions on labor questions. He who reads them will find the
expression, “judgment of the lower court reversed,” recurring with a
rather painful iteration; unless, indeed, the decision of the lower
court has rebuked the plaintiff, when the expression, “judgment of
the lower court affirmed,” is usually found. Mr. George W. Alger, in an
article on “The Courts and Factory Legislation,” in the American
Journal of Sociology for November, 1900, gives the following careful
and temperately worded summary of recent reversals in employers’
liability cases in New York State:—
“The percentage of reversals on appeal in master-and-servant cases
of this kind, when the verdict of the juries in the courts below had
been in plaintiff’s favor, is perhaps larger than in any other branch of
litigation. In New York, for example, an examination of twenty
volumes of the Court of Appeals reports (126 N. Y.-156 N. Y.) shows
written opinions in thirty-seven such cases. Of these: (1) in three
cases the juries in the lower court had found for defendant, and
plaintiff was the appellant; (2) in four cases the court below had
dismissed plaintiff’s case as insufficient, without requiring defendant
to introduce any testimony; (3) in thirty cases the juries below had
found for plaintiff with substantial damages. The Court of Appeals in
class (1) affirmed all of the cases where plaintiff was defeated
below. In class (2) it reversed the four cases where plaintiff had
been summarily non-suited and sent the cases back to trial courts to
hear defendant’s testimony: a partial victory at most for plaintiff. In
class (3), where plaintiff had actually received a verdict, of the thirty
cases twenty-eight were reversed. These statistics are interesting as
showing how complete is the lack of harmony between the courts, at
least in New York, and the moral sense of the people by whom the
courts were created, in regard to these cases. Twice in thirty times
do the opinions of the learned judges of New York’s highest court
coincide with the opinions of juries of citizens as to the requirements
of justice.”
The tendency, which is most clearly indicated by the mass of
decisions in cases demanding damages for injuries or death, is the
growing disposition to make property paramount and life
subordinate. It is a common practice to set aside verdicts of
damages on the score that they are excessive. It is no less a
common practice to instruct the jury to decide for the defendant in
order to rebuke litigation. The language of the leading work on one
phase of this subject—Shearman and Redfield’s “A Treatise on the
Law of Negligence”—sums up the matter in a few words:—
“It has become quite common for judges to state as the ground of
decisions the necessity of restricting litigation. Reduced to plain
English, this means the necessity of compelling the great majority of
men and women to submit to injustice in order to relieve judges
from the labor of awarding justice.... The stubborn resistance of
business corporations, common carriers, and mill-owners, to the
enforcement of the most moderate laws for the protection of human
beings from injury, and their utter failure to provide such protection
of their own accord, ought to satisfy any impartial judge that true
justice demands a constant expansion of the law in the direction of
increased responsibility for negligence.”

