A WEB-BASED ANNOTATION SYSTEM FOR ...

A WEB-BASED ANNOTATION SYSTEM FOR IMPROVING COOPERATION IN A CARE NETWORK∗

MYRIAM LEWKOWICZ, GAËLLE LORTAL, AMALIA TODIRASCU, MANUEL ZACKLAD, MOHAMED-FOUED SRITI* University of Technology of Troyes / ISTIT lab/Tech-CICO team 12, rue Marie Curie . BP 2060 . 10010 Troyes Cedex - France E-mail : {surname.name}@utt.fr, *[email protected] Coming from the needs of a care network for a cooperating system enabling its members to share information and argue cases, this paper presents a work in progress which aim is to propose a Web-Based system using annotation for cooperation through an electronic patient le (EPF). We rst describe the EPF, and then we focus on our socio-semantic web positioning, on existing web standards for annotation, and on requirements for the EPF annotation tool. Finally, we present a distributed architecture for this web-based application. 1

Introduction

A distributed care network is an organizational device enabling a group of people to work together in order to improve care on a particular disease. The health network studied here was created in order to deal more eciently with the memory disorders of elderly people living in a restricted geographical area. It brought together various actors, such as health professionals (family doctors, neurologists, geriatricians, speech therapists, nurses, etc.) and social workers. It will be extended to include representatives of the patients, the "care-takers", namely close parents of the elderly who are actively involved in supporting their relatives, and who are in close contact with the other participants. In our case, generalist and specialist doctors are working together on cases of Alzheimer's disease during sometimes long-way o meetings. In order to complete these synchronous phases of cooperation by asynchronous phases, a technical device has been added to the organizational one : a web-based system playing the role of a boundary object 1,2,3 between all the members of the network. This web-based system is an Electronic Patient File (EPF) with dierent types of information about a case. In this context, the EPF helps participants to work together and share data, but participants expressed the need to enlarge the structured EPF, with less structured exchanges, in order to argue on patient cases, or on prescription for treatment, to clarify interpretations by explaining a context, or to interact in this care network for building new initiatives. Thus, analysing actors' exchanges while face-to-face meetings led to the denition of needs in annotating as : 1) Annotating data laid in the EPF to specify the context ∗ THIS WORK IS A PART OF A PROJECT SUPPORTED BY THE CNRS MULTIDISCIPLINARY PROGRAM TCAN

227

(kind of comment sometimes done by the author him/herself), or to ask for its justication or review (possibly based on other items of data from the EPF) 2) Annotating one or several pieces of information laid by a colleague to ask for details or question a result 3) Annotating a diagnosis to inquire about it or assess it (possibly based on other items of data from the EPF). In order to permit these discussions, or negotiations, and then improve the cooperation into the network, we propose to add to the EPF a web-based annotation system. This way, annotations not only help in structuring and improving information access as in classical Semantic Web approach, but also in structuring social exchanges in a Socio-semantic web orientation 4 . In this paper, we rst describe the existing EPF, and then we focus on our socio-semantic web positioning, on existing web standards and tools for annotation, and on requirements for the EPF annotation tool. Finally, we present a distributed architecture for this web-based application.

2

Electronic Patient File : A Web-Based System for Information Sharing in a Care Network

The EPF takes into consideration static medical observations and activities, as well as dynamic, structured, creative pieces of information. It is structured on the medical SOAP format 5 , which contains : Subjective data ; new Objective data ; Assessment of the new data and new Plans that are determined by the new data. Then our EPF contains Symptoms and clinical signs of a patient, Observations and analysis of doctors about a case, Assessment of the disease within a case, and Procedures or action plans for resolving health problems. The EPF is included in a public web site and is structured by 8 tabs pointing at dynamic page : Summary form (basic civil information, contact), Civil status record, Familial status, Social le, Medical le, Medical examination (Biological tests, Neurological tests, Cognitive tests). These data are shared by participants from specialized areas : medical (general practitioner, specialist...), social (social worker...), and administrative (secretary...). Each category is allowed to consult certain types of data and to act on certain types of data. As we already introduced, it appears that this EPF is not sucient for asynchronous phases of cooperation among doctors. Actually, they can just put some information on the patient and see information that the others have put. They cannot discuss on analysis results, on primary diagnosis, on decisions like putting the patient in hospital, leaving him/her at home... In order to permit these informal discussions through all the professionals of the care network, we propose to add a web-based annotation system to the EPF. 228

Fig.

