*INRIA Sophia-Antipolis, Projet ACACIA, 2004 route de Lucioles,BP 93, 06902 SOPHIA-ANTIPOLIS CEDEX, FRANCE, E-mail: email@example.com
+Université René Descartes, UFR Mathématiques et Informatique, 45 rue des Saints-Pères, 75006 PARIS, FRANCE
This article presents an experiment of expertise capitalization in road traffic accident analysis. We study the integration of models of expertise from different members of an organization into a coherent corporate expertise model. We present our elicitation protocol, and the generic models and tools we exploited for knowledge modelling in this context of multiple experts. We compare the knowledge models obtained for seven experts in accidentology and their representation through conceptual graphs. Last, we discuss on the results of our experiment from a knowledge capitalization viewpoint.
There is an increasing industrial interest in the capitalization of knowledge (i.e. both theoretical knowledge and practical know-how) of groups of people in an organization, such groups being possibly dispersed geographically. The coherent integration of this dispersed knowledge in a corporation is called "corporate memory" ( Steels, 1993 ). The memory of an enterprise includes not only a ";technical memory" obtained by capitalization of its employees' know-how but also an "organizational memory" related to the past and present organizational structures of the enterprise (human resources, project management, etc.). The construction of a corporate memory requires abilities to manage disparate know-how and heterogeneous viewpoints, to make this knowledge accessible to the adequate members of the enterprise, to integrate and store this knowledge in written or electronic documents, or in knowledge bases which should be easily accessible, usable and maintainable. The solutions offered by research ( Macintosh, 1994) to this problem crucial in industry ( Morizet-Mahoudeaux, 1994 ) may be related to: (1) the analysis and modeling of an enterprise (Fox, 1993; Fox, Chionglo, and Fadel, 1993), its evolution through time, the experience acquired from past projects; (2) the integration of models of expertise from different groups in an organization into a coherent corporate expertise model; (3) the construction and integration of distributed, heterogeneous knowledge bases or knowledge-based systems, possibly stemming from multiple experts; (4) the development of "intelligent" electronic documentation (Poitou, 1995) with knowledge bases linked to this documentation (Martin and Alpay, 1996); (5) knowledge sharing between different groups; the exploitation of artificial intelligence techniques such as case-based reasoning (Simon and Grandbastien, 1995; Kitano, Shibata, Shimazu, Kajihara, Sato, 1992), exploitation of multi-agent systems (Oliveira and Shmeil, 1995; Vandenberghe and de Azevedo, 1995), exploitation of Web (Huynh, Popkin and Stecker, 1994), natural language document analysis techniques (Trigano, 1994).
As noticed in (Nonaka, 1991), (Van Engers, Mathies, Leget and Dekker, 1995), the knowledge chain consists of seven links: listing the existing knowledge, determining the required knowledge, developing new knowledge, allocating new and existing knowledge, applying knowledge, maintaining knowledge, disposing of knowledge. So, we can consider the building of the corporate memory as relying on the following steps: 1) Identification and selection of the sources from which the corporate memory can be built: specialists, written or electronic documents, existing databases or case libraries, dictionaries; 2) Identification of the needs in corporate memory: determination of its potential users and of a mode of exploitation useful and adapted to their work environment; 3) Building of the corporate memory, that can consist of written documents or of computational tools: construction of its contents, organization of the corporate memory; 4) Integration, diffusion and exploitation of the corporate memory in the enterprise.