II
“Law,” wrote Sir Edward Coke, “is the perfection of reason.” This
may be true; but, if so, it tends to throw mankind over to the
position of the Catholics, that the reason itself needs considerable
perfecting. This is not only the disposition of the lay mind, but,
evidently, also of the supreme judicial mind; for a large part of the
higher judicial activity during recent years has been expended in
declaring null and void laws passed by two houses of the people’s
representatives and signed by an elected Governor or President. Mr.
Stimson, in his summary of labor legislation for the years 1887-97,
found that only 114 out of the 1639 laws passed had been declared
unconstitutional. But these 114 comprised examples from 19 out of
the 35 classes of legislation passed, and must therefore have
reacted upon a very considerable number of the remainder. It is a
coincidence which has been noted before, and need not be specially
insisted upon here, that the overwhelming majority of laws which fail
to reach the constitutional standards set by our judges are those
intended to safeguard the interests of the industrially subordinate
and to set some limitation to the powers of the industrially mighty.
The judicial mind, however, affects to know no difference between
high and low, between weak and strong; and thus its decisions,
ignoring actual conditions, tend more and more to strengthen the
powers of one class and to weaken the powers of another. “Liberty”
is the shibboleth; the citizen must be free to act as he wills.
Somewhat curiously, though, liberty of speech, press, and
assemblage is not so strenuously insisted upon; and, indeed, by
injunctions and other judicial determinations is at times rather
severely limited: the miners of West Virginia have been recently
enjoined from holding meetings on their own grounds. But economic
liberty—the liberty of the dependent classes to do acts which, in the
nature of things, they cannot possibly do—is held for a sacred
principle. The doctrine of the extension of the State’s police power,
limiting the foregoing doctrine, has gained some headway since the
Utah decision confirmed a State’s right to limit the hours of work for
men in dangerous trades; but the determination of how far it is to
be applied rests largely with the forty-eight State and Territorial
courts; and it is a safe guess that it will meet with stiff resistance if
incarnated in further “advanced” legislation.
“No discrimination,” which in effect means much discrimination,
follows the judicial shibboleth of “liberty.” Especially zealous for the
protection of liberty and keenly watchful of proposed discrimination
is that eminent tribunal, the Supreme Court of Illinois. Some six
years ago it discovered that the statute regulating the hours of
women workers in the factories contravened the Federal and State
constitutional guarantees of “life, liberty, and property.” A woman’s
labor was her property, and any limitation of it was a deprivation
“without due process of law.” On December 20, 1900, it fell to the
lot of this tribunal to pass upon two labor laws,—to the lay mind
entirely different in principle,—and, by a somewhat difficult
struggling along parallel lines of argument, triumphantly to reach
conclusions adverse to both of them. One was the Chicago
ordinance requiring union labor and an eight-hour day on all public
work contracted for; the other the State statute prohibiting
discharge of an employee for belonging to a labor union. Regarding
the ordinance, the union requirement, in the words of Associate
Justice Magruder, “amounts to a discrimination between different
classes of citizens.” It is therefore void, and the eight-hour provision
is also void, because it “infringes upon the freedom of contract, to
which every citizen is entitled under the law.... Any statute providing
that the employer and laborer may not agree with each other as to
what time shall constitute a day’s work is an invalid act.” (58
Northeastern Reporter, 985.)
Without venturing to discuss this ruling, one may at least compare it
with the ruling on the State statute. The latter was a law intended to
prevent discrimination against union men. But, curiously to the
unlegal mind, it is discovered to be discrimination in favor of the
union man. “The act certainly does grant to that class of laborers
who belong to union labor organizations a special privilege.” (58
Northeastern Reporter, 1007.) The act was also found to “contravene
those provisions of the State and Federal constitutions which
guarantee that no person shall be deprived of ‘life, liberty, or
property without due process of law.’” “That strain again,” as Orsino,
in “Twelfth Night,” exclaims. It has not, however, a “dying fall,” for it
has been taken up and echoed in other quarters since.