3

1: Use cases in EPF

Proposition of a socio-semantic Web application for Annotation

3.1

Socio-Semantic Web approach

Socio-Semantic Web (2SW), as dened by Zacklad and Barbaud 4 , is an extension of a Cognitivo-Semantic Web proposed by Caussanel et al. 6 . The Cognitivo-Semantic Web aimed to add to Semantic Web researches the cognitive aspects of an user making a request on the web : initial design of the request, updating of the request according to the increase of knowledge, and evaluation of the results. In order to take into account these cognitive activities, we recommended the use of more simple knowledge representation languages, as Topic Maps 24 , rather than using formal logics. While extending this perspective, the Socio-Semantic Web (2SW) is dened beside a "Social Web" which corresponds to all the web applications dedicated to the increase of a mutual awareness in distant interactions (newsgroups, chat, instant mes229

senger...). The 2SW aims at supporting more structured design activities, in which interactions are based on shared documents or information by a group, pursuing common objectives, at least for a while. Considering these objectives, 2SW has to contribute to the building of a structured representation, both of the domain and of the group. Cooperation spaces proposed by the 2SW have then to supply communicational and documentary functions, near the ones of groupware. But, contrary to these applications which were based on proprietary environments, 2SW is following the Open philosophy of Semantic Web. This open and distributed nature of Semantic Web, enlarged both with Cognitivo-Semantic Web and Socio-Semantic Web, seems then to oer good design principles for our application. 3.2

Existing Web standards for Annotation

Our aim to allow an user to add some comments on a patient le questions us on the anchoring and the form of these meta-information that we want to add to the original le. This problematics is tackled in the Semantic Web eld of research, which goal is to enrich web resources with structured and descriptive information in order to improve accessibility, research and use of information. We are now going to describe existing tools in this eld, which we could reuse and enrich for our project. Semantic Web identies three annotation types : simple meta-data (modication date, author...), annotations which we will call computational because they address programs and they allow them to exploit resources in a better way 7,8,9 , and annotations which we will call cognitive because they address the reader, allowing him/her to be an active actor of the Web. This aim to enrich online documents, to go to a collaborative Web is not new. In fact, several tools have been developed since the early 90. These tools permit to review texts with comments and/or explanations, to justify decision... They are generally made of several elements which permit to visualize, create, store, and search annotations, each annotation being dened by an anchor, attributes, and a body. Annotations are stored on a dedicated server and can be sorted according to their attributes, be public, shared by a dened group, or private. These servers contain information on the localization (the document in which the annotation has been made, the place in the document), the style (police, color...), content (text and attributes), and the function of the annotations (link for example). Annotations are generally linked together as a tree, which facilitates their management and browsing. All these researches lead to dene the W3C standard Annotea 10,11 which is based on a RDF (Resource Description Framework 12 ) description of annotations and "enhances collaboration via shared metadata based Web annotations, bookmarks, and their combinations". Several annotations' servers (ZAnnot, Annotea) and clients (Annozilla, Amaya) implement the Annotea 230

standard. The "ZAnnot" 13 annotations' server stores the annotations in a RDF database, and the users can interact with this server via Annozilla client 14 to search an annotation, to create a new one, or to delete another. An annotation is described as a set of meta-data (its attributes), and a body. The meta-data are dened by a RDF schema and give information as the author of the annotation, its date of creation, its type (comment, question, correction, etc.), the original document, the part of document which is annotated. The advantage of this RDF notation is that it is possible to personalize it by adding for example some attributes and a set of values of these attributes to the schema of an annotation. It permits to t collaborative needs of the concerned team. This technical solution is then interesting for our project, because, as we will expose in Sec. 3.4, we wish to index the annotations according to several points of view. These points of view are dierent from existing Annotea attributes, and are related to a cognitive use of the annotations in the care network, contrary to other studies in Semantic Web which focus on computational use of annotations or attributes. We are now going to describe and classify existing Annotation tools, to clarify our positioning. 3.3