We have worked with INRETS (1) Salon de Provence in a project that resulted in the building of a partial corporate memory of a team of specialists in road traffic accident analysis ( Alpay et al, 1996 ). INRETS Salon de Provence comprise researchers (i.e. experts) and investigators. The experts of INRETS stem from various disciplinaryfields: psychologists specialists of the driver's behaviour, engineers specialists of the vehicle, engineers specialists of road infrastructure. After an accident, INRETS is immediately informed by firemen. Two investigators from INRETS go to the scene of the accident. One of them carries on interviews with the persons involved in the accident, while the other investigator tries to keep track of the accident scene (shots, measurements...). Then, they exchange their first ideas or hypotheses. They determine the missing data, that will require an additional collection, after which both investigators analyze the case together and write an account called "synthesis". Once finished, the accident dossier obtained thanks to this "Detailed Study of Accident" is put in archives. Then, such dossiers are used by the INRETS researchers for thematic analyses, directed towards research topics (e.g. driving behaviour of old persons). The experts of INRETS are presently researchers working mainly on thematic analyses, but several of them had been investigators for several years, several years ago. So, they were interested in an experiment of knowledge capitalization that would enable to formalize their know-how (this formalization could for example result in the writing of a book), to share this expertise within the whole institute, to enhance the work of the new investigators or even of the researchers (in particular by making explicit the competences of researchers from other disciplines), and to improve the communication and cooperation between the experts. In addition to the purpose of formalization of their tacit, implicit knowledge (Nonaka, 1991), they were also interested in a computational tool that could have several purposes: 1) to support traffic accident analysis, in particular to help the new investigators by making available the experts' know-how, 2) to help the experts by making available the abilities of the other researchers from other disciplines.
From the viewpoint of the building of a corporate memory, this experiment helped us to focus on the following questions: how to analyze the evolution of a group through time, the experience acquired from past projects? How to integrate models of expertise from different members of an organization into a coherent corporate expertise model? In this article, we present this experiment of expertise capitalization. First, we describe our method (our elicitation protocol and the generic models and tools we exploited for knowledge modelling). Then, we describe and compare the knowledge models obtained for seven experts in accidentology and their representation through conceptual graphs. Last, we discuss on the results of our experiment from knowledge capitalization viewpoint.
We made an investigation (2) with 18 French or foreign enterprises or research laboratories, that had faced the problem of knowledge acquisition from multiple experts (Amergé et al, 1994). Thus we took stock of: the expertise elicitation techniques and knowledge engineering methods used and their advantages and drawbacks for multi-expertise; the type of expertise conflicts and the methods used for solving them; the other kinds of multi-expertise-related problems encountered; the architecture used for implementation.
Relying on the analysis of previous work, we developed our own method for acquiring knowledge from multiple experts. This method is inspired of the principles of the CommonKADS method (Breuker and Van de Velde, 1994) and relies on several generic models, among which a model of cognitive agent inspired by distributed artificial intelligence. Our method is based on the following steps: 1) Document collection and Knowledge elicitation; 2) Knowledge analysis and modelling; 3) Validation of the knowledge models obtained; 4) Building of the corporate memory. The validated knowledge can be represented in a formalism such as conceptual graphs ( Sowa, 1984) and compared thanks to (informal or formal) techniques of comparison of expertise models.
We performed our observations on seven experts: two specialists of the driver's behaviour, two vehicle engineers and three infrastructure engineers. We needed an elicitation protocol that would enable to build a coherent corporate memory from the integration of the expertise models of the different experts. Our objectives were to determine the individual expertise of each expert, and the influence of other experts upon an individual resolution, to identify the points leading to discussions between the experts, and to determine the help expected from the future computational corporate memory, according to the expertise domain. So we had to be able to compare: (1) individual resolutions of the different experts, (2) an individual resolution and a collective resolution, (3) a resolution of a homogeneous group of experts and an heterogeneous one (homogeneity w.r.t. to the expertise domain).
Taking into account our objectives, we developed an elicitation protocol composed of:
Remarks: It was not possible to observe the experts during actual data collection, in the scene of the accident. The task proposed to the experts during the case studies corresponds to a Detailed Study of Accident, where the analysis of the dossier is aimed at understanding the sequencing of the accident. Five of the seven experts had carried out this activity for several years, several years ago. Presently, they rather moved towards thematic analysis of dossiers constituted by new investigators. Contrarily to the interaction situations in the protocol, usually, the experts don't analyze a dossier together.