The liberty of the employer to pay his employees in brass checks or
store orders was affirmed by the Kansas Supreme Court on
December 9, 1896, and the act requiring payment in lawful money
was declared invalid. “To say that a free citizen can contract for or
agree to receive in return for his labor one kind of property only, and
that which represents the smallest part of the aggregate wealth of
the country, is a clear restriction of the right to bargain and trade, a
suppression of individual effort, a denial of inalienable rights.” Anti-
truck acts were also declared unconstitutional by the courts of
Pennsylvania, Ohio, Illinois, and West Virginia. The Kentucky
Supreme Court, however, nine months after the Kansas decision,
found that liberty and the compulsory payment of wages in lawful
money were compatible, so that the question is at least open.
Decisions like that of the Kansas court, and the somewhat similar
decisions rendered in Pennsylvania, Illinois, and Tennessee, of
course fasten the laborer to the company store; but of this the
courts usually take no cognizance. Actual liberty may be restrained,
but theoretical liberty must not be tampered with.
Weekly payment laws are found to conflict with liberty in
Pennsylvania, Illinois, Missouri, West Virginia, and Indiana. Moreover,
the liberty of a legislature to determine that prevailing wages shall
be paid to employees of city and State must not be confused by the
lay mind with the liberty of the wage-earner to work under what
conditions he must. For the former is clearly unconstitutional, as
decided in New York by the Court of Appeals in February, 1901. “The
effect of this statute [the Prevailing Rate of Wages act],” reads the
decision of Judge Denis O’Brien, “was to make the city [of New York]
a trustee or instrument for the enforcement of the law in the
interests of the persons for whose benefit it was enacted, and thus
the powers and functions of the municipality are employed for
purposes foreign to those for which they were created and exist
under the Constitution.” The eight-hour laws passed in several of the
States have generally suffered the Illinois fate, although Kansas
proved an exception. Regulation of the working hours of women was
nullified not only in Illinois, but in Nebraska and California. The
police-power doctrine, as voiced in the Utah decision, may justify a
limitation of the working day in dangerous trades, but otherwise
such a limitation appears to be an infringement of the right of
contract, or a deprivation of “property” without “due process of law.”
Even the National Eight-hour law of 1868, while not strictly
unconstitutional, is held to be merely advisory. “We regard the
statute,” says the Supreme Court (94 U. S. 404), “chiefly as in the
nature of a direction from the principal to his agent that eight hours
is deemed to be a proper length of time for a day’s labor, and that
his contract shall be based upon that theory.”
Anti-trust laws may be quite as lacking in constitutional decorum as
are eight-hour and prevailing-wages laws; and the judiciary reserves
to itself the right to determine what are the standards. The Texas
Anti-trust law of 1889, for instance, overleapt judicial sanction. “It is
not every restriction of competition or trade,” reads the decision of
District Judge Charles Swayne (February 22, 1897), “that is illegal or
against public policy, or that will justify police regulation, but only
such as are unwarrantable or oppressive; and a State statute which
prohibits combinations formed for the purpose of reasonably
restricting competition violates the rights of contracts guaranteed by
the Federal Constitution.” (79 Federal Reporter, 627.) Another
legislature, with this lesson before it, will know better where to set
bounds to its attempt at interference.
One cannot pass this phase of the general subject without recurring
to the pertinent advice of the wise Sir Francis Bacon. “Judges,” he
wrote in his essay, “Of Judicature,” “ought to remember that their
office is jus dicere, and not jus dare, to interpret law, and not to
make law.... Judges ought to be more learned than witty, more
reverend than plausible, and more advised than confident.... A judge
ought to prepare his way to a just sentence, as God useth to
prepare his way, by raising valleys and taking down hills; so when
there appeareth on either side a high hand, ... cunning advantages
taken, combination, power, great counsel, then is the virtue of a
judge seen to make inequality equal; that he may paint his judgment
as upon an even ground.” Wise counsel! though it seems to have
lacked something in observance two hundred and seventy-five years
ago, and may be suspected, even yet, of not always and everywhere
reaching entire fulfilment.