Existing Web tools for Annotation

Several annotation clients are available nowadays, coming from the Semantic Web initiative which we have described above. Most of them adopt what we have called a Computational Semantic Web approach aiming at indexing Web pages more or less automatically. They are used for meta-data creation and some of these are based on ontologies to help supporting these computational annotations (OntoMat-annotizer 15 , Melita 16 , MnM 17 ). Computational annotations are geographically bound to a part of a Web page, but do not enrich the page at a knowledge level, neither help cooperation nor interaction between readers of the page. They are meta-data indexing computationally a page, and they allow search engines to better retrieve information or pages. Some other clients adopt a more social approach, aiming at supporting communication without any plan on indexing or retrieving annotations, even if these annotations can be sorted by basic meta-data as date of creation, or author (Yawas 18 , CritLink 19 , XLibris 20 , etc.). In these annotation tools, we can consider textual annotations as comments, as found in some proprietary software or specic plug-in for software, where comments are not indexed nor dierentiated from the document (Windows Word comments 21 ). Textual annotations can sometimes be stored apart, in an annotations server (Acrobat PDF 22 ) and little organised. But these social annotations are not related to each other and so can not represent a discussion or exchanges between human users about a document. They then cannot support interaction more than computational annotations tools. Finally, we could roughly say that we are facing two annotation tools fa231

milies. The rst one is focusing on Web pages indexation for a better retrieval, and the second one is focusing on human communication through comments. In a collaborative environment, all these tools lack arguing/reviewing management, as well as cooperative work. Our proposal is then to enrich social annotation tools by semantic web techniques of indexation and to support user adding annotations to a document in order to help him/her to work in collaboration. Moreover, we will propose other new functions as multi-anchoring (binding several fragments) and replying to an annotation. We are now going to present our requirements for a Socio-Semantic webbased application supporting cooperation around the EPF. 3.4

Requirements for a Socio-Semantic web-based annotation tool

With Zacklad et al. 23 , we dene annotation as a located meta-data, bound to another document. It is related to several parameters as time, place, participant, public/private access, semantics... which means that an annotation is a multi-parts entity composed of an anchor(s) in document(s), attribute(s), and a body (the text of the annotation). We also consider annotation as a track of a collaboration process, which has two main functions : Planning (project management, micro-organization), and Reviewing (Argument, annotations as body of a written document). Meta-data proposed by existing web standards (as Annotea which we described above) for indexing annotations (Author name, Date, Subject, Type of the annotation...), are not sucient for our purpose and to take into account the role of annotation in a collaborative process. In fact, with this kind of index, we miss the organizational context (roles, prole of the participants, etc.), the domain context (professional keywords, domain-specic terms, concepts, etc.), and arguing types (proposition, opposition, etc.). In order to allow a more subtle information retrieval, we then propose to extend annotation indexing with regard to domain-specic point of view, but also to add social dimension of indexing : the arguing point of view (keeping a track of the decisions and negotiations between human participants), as well as the organizational point of view, using the actors' role to increase the importance of the decision. Using Ontologies for Indexing Annotations

Each of the three points of view dened above is described by an ontology. From a Semantic Web perspective, ontologies should represent exhaustively domain-specic knowledge structured as a hierarchy of concepts and relations between concepts. Each concept is well dened by its properties, so the expert should completely specify these relations. Generally speaking, ontologies' coherence and consistency should be computed automatically, by specic inference mechanisms. Building such ontologies is a dicult, time-consuming and expensive task. Human experts have conicting denitions of certain concepts and several denitions are possible. Computing on-

232

tology's coherence is also a time-consuming operation. Reusing existing ontologies is not always possible. On one hand, generic ontologies (EuroWordNet, DOLCE24 ) are not adapted to domain-specic applications ; they do not contain domain-specic concept denitions. And on the other hand, domain-specic ontologies are not available or they are very expensive, even if their portability is increased by the use of W3C standards (OWL, RDF). Building a representation of the semantic content of Web pages is still a dicult task, even when using ontologies. To avoid this drawback, the Socio-Semantic Web proposes the use of less formal ontology, which main purpose is to help the user navigating through Web pages and not to compute automatically the semantic representation of the document content. From this perspective, the concepts should be less-specied ; there is no need to identify all the concepts' properties. Standards as Topic Maps (TM) ISO norm 25 are dened for these semi-formal ontologies. TM formalism denes a network of topics covering domain-specic knowledge. Topics' denitions are dened via simple URL, so all the users share the same denition. The topics are hierarchically organised (related by "isa" relations) and associated by horizontal relations ("partOf", "used") (Fig. (2)). No coherence checking mechanism is used by Topic Maps. While TM do not require a precise denition of concepts, and are designed to support user browsing Web pages ; we adopted this formalism for representing the various point-of-view ontologies. In our system, the micro-organizational and argumentative ontologies are built manually. The rst one is based on a social analysis of the network, and the second one is based both on a cognitive and a pragmatic analysis of interactions in the network. The domain-specic ontologies require a combination of Natural Language Processing (NLP) techniques and manual choice of terms and concepts. These three ontologies are saved on an ontology server that permits to easily retrieve concepts. We focus now on these NLP techniques. Using NLP for building domain-specic ontology