Thanks to the elicitation sessions conducted according to the previously described protocol, we collected the following expertise documents:
The analysis of the expertise documents enables the elaboration of three main types of knowledge models: task model ( Duursma, 1993), expertise model (Breuker and Van de Velde, 1994) and agent model (Dieng, Corby and Labidi, 1994).
… We pre-processed some expertise documents with the statistics-based analyzer of natural language texts, ALCESTE - see (Lapalut, 1996) for more details.
We view the realization of a corporate memory in an organization as a cooperative activity between experts, knowledge engineers, developers and potential end-users of this memory. We proposed a model of cognitive agent ( Dieng, Corby and Labidi, 1994b ; Labidi, 1995, 1996) to help the knowledge engineer to model the organization (i.e. human, material and software environment in which the corporate memory will be integrated) and the interactions/cooperation between the experts in situation of problem solving. Our agent model indicates individual characteristics and social characteristics. We also distinguish static characteristics and dynamic characteristics. All such characteristics must be instantiated by the knowledge engineer for the application, thanks to the knowledge acquisition process. The agent's expertise model is described using CommonKads framework. Figure 3 illustrates the method for exploiting the agent model during analysis of the expertise documents.
For the analysis of the expertise documents, we used the previously described generic models; the types of knowledge obtained depended on the nature of the analyzed documents.
From the reports, articles and interviews of an expert, we can obtain: 1/ the explicit view of this expert on the characteristics of the other experts; 2/ his thematic specificities; 3/ his model of the task of collection and analysis of road traffic accidents; 4/ his expertise model, with: (a) at the domain level, his terminology (i.e. ontology), comprising definitions of some terms and examples, typologies of domain concepts, domain models, expertise rules indicating which hypotheses are generated by this expert and from which clues extracted from the accident dossier; (b) at the inference level, the inference structure modeling this expert's reasoning; (c) at the task level, the controlinformation added to this inference structure.
From the analysis of an expert's individual case, we can deduce: 1/ his view on the characteristics of some other absent experts; 2/ the exploitation of his thematic specificities; 3/ his model of the task of traffic accident analysis in situation; 4/ his model of expertise, with: (a) at the domain level, the domain models used, and expertise rules on the generation of hypotheses from clues; (b) at the inference level, the inference structure modeling this expert's reasoning; (c) at the task level, the controlinformation added to this inference structure.
From the analysis of a case processed collectively by this expert and another expert (Dieng, 1995b), we obtain knowledge on: 1/ his view on the characteristics of some other experts: explicit call to the other present expert or explicit need of the capabilities of an absent expert; 2/ the exploitation of his thematic specificities; 3/ his model of the task of analysis of an actual accident, processed in situation (we noticed the parts where all the experts showed their competence and the parts revealing the exclusive competence of one of them); 4/ his model of expertise. It was sometimes difficult to dissociate the experts and to distinguish the individual expertise of each of them, and the collectiveexpertise emerging from their gathering.
In all cases, the knowledge modeled can be structured in a model of agent associated to this expert. So the following instantiated agent models can appear (see Figure 5): 1/ an agent corresponding to the knowledge modeled from the articles, reports and interviews (i.e. "rationalized" knowledge); 2/ an individual agent corresponding to the work of the expert alone, in situation; 3/ a specific agent corresponding to the aspects specific to this expert when he was solving a case with another expert; 4/ an agent common to this expert and to another expert, and corresponding to their collective expertise during their collective case studies. We modeled the agents' individual aspects such as task model or expertise model and their social aspects (e.g. viewpoint on the other agents, interaction with the other agents).
We carried out two types of informal validation of the models built: with the other knowledge engineers and with the experts themselves. The knowledge engineer presented the models to the experts, and gave him explanations if needed. If the expert did not agree with such models, he was required to give explanations. These comments and remarks were then studied and, if needed, the models were modified. The validation could be verbal (i.e. a model was submitted to the expert and we discussed verbally about its validity) or written (i.e. the graphical representation of a model was submitted to the expert, and he wrote alone his comments about this model).