III
We have the testimony of no less eminent an authority than United
States District Judge John J. Jackson, of the Northern District of
West Virginia, that in all his experience on the bench he could not
recall a single occasion when any court, either Federal or State, ever
abused the writ of injunction in strike questions. It is a definite and
authoritative pronouncement; and the restrained and careful
language accompanying it, wherein the officials of labor unions are
described as “a professional set of agitators,” and “vampires that
fatten on the honest labor of the coal miners,” certainly proves that
it cannot be an ex parte statement. Yet, for all that, there is a widely
diffused sentiment that the writ of injunction has occasionally been
abused in strike questions. In the same locality, at about the same
time, an injunction issued by United States District Judge B. F. Keller,
of the Southern District of West Virginia, declared, among a
multitude of other prohibitions, that the strikers “are further
inhibited, enjoined, and restrained from assembling in camp or
otherwise,” even on grounds leased by them for their meetings.
A pamphlet, prepared by five members of the New York Bar and
issued by the Social Reform Club, of New York City, in the summer of
1900, gives the substance of a number of injunctions that have been
issued against striking workmen. “In the case of the Sun Printing
and Publishing Company vs. Delaney and others in December
(1899),” says the pamphlet:—
“The Supreme Court of New York, among other things, enjoined the
defendants from the exercise of their right to give the public their
side of the controversy with the Sun as an argument against
advertising in a paper which they claimed had treated them unjustly;
it also forbade them from attempting to persuade newsdealers from
selling the paper; and finally wound up with a sweeping restraint
‘from in any other manner or by any other means interfering with
the property, property rights, or business of the plaintiff.’ It should be
added that, on appeal, the Appellate Division struck out these
commands; but they were so plainly subversive of fundamental
rights that it is difficult to see how they could have been granted in
the first instance.
“In still another case last year—The Wheeling Railway Company vs.
John Smith and others (so runs the title of the action without
naming the others)—in the United States Circuit Court, West
Virginia, two men not parties to the action, nor found to be agents
of ‘John Smith and others,’ whoever they may have been, were
punished for contempt of court, for, among other things, ‘reviling’
and ‘cursing’ the court? not at all, but for ‘reviling’ and ‘cursing’
employees of the railroad company. If these men had not actually
served out an imprisonment in jail for thirty days as a punishment
for contempt of corporation, it might be thought that your
committee had taken this example from opera bouffe. The legality of
this punishment was never passed on by the Supreme Court, for the
reason, as your committee understand, that the parties were unable
to bear the expense of taking it there, and so served their term in
jail.
“During the final drafting of our report a temporary injunction has
been granted by a Justice of the Supreme Court in New York City....
This injunction forbids the defendants [certain members of the Cigar
Makers’ International Union] even from approaching their former
employers for the laudable purpose of reaching an amicable result; it
forbids them from making their case known to the public if the
tendency of that is to vex the plaintiffs or make them uneasy; it
forbids them from trying in a perfectly peaceable way in any place in
the city, even in the privacy of a man’s own home, to persuade a
new employee that justice is on their side, and that he ought to
sympathize with them sufficiently not to work for unjust employers;
and, finally, it forbids the union from paying money to the strikers to
support their families during the strike.”
Such instances, as the pamphlet states, can be multiplied. Perhaps
they do not wholly controvert Judge Jackson’s declaration. But, at
least, they illustrate an unbridgeable disparity between the
definitions of justice held on the one hand by our interpreters of law,
and on the other by the overwhelming majority of the citizenship.
That disparity has been great in all recent times; but weekly and
daily it grows greater. The stronger inclination of the judiciary to
make property the paramount interest is everywhere observed; and
the magnates, with an exultant recognition of the fact, make haste
to enjoy the fruits of the new dispensation.