Due to the low availability of domain-specic ontologies and to the fact that generic ontologies are of little use for domain specic applications, many projects aimed to use NLP techniques to extract semi-automatically terms (concept instances) 26 , to create term clusters (concepts) 27 , as well as to extract relations between terms 28 . The expert should name the clusters as concepts and eventually should dene relations between concepts. In our system, NLP techniques are used for two main purposes : for building and for maintaining the domain-specic ontology from corpora, but also for browsing and for indexing annotations. The rst task is done o-line, by extracting terms from a selected corpus and by proposing a simple topic hierarchy (a term is equivalent as a topic). The second task is to help the user indexing his annotation regarding the three points of view (other indexes like author name, date, title, are automatic), by proposing him/her a semi-automatic indexation of his/her annotation. For the rst task, it was not possible to use existing medical ontologies (MeSH 29 , UMLS 30 ), as they are too generic or cover other domains (Menelas 31 ) than our application's one. To build an initial ontology from corpora, we start from text to create a semi-formal ontology (structured with topics) from scratch, and to do so, we need to use a term extractor. We use a simple term extractor, LIKES 32 ,

233

Fig.

2: Topic Map formalism

to identify sequences of words (collocation, or repeated segments) occurring in the corpora. The repeated segments are potential candidate terms and tree-organized, grouped by their heads and displaying their frequency. Candidate terms will be used to select the topics of our ontology. Outputs should be ltered out to eliminate incorrect candidate terms (terms nishing with a preposition, conjunction). Most of the candidate terms match a Head+Modier pattern. We developed a tool (GenTMInd), identifying hierarchical relations between terms via heuristic rules and structuring them in Topic Maps format. This way, a term matching a pattern Head + Modier is a subconcept of the Head concept. For the moment, candidate topics should be identied among simple noun phrases (a noun group followed by at most one prepositional phrase). These hypothesis and heuristic rules are not enough to identify all the hierarchical relations or all the relevant candidate topics. Once we will have a relevant reference corpus, we intend to extend searches among a set of domain-specic verbs, for new candidates. We will explore the context of each candidate topic (a window limited by the phrase boundaries) to identify more relations between topics. If we nd other candidate topics occurring frequently in the context, this means that a horizontal relation should be added between the two candidate topics. For example, the context of "Alzheimer's disease" in the test corpus contains "old person", which means that a relation "concern" between the two topics could be established. (Fig. (3)). For the second task, proposing terms (keywords) to users for indexing annotation means that NLP tools should be able to scan a short text, and match it to the ontology concepts, for each viewpoint. For that point, we want to use a matching

234

Fig.

3: A fragment of the Topic Map ontology for medical domain

algorithm between the text, body of the annotation, and a part of a document or the Topic Maps built by the system. The annotation system will then propose some domain-specic keywords and phrases, as well as typical argumentative phrases to the user. The user will then decide whether the proposed indexation is relevant and needs to be stored as meta-data of the annotation. In the same perspective, the user decides where to put the annotation, and to what part(s) of document, document or documents s/he can bind and anchor the annotation. So, we foresee a complex indexation, with multi-anchoring. The annotation is then stored in the annotation server. The next step to be implemented is to adapt a more ecient term extractor, as FASTR is 33 to identify candidate terms in annotation's body, and to extract the concept hierarchy by adapting some clustering techniques 27 . We are now going to present the distributed architecture of our Web-based annotation system, integrating W3C standards, and NLP tools. 3.5

Architecture

According to Annotea W3C standard, the architecture of our annotation system is a distributed client/server architecture (Fig. (4)) : - Client part : As we mentioned before, we need to annotate the EPF pages, which are accessible via a Web browser. Annozilla, a plug-in for Mozilla Navigator, is an Annotea client that permits to reach this aim. Using XPointer and DOM standards and a lot of Mozilla Framework facilities (XPConnect and XPCom components, see 34 for details), Annozilla gives the user the ability to create, update and remove annotation on a document or on a part of the document, and gives the possibility to store it in a local (individual use) or in a remote server (shared use).

235

Fig.