When the expertise capitalization in an enterprise involves several experts for knowledge acquisition phase, the knowledge engineers must detect and solve several kinds of conflicts: (a) differences of terminology, (b) incompatibility between terminologies, (c) differences between compatible reasonings (i.e. the experts use different problem solving methods but obtain non contradictory results), (d) incompatibility of reasonings (i.e. the different problem solving methods used by the experts lead to contradictory results). In knowledge acquisition methods, expertise conflict management is taken into account by study of terminology conflicts (Gaines and Shaw, 1989), management of several viewpoints (Easterbrook, 1991), conflict detection in the framework of KADS-I methodology (Dieng, 1995a) or library of concurrent design conflict models in the framework of CommonKADS (Matta, 1996).
The knowledge engineer can analyze the expertise documents in order to build: a) a common model corresponding to the kernel of knowledge common to all experts and perhaps models common only to subgroups of experts, b) specific models corresponding to knowledge specific to an expert and not shared by other experts. Two approaches are possible: (1) either the knowledge engineer tries to build such models directly from the expertise documents, or (2) he builds separately each model of expertise corresponding to each expert and then tries to compare the obtained models of expertise in order to find their common parts and their specific parts. In our case, eight knowledge engineers were involved in the construction of the expertise models: each of us was responsible for modeling (a specific aspect) of one expert.
We represent each expert by an artificial agent (Dieng, Corby and Labidi, 1994; Labidi, 1995, 1996) whose expertise model is described in CommonKads. Moreover, we represent the concepts and relations of the domain layer by conceptual graphs (Sowa, 1984). As shown in Figure 4 , an agent has (a) a support indicating the conceptual vocabulary (a support is composed of a concept type lattice, a relation type hierarchy, a set of markers and a conformity relation) and (b) a base of canonical conceptual graphs, built on this support and representing the view of this agent on the world, as well as his expertise. This base of canonical conceptual graphs is subdivided into several different viewpoints (see Figure 5). Therefore, the detection of conflicts between several expertises is based on the comparison of the domain levels of the expertise models of the agents associated to the experts, such domain levels being represented through conceptual graphs.
Searching the common support associated to several experts of the same domain can be seen as a part of the search of a common, shared or accepted ontology among the experts (Garcia, 1996). One can work either at the knowledge level (Newell, 1982), without choosing a representation formalism or at the symbol level, once chosen a representation formalism. Our choice of the framework of the conceptual graph formalism allows us to propose algorithms based on the notions underlying conceptual graphs. In (Dieng, 1996), we proposed a procedure for comparison of the expertise models of two experts, procedure based on the following steps: 1/ comparison and integration of both supports, so as to build a common support, 2/ comparison of the two bases de CGs , 3/ construction of the base of integrated CGs, according to the chosen integration strategy.
The knowledge models obtained for each expert comprise: his terminology, his task model, his expertise model (with, in particular, the domain models used by this expert and his expertise rules), his agent model and his links with the other experts. We applied our research on a model of cognitive agent to the design of the agents corresponding to the experts of INRETS. The construction of expertise models associated to these agents relied on CommonKADS methodology and exploited the generic models for the tasks of modeling and diagnosis, offered in the CommonKADS library (Breuker and Van de Velde, 1994). For each expert, sessions of individual or collective validation allowed to correct, deepen, refine the obtained models.
The next figures give examples of the knowledge models of the different experts.
We used CGKAT (Conceptual Graph Knowledge Acquisition Tool), a tool developed in the Acacia project, (Martin, 1995) to build the hierarchies of concept types and of relation types for the different experts as well as conceptual graphs describing the reasoning strategies of the different experts (Alpay, 1996). We exploited the predefined ontology offered by CGKAT, as well as its capabilities of visualizing conceptual graphs. CGKAT also enables to associate a base of conceptual graphs to the structured documents constituted by the expertise documents (Martin and Alpay, 1996). Figure 5 shows an example of conceptual graph base subdivided into several viewpoints.