IV
From judgeship to attorneyship of a great corporation has recently
become a common promotion. The number of ex-judges who have
been thus translated to higher sees is notable: one finds or hears of
them in many places. Republics may be ungrateful, as the adage
runs, but not so the magnates. The gratitude of the latter may not
be wholly platonic; it includes, no doubt, a lively sense of favors to
come. But whether prospective or retrospective, it expresses itself in
deeds of recompense, and that is the main test. It is a discriminating
gratitude, moreover. Keenly enough, it recognizes the comparative
value of service. Other servitors of the magnates may toil faithfully,
and receive but moderate reward. The moulders of opinion, such, for
instance, as the newspaper men, may ask for preferment, and be
met by the impatient retort of Richard III to Buckingham, “I am not
in the giving vein to-day.” But for one who can interpret the law as it
should be interpreted, there are glory and riches to be had for the
asking.
 CHAPTER VII
Our Moulders of Opinion

“There never was a time,” says Justice Brewer, in the concluding

lecture of a series recently delivered by him at Yale University, “when
public opinion was more potent.” Possibly the saying is true; but
whatever force it may have lies in the application. Public opinion may
make for a general passivity—an acquiescence in things as they are
—quite as much as for a general strenuousness. Nowhere, for
instance, among civilized peoples, is public opinion more powerful
than in a quiet and isolated community, held fast to certain habitual
modes of speech and action. Only a brave man, or a desperate
woman, so environed, would dare defy the tribal customs.
Public opinion in these United States may be more potent than ever
before, but the personal attitude which it supports and encourages
becomes more and more one of acquiescence in the existing régime.
A legislative reaction and a judicial reaction are manifested; and a
growing irritation is expressed, as from time to time those rude
disturbers of the public peace, the social reformers, come forward
with plans for curing imputed evils. Social and political quietism
becomes our everyday philosophy. An “air of contentment and
enthusiastic cheerfulness ... characterizes our society,” writes
Professor William G. Sumner, of Yale, in a recent number of the
Independent; and though the judgment might be somewhat more
accurately worded, he is not far wrong. A keen-eyed observer from
Italy,—Professor Angelo Mosso, of Turin,—who visited us a few years
ago, gives somewhat similar testimony. The fact astonishes him, as
he confesses, since he saw much of political and industrial evil which
he could not comprehend a democracy enduring; yet for all that the
evidence was convincing.
 I
Among the causes making for this acquiescence in existing social
conditions, there are three which may be considered here. The first
is the one which so strongly impressed Professor Mosso. It is the
rage for individual exploitation. The imaginations of most men are
fired by the spectacle of the few achieving great fortunes; each
believes that a like fortune lies somewhere within his own reach, and
with blind fatuity he tolerates conditions which he instinctively feels
to be inequitable, simply because he expects himself to master
them. “I believe,” writes Professor Mosso, “that the desire to become
wealthy is so strong and powerful in every American that, in order to
reserve the opportunity of realizing such desire, Americans willingly
submit to the continuance of laws which allow such accumulations.”
It is the petty gambler’s faith, the conviction that, though everything
be against him, he will somehow “beat the game.” And just as the
petty gambler’s faith is fostered by the runners and “cappers” for
faro, policy, roulette, and keno, so the faith of the industrial
underling is fostered by a tremendous trumpeting of the ways and
means to worldly “success.” The preaching of “success” has become,
in these last five years, a distinct profession, honored and well
recompensed.
A second cause of the prevailing acquiescence in the present régime
applies more particularly to social reformers, and to those who, while
not actively enlisted as “come-outers,” do yet sympathize with the
activities of their more aggressive brethren. It is a feeling, born of
years of experience in promoting some collective good, of the
hopelessness of achievement. Opposed at all points, frustrated at
many, there comes a time, sooner or later, when all but the most
resolute reformers are forced to admit that little or nothing can be
done. Many thereupon fall back into the ranks of the do-nothings
and the care-nothings; while others, in whom the fire of purpose is
not entirely quenched, reluctantly exchange their radical and
comprehensive plans of social changes for more narrow and
immediate purposes,—the giving of small charities, the doing of
near-at-hand services, and the occasional support of a particular
public measure.

II
A third, and perhaps the most important, cause is the continual
output from pulpit, sanctum, forum, and college chair, of our
professional moulders of opinion. Now not all of this output, it is
freely conceded, makes for acquiescence; but the overwhelming
mass of it unquestionably does. From these instructors of the people
we learn that conditions, while not perfect, either are reasonably
near to perfection, or, if evil, are not to be corrected except by
individual regeneration. We learn of the irrationality or the moral
obliquity of discontent; the viciousness or fanaticism of impertinent
persons who seek to change things; the virtues of obedience; the
obligation of toil (specifically directed to those who are doing most
of the world’s work, for the profit of others), and of the worth,
benevolence, and indispensability of our magnates.
The denunciation of discontent becomes more common and more
emphatic. A plentiful crop of instances is always forthcoming to any
one who cares to look for them. The generation of Rousseau and the
following generation of Jefferson set high hopes for mankind on the
faculty of discontent. The past generation, compromising between
theology and evolution, found in discontent a perpetual factor
making for the creation of a better environment. But our present
reaction takes us back to the days of the Stuarts. The magnificent
invectives of Dryden, voiced in that—

“full resounding line,

The long majestic march and energy divine,”

against the sedition and discontent frequently manifested during the

reign of Charles II, might serve for a thousand texts for present-day
sermons, lectures, and editorials. The thought, common these last
hundred years, that discontent is usually the result of privation,
wrong, or oppression, is given over; and our modern moulders of
opinion revert to the notion that it is fostered by ease and comfort.