4: Web-based Annotation System Architecture

- Server part : We have chosen the Annotea client ZAnnot developed under Zope Platform 35 which owns an Object DB and a Web server and some of other components (ZClass, Zope Products...). ZAnnot as a Zope product takes advantage of the Zope Platform and handles all annotations queries sent by the client Annozilla ; in addition it handles and manages reply services. The particularity of our approach is that we have to annotate dynamic (PHP) Web pages. To achieve this, we use the EPF pages to exchange some hidden parameters between EPF Server and ZAnnot like the EPF identier. After we adapted Annozilla to annotate the EPF, we implemented the reply ability between annotations and the indexing function. To classify annotations, we extended the Annotea Annotation Scheme, by adding some indexing metadata (the three viewpoints) which will be saved in RDF format with the other annotations metadata and body. For coherence reasons, the Topic Maps of the dierent points of view are actually stored also in RDF format and they are not modiable by user. We provide users an interface allowing him/her to manage the topics in the dierent points of view and browse the annotations' list through it. When a member of the care network will open an EPF, he/she will be able to open on the left-side

236

of the screen the Annozilla plug-in which permits both to annotate and to retrieve past annotations, sorted according to the indexes which we have dened above. If he/she decides to create a new annotation, this annotation appears in a new window containing its body and the elds for indexation. The next step is to involve the Ontology server and NLP Tool, and thus, the Topic Maps will be stored in the Ontology server and will be continuously updated by the user or by the NLP tool. 4

Conclusion

In this article, we rst described the need of members of a care network to cooperate around information laid in an Electronic Patient File. In a medical context where data is strongly coded and xed, annotation seems to be a natural way of allowing arguing and collaborative work. We have then presented a web-based tool to support collaboration through annotation. We propose to take into account the working context of participants by indexing annotations according to domain, micro-organization structure, and argument process enabling the emerging of new solutions. Our annotation application respects the W3C Annotea Standard, and uses NLP tools both to create semi-formal ontologies which will permit to index annotations, and to help users to index their annotation by the way of a semi-automatic indexation. Some technical points have yet to be improved, and an experiment in the care network is planned. Références

1. Star, S. and Griesemer, J.Institutional ecology, "translations" and boundary objects : amateurs and professionals in Berkeley's museum of vertebrate zoology, 1907-1939, Social Studies of Science, 19, 387-420 1989. 2. Wenger, E.Communities of Practice - Learning, Meaning and Identity, Cambridge University Press, Cambridge 1998. 3. Bannon, L., Kuutti, K. Shifting expertises on organisational memory : from storage to active remembering, in IEEE HICSS 96, Vol. III, Information Systems - Collaboration Systems and Technology IEEE Computer Society Press, Hawaii, pp. 156-1671996. 4. Zacklad M. Barbaud X. Vers une application du Web Socio Sémantique pour la réalisation d'un système d'information destiné aux réseaux de santé, In Second séminaire francophone du Web Sémantique Médical, Rouen, 2004. 5. Weed L.D., The Problem Oriented Record as a Basic Tool in Medical Education, Patient Care and Clinical research. Ann Clin Res 3(3) :131-1341971 6. Caussanel J.,Cahier J-P.,Zacklad M., Charlet J. Les Topic Maps sont-ils un bon candidat pour l'ingénierie du Web sémantique. In Bruno Bachimont, editor, Actes des 6emes Journées Ingénierie des Connaissances, pages 233- 52, Rouen, France, 28-30 mai 2002. 7. Bremer J.M., Gertz M., Web Data Indexing through External Semantic-carrying Annotations. In 11th IEEE Int'l Workshop on Research Issues on Data Engineering : Document management for data intensive business and scientic applications (RIDE-DM'2001), IEEE Computer Society, pp. 69-76 2001.