The structures of the expertise and agent models (see Figure 7, Figure 8 and Figure 9) help to identify natural criteria of comparison between the knowledge models of different experts.
There were different viewpoints on the concepts of scenario and of factor.
(b) Inference and task
The vehicle engineers (cf. Figure 8 ) rather carry out modeling since their main task, kinematics reconstitution, can be modeled as a modeling task (see Figure 12 ). The psychologists (cf. Figure 7 )
(c) Domain models
Some domain models are used by all the experts and seem to characterize accidentology, independently of any disciplinary aspect: 1/ the CVI model (cf. Figure 10 ): all the experts noticed the importance of the interactions between the components of this model; 2/ the model of cutting of the accident into phases: driving, accident, urgence and crash situations. But some experts personnalize this model by introducing an approach situation and a pre-accident situation.
Some models seem specific to a discipline: only the vehicle engineers made explicit a model of vehicle mechanical defaults and a model of kinematics sequences; the models of tracks are exploited by the infrastructure engineers and by the vehicle engineers; the cognitive models of the driver are typical to the psychologists and most of the infrastructure engineers (cf. influence of the infrastructure on the roaduser's behaviour). Within a given discipline, we can also take into account the specific models acquired by an expert thanks to his thematic research: for example, one psychologist has a model of drivers' malfunctionings and a model of help to driving while the other psychologist has a model of the crossroad driver and of the GTI vehicle driver. The detailed models of infrastructure are specific to the infrastructure engineers (as an exception, a psychologist has an expertise due to his thematic analyses on the drivers in crossroads; this expertise appears through his deep model of crossroads). We can also distinguish the explicit use and the implicit use of a model by an expert.
For an expert, we can notice differences between: (1) the model of the «rationalized» task, obtained from his articles and interviews, (2) the task model obtained from his individual case studies; moreover, we could have studied the evolution of this task model as the expertise elicitation sessions advanced, thanks to the influence of the collective case studies, (3) the task model obtained from the collective case studies. According to the discipline, some subtasks were emphasized: kinematics reconstitution by the vehicle engineers, drivers' interview by the psychologists, analysis of the tracks on the pavement by the infrastructure engineers.
Plan, maps, check-lists, drivers' interviews are used by all the experts. The exploitation of photos depends on the discipline. Only the vehicle engineers use the ANAC software. The infrastructure engineers use very specific resources such as inking-pad, etc.
(b) Interaction points
The input interaction points of the agents (i.e. the requests the corresponding experts can receive) depend on the discipline: the elicitation techniques during the interviews with the drivers, the analysis of the drivers' reliability degree are typical of the psychologists; the task of kinematics reconstitution, the analysis of the vehicle mechanical defaults and of the tracks on the road are characteristic of the vehicle engineers; the design of the infrastructure, the characteristics of the roadway, the analysis of the tracks are logically the concern of the infrastructure engineers.
(c) Models of the other agents
Psychologists as infrastructure engineers described their view of the task of kinematics reconstitution, task specific to their colleagues vehicle engineers. The analysis of the case study common to a psychologist and an infrastructure engineer revealed their respective implicit view on each other, as well as their view on the task of kinematics reconstitution that was the concern of the discipline absent from this collective case study. Likewise, some experts made explicit their view on the terminology of other experts (in particular, on the notions of scenario and of factor).
The common points revealed between the experts attest the existence of an expertise common in accidentology, and based on the task of collection and accident analysis, exploiting the model of the CVI system and the model of cutting of the accident into phases.
At the level of the expertise models, the specificities of the experts show the diversity and complementarity of their disciplines, as well as the potential interest of a computational corporate memory that would include an interdisciplinary expertise: for example, the model of information processing by the driver and the cognitive model of the driver in some infrastructure types or on board of some vehicle types (models exploited by the psychologists), the expertise on collection and analysis of tracks, on the design of road infrastructures and of pavements (expertise of the infrastructure engineers) and the expertise on the deep kinematics reconstitution (speciality of the vehicle engineers).