“To what would he on quail and pheasant swell

That even on tripe and carrion could rebel?”

asks “Glorious John” in satirizing his rival Shadwell. Tripe and carrion
did not form the usual nourishment for rebellion. We find the same
idea constantly echoed in very recent days; and the demands of
organized workmen for better pay are almost invariably regarded in
certain intellectual circles as evidences, not of need, but of the pride
and rebelliousness engendered by an already attained competency.
Honors are even between churchmen and lay publicists, when it
comes to the denunciation of discontent. The pulpit, the stump, the
college chair, and the editorial sanctum are alike busied with its
condemnation. Perhaps a typical protagonist in the work was the
late E. L. Godkin. The thought recurs again and again in his writings.
“I must frankly say,” he avers in his essay, “Social Classes in the
Republic,” “that I know of no more mischievous person than the man
who, in free America, seeks to spread among them [the workers]
the idea that they are wronged and kept down by somebody; that
somebody is to blame because they are not better lodged, better
dressed, better educated, and have not easier access to balls,
concerts, or dinner parties.” Whereupon, to make clear his
contention, he tells of the following pathetic little episode:—
“Two years ago I was in one of the University Settlements in New
York, and was walking through the rooms of the society with one of
the members. They were plain and neat and suitable, and he
explained to me that the purpose in furnishing and fitting them up
was to show the workingmen the kind of rooms they ought to have
‘if justice were done.’ To tell this to a workingman, without telling
him in what the injustice consisted and who worked it if he had not
such rooms, was, I held, to be most mischievous.”
Even President Roosevelt, doubtless impressed by the modern
reiteration of the notion, felt called upon, in his Providence speech
(August 23d), to rebuke discontent, and incidentally to identify it
with envy. “Not only do the wicked flourish,” he says, “when the
times are such that most men flourish, but what is worse, the spirit
of envy and jealousy and hatred springs up in the breasts of those
who, though they may be doing fairly well themselves, yet see
others, who are no more deserving, doing far better.”
Education, in the modern view, is largely responsible for discontent,
and should be restricted. Judge Simeon A. Baldwin, of the
Connecticut Supreme Court, and lecturer in the Yale Law School, is
quite certain upon this point. His “signed editorial,” in the April 9th
issue of a New York newspaper published by the Yale lecturer on
journalism, expresses a view which is coming to be widely held. Our
young men, he notes with great complacency, are obliged to leave
school early, in order to go to work; and he thereupon urges that
young women also should clip their education at an early age. “Girls
would make better wives and mothers and housekeepers,” he writes,
“if they finished school at from fourteen to sixteen years of age. As it
is, they obtain a smattering of many studies, which in my opinion
cannot do them much good. They are possessed by a spirit of unrest
to-day, and develop ambitions not compatible with the happiest
homes.”
Professor Harry Thurston Peck expresses the modern view more
succinctly. Professor Peck, it may be stated for the benefit of the
unenlightened, is an instructor of Latin in Columbia University. No
pent-up Utica, however, contracts his powers; he has courageously
sallied forth from his particular domain and has taken all knowledge
for his province. Over this province he ranges with unconstrained
freedom, noting what he will, and, with something of the “large
utterance of the early gods,” making known to a waiting world his
impressions and beliefs. What a great lexicographer said of an
amiable poet may be repeated in present praise: He touches nothing
that he does not adorn. Some intellectual limitations it is possible he
may have; but as a reflector of certain current views obtaining in
Welcome to our website – the ideal destination for book lovers and
knowledge seekers. With a mission to inspire endlessly, we offer a
vast collection of books, ranging from classic literary works to
specialized publications, self-development books, and children's
literature. Each book is a new journey of discovery, expanding
knowledge and enriching the soul of the reade

Our website is not just a platform for buying books, but a bridge
connecting readers to the timeless values of culture and wisdom. With
an elegant, user-friendly interface and an intelligent search system,
we are committed to providing a quick and convenient shopping
experience. Additionally, our special promotions and home delivery
services ensure that you save time and fully enjoy the joy of reading.

Let us accompany you on the journey of exploring knowledge and

personal growth!

ebookultra.com