237

8. Volz R., Oberle D., Motik B., Staab S., KAON SERVER A Semantic Web Management System ? In : Proceedings of the 12th World Wide Web, Alternate Tracks - Practice and Experience, Hungary, Budapest, 2003. 9. Roussey C., Calabretto S., PINON J.-M., SyDoM : A Multilingual Information Retrieval System for Digital Librairies in proc. International Conference ICCC/IFIP On Electronic Publishing (ELPUB'2001), Canterbury (UK), 5-7 july 2001, p. 150-164, 2001 10. Kahan J., Koivunen M.-R., Prud'Hommeaux E., Swick R.R. Annotea : an open RDF Infrastructure for Shared Web Annotations, in Proceedings of WWW10, May 1-5 2001, Hong-Kong, pp. 623-632 2001. 11. http ://www.w3.org/2001/Annotea/ 12. D.Brickley, R.V.Guha, Resource Description Language http ://www.w3.org/TR/rdf-schema/, February 2004 13. http ://www.zope.org/Members/Crouton/ZAnnot/ 14. http ://annozilla.mozdev.org/ 15. S Handschuh, S Staab, and F Ciravegna, S-cream - semi-automatic creation of metadata. In 13th International Conference on Knowledge Engineering and Knowledge Management (EKAW02), October. 2002 16. Alexiei Dingli, Next Generation Annotation Interfaces for Adaptive Information Extraction. In 6th Annual Computer Linguists UK Colloquium (CLUK 03), January, Edinburgh, UK, 2003 17. J B Domingue, M Lanzoni, E Motta, M Vargas-Vera, and F Ciravegna, Mnm : Ontology driven semi-automatic or automatic support for semantic markup. In 13th International Conference on Knowledge Engineering and Knowledge Management (EKAW02), October, 2002 18. L.Denoue, L.Vignollet, An annotation tool for Web browsers and its applications to information retrieval, in proceedings of RIAO 2000, 2000 19. Ka-Ping Lee CritLink : Better hyperlinks for the WWW http ://crit.org/ping/ht98.html, 1998 20. Price M., Schilit B., Golovchinsky G., XLibris : The active reading machine. In proceedings of CHI'98 Human factors in computing systems, Los Angeles, California, USA, vol.2 of Demonstrations : Dynamic Documents, pages 22-23, 1998 21. http ://oce.microsoft.com/fr-fr/assistance/HA010714941036.aspx 22. http ://www.adobe.com/support/techdocs/ac76.htm 23. Zacklad, M., Lewkowicz M., Boujut, J-F., Darses, F., Détienne, F., Formes et gestion des annotations numériques collectives en ingénierie collaborative, actes des journées Ingénierie des Connaissances 2003, Laval 2003. 24. Gangemi A., Guarino N., Masolo C., Oltramari, A. Sweetening WordNet with DOLCE, AI Magazine 24(3) : Fall 2003, 13-24 2003. 25. M. Biezunski, M. Bryan, S. R. Newcomb, Topic Maps, spécication ISO/IEC 13250 ,December the 3thd 1999. 26. Jacquemin C.,Bourigault D., Term Extraction and Automatic Indexing, in Mitkov R. (ed), The Oxford Handbook of Computational Linguistics, Oxford University Press, 2003, pp. 599-615 2003. 27. Cimiano, P. Hotho, A., Staab S. Clustering Concept Hierarchies from Text, in Proceedings of LREC'2004, Lisbon, 25-28 mai 2004, ISBN 2-9517408-1-6, pp. 1721-1724 2004.

238

28. Buitelaar P., Olejnik D., Hutanu M., Schutz A., Declerck T., Sintek, M. Towards Ontology Engineering Based on Linguistic Analysis, in Proceedings of LREC'2004, Lisbon, 25-28 mai 2004, ISBN 2-9517408-1-6, pp.7-11 2004. 29. Medical Subject Headings, http ://disc.vjf.inserm.fr :2010/basismesh/meshv04.html 30. UMLS Knowledge Source Documentation, 2004 http ://www.nlm.nih.gov/research/umls/umlsdoc.html 31. ZWEIGENBAUM (P.) et Consortium MENELAS, MENELAS : an access system for medical records using natural language. In Computer methods and programs in Biomedicine, 45 :117-120, 1994 32. Rousselot,F., Frath, P., Oueslati, R., Extracting concepts and relations from Corpora. In Proceedings of the Workshop on Corpus-oriented Semantic Analysis, European Conference on Articial Intelligence, ECAI 96, Budapest, 12 August 1996. 33. Jacquemin, C.,and Tzoukermann, E., NLP for Term Variant Extraction : A Synergy of Morphology, Lexicon and Syntax. In T. Strzalkowski, editor, Natural Language Information Retrieval, pages 25-74, Kluwer, Boston, MA 1999. 34. N. McFarlane, " Rapid Application Development with Mozilla ", Bruce Perens' Open Source Series,2003. 35. A. Latteier, M. Pelletier, C. McDonough, P. Sabaini, The Zope Book, Edition 2.6. 2003

239