We proposed a general method for knowledge acquisition from multiple experts, based on:
Concerning the questions related to the construction of corporate memory, we analyzed the INRETS-Salon de Provence, its history and its evolution through time, the experience acquired from the Detailed Study of Accident period. We studied how the expertise models of several experts of different specialties could be integrated in a common expertise model that would be part of the INRETS corporate memory. We applied our solutions to the case of experts in accidentology: elicitation protocol applied with seven experts of INRETS, modeling of the knowledge of each expert (e.g. terminology, expertise model, cognitive agents associated to this expert), representation of some expertise models through the formalism of conceptual graphs, implementation of conceptual graph bases in the knowledge acquisition tool CGKAT - such bases, rather aimed at the knowledge enginners, will not constitute the actual computational corporate memory - , comparison of the knowledge models of the different experts. For accidentology, our research allowed to build a significant base of expertise models of several experts in accidentology, stemming from different specialties (psychology, vehicle engineering, infrastructure engineering). Thanks to this capitalization of multiple expertises, the knowledge thus modeled constitute a (partial) corporate memory of INRETS (Alpay et al, 1996). They can be used as basis for the construction of a system for support to road traffic accident analysis, system that would take advantage of the competences of several specialists of various disciplines.
One result of this experiment was the improvement of the practice of the experts: for example, after being implicitly influenced by an infrastructure engineer during the collective case studies, one psychologist started to use more thoroughly the photos. He had realized that the information he previously extracted from the infrastructure or vehicle textual check-lists was more accurate when visualized through photos. Nonaka's theory (Nonaka, 1991), as described in (Morizet-Mahoudeaux, 1994) considers that knowledge spreads through socialization, articulation, combination and internalization. He distinguishes tacit knowledge (know-how that an expert gained from practice) and explicit knowledge. As INRETS is a research center, some knowledge was explicit within written documents such articles, reports. As the dossiers of the processed accidents were kept in archives and exploited by the present researchers for their thematic analysis, such dossiers can be seen as cases describing both problem data (i.e. accident data) and a solution (i.e. the accident synthesis, with the accident reconstitution and the accident factors determined). But the reasoning leading to the synthesis was tacit and our elicitation protocol (cf. the case studies) helped to make this knowledge explicit. We also made explicit the models used by the experts in their reasoning, in particular models dependent on their discipline and models dependent on their previous experience such as the thematic analyses they had carried out in the past. An explicit base of generic scenarios is a part of the corporate memory. If several accident dossiers characterized by a given parameter value (such as a type of driver - e.g. old drivers - or a infrastructure type - e.g. crossroads - or a vehicle type - e.g. GTI -) reveal similar types of accident factors, the expert performing this thematic analysis builds some generic scenarios of accident, based on such cases. The expert's reasoning seems close to a case-based reasoning: in front of a new accident, he retrieves similar previous cases or he relies on a generic scenario he built thanks to his previous thematic analyses. So, case-based reasoning can play an important role for building of a corporate memory (Simon and Grandbastien, 1995; Kitano, Shibata, Shimazu, Kajihara, Sato, 1992).
During our experiment, the INRETS researchers wrote a book on their methodology of Detailed Study of Accident (Ferrandez & al, 1995). This "guide of good practice" will be typically useful for the new investigators. So our experiment had an influence on (individual and collective) learning in INRETS. As cited in (Morizet-Mahoudeaux, 1994), «making personal knowledge available to others is the central activity of the knowledge creating company». We enabled a process of articulation (i.e. making tacit knowledge explicit) and a process of internalization (i.e. extending one's tacit knowledge by explicit one). Figure 15 shows the architecture of the future computational corporate memory that will be used by the investigators at INRETS.
We thank the «Ministère de l'Enseignement Supérieur et de la Recherche» (contract n. 92 C075), the «Ministère de l'Equipement, des Transports et du Tourisme» (contract n. 93.0033) that funded this research, and ModelAge WG (EP:8319). We thank very much the experts of INRETS, Salon de Provence (Francis Ferrandez, Dominique Fleury, Yves Girard, Jean-Louis Jourdan, Daniel Lechner, Jean-Emmanuel Michel, Pierre Van Elslande) for their cooperativeness. We thank Stéphane Boyera that had partially taken part to the modeling work in accidentology in 1994. We thank the whole ACACIA team for very fruitful discussions on corporate memory and Johan Vanwelkenhuysen for his help in the preparation of this KAW'96 Track.
Alpay, L. (1996) . Modelling of reasoning strategies, and representation through conceptual graphs: application to accidentology. Rapport de recherche INRIA, n. 2810.
Alpay, L. , Amergé, C., Corby, O., Dieng, R., Giboin, A., Labidi, S., Lapalut, S., Ferrandez, F., Fleury, D., Girard, Y., Jourdan, J.-L., Lechner, D. Michel, J. E., Van Elslande, P. (1996) . Acquisition et modélisation des connaissances dans le cadre d'une coopération entre plusieurs experts : Application à un système d'aide à l'analyse de l'accident de la route. Rapport final du contrat DRAST n. 93.0033.
Amergé, C. , Corby, O., Dieng, R., Fleury, D., Giboin, A., Labidi, S., and Lapalut, S. (1994) . Acquisition et modelling des connaissances dans le cadre d'une coopération entre plusieurs experts : Application à un système d'aide à l'analyse de l'accident de la route. Rapport intermédiaire du contrat DRAST n. 93.0033.
Breuker, J. and Van de Velde, W., eds (1994) . CommonKADS Library for Expertise Modelling, Reusable Problem Solving Components. IOS Press, Amsterdam.
Dieng, R (1995a) . Conflict Management in Knowledge Acquisition. AIEDAM, Special Issue on Conflict Management in Design, 9(4): 337-351.
Dieng, R (1995b). Specifying a Cooperative System through Agent-Based Knowledge Acquisition. KAW'95, pp. 20-1 20-20.
Dieng, R (1996). Comparison of Conceptual Graphs for Modelling Knowledge from Multiple Experts. In Ras Z. W. & Michalewicz M. eds., Foundations of Intelligent Systems, ISMIS'96, Springer-Verlag, LNAI 1079, pp. 78-87.
Dieng, R, Corby, O., Labidi, S. (1994). Agent-based Knowledge Acquisition. In Steels, L., Schreiber, G., Van de Velde, W., eds, A Future for Knowledge Acquisition : EKAW'94, Bruxelles, Springer-Verlag, LNAI n. 867, pp. 63-82.
Duursma, C. (1993). Task Model Definition and Task Analysis Process. Report # KADS-II/M5/VUB/RR/004/1.1b.
Easterbrook, S. (1991). Handling conflict between domain descriptions with computer-supported negotiation. Knowledge Acquisition, 3(3):255-289.
Ferrandez, F., Brenac, T., Girard, Y., Jourdan, J.-L., Lechner, D. Michel, J. E., Nachtergaele, C. (1995). L'étude détaillée d'accidents orientée vers la sécurité primaire : Méthodologie de recueil et de pré-analyse. INRETS, Presses de l'E.N.P.C.
Fox, M. S. (1993). Issues in Enterprise Modelling. Proc. of the IEEE Conf. on Systems, Man and Cybernetics, Le Toquet France.
Fox, M., Chionglo, J.F., and Fadel, F.G. (1993) . A Common Sense Model of the Enterprise. Proceedings of the 2nd Industrial Engineering Research Conference , pp. 425-429, Norcross GA: Institute for Industrial Engineers.
Gaines, B. R. and Shaw, M. L. G. (1989) . Comparing the Conceptual Systems of Experts. Proceedings of the 9th IJCAI (IJCAI'89), pp. 633-638, Detroit, USA.
Garcia, C. (1996) . Co-operative Building of an Ontology within Multi-Expertise Framework. In Proc. of the Second International Conference on Cooperative Systems (COOP'96), Juan-les-Pins, pp. 435-454.
Huynh M., Popkin L., Stecker M. (1994). Constructing a Corporate Memory Infrastructure from Internet Discovery Technology. White Papers, Marble Associates Inc., http://www.marble.com/chi-bin/list-white-papers.
Kitano H., Shibata A., Shimazu H., Kajihara J., Sato A. (1992). Building Large-Scale and Corporate-wide Case-based Systems: Integration of Organizational and Machine Executable Algorithms. Proc. of AAAI'92, pp. 843-849.
Labidi, S. (1995). Ingénierie de la connaissance dans le cadre de projets multi-experts : Méthode, techniques et outils. Thèse de Doctorat en Informatique. Université de Nice-Sophia-Antipolis.
Labidi, S. (1996). Supporting Inter-Experts Cooperation in Knowledge Acquisition Stage. In Proc. of the Second International Conference on Cooperative Systems (COOP'96), Juan-les-Pins, pp. 207-222.
Lapalut, S. (1996). Text Clustering to help Knowledge Acquisition from Documents. In Shadbolt, N. , O'Hara, K., Schreiber, G. eds, Advances in Knowledge Acquisition, EKAW-96, Nottingham, England, Springer-Verlag, LNAI 1076, pp. 115-130.
Macintosh, A. (1994). Corporate Knowledge Management State-of-the-Art Review. Proc. of the 2nd Int. Symposium on the Management of Industrial and Corporate Knowledge (ISMICK'94), Compiègne, France, pp. 131-145
Martin., Ph. (1995). Knowledge acquisition using documents, conceptual graphs and a semantically structured dictionary. Proc. of KAW'95, Banff, Canada, pp. 9-1 9-19.
Martin, Ph. & Alpay, L. (1996). Conceptual Structures and Structured Document. In Eklund, P. W. , Ellis, G., Mann, G. eds, Conceptual Structures: Knowledge Representation as Interlingua, ICCS'96, Springer-Verlag, LNAI 1115, pp. 145-159.
Matta, N. (1996). Conflict Management in Concurrent Engineering: Modelling Guides. In Mueller, J. & Dieng, R., eds, Proc. of ECAI'96 Workshop on Modelling Conflicts in A.I., Budapest, Hungary, pp. 29 - 33.
Morizet-Mahoudeaux, P. (1994). Basic Research Knowledge Management in Industry, Academia and Administration: A Japanese Perspective. Proc. of ISMICK'94, Compiègne, France, pp. 3-18.
Newell A. (1982). The Knowledge Level. Artificial Intelligence, 18(2):87127.
Nonaka, I. (1991). The knowledge-creating company. Harvard Business Review. Nov-Dec. 1991, pp. 96-104.
Oliveira E. and Shmeil M. A. H. (1995). ARTOR - An Approach towards Artificial Organizations. Proc. of the 3rd Int. Symposium on the Management of Industrial and Corporate Knowledge (ISMICK'95), Compiègne, France, pp. 29-39.
Poitou, J.P. (1995). Documentation is Knowledge: An Anthropological Approach to Corporate Knowledge Management. Proc. of the 3rd Int. Symp. on the Management of Industrial and Corporate Knowledge (ISMICK'95), Compiègne, France, pp. 91-103.
Simon, G. and Grandbastien, M. (1995). Corporate knowledge: a case study in the detection of metallurgical flaws. Proc. of ISMICK'95, Compiègne, France, pp. 43-52.
Sowa, J. F. (1984). Conceptual Structures, information processing in mind and machine. Addison-Wesley System Progr. Series.
Steels, L. (1993). Corporate knowledge management. Proc. of ISMICK'93, Compiègne, France, pp. 9-30.
Trigano, P. (1994). Automatic Indexation and Knowledge Storing. Proc. of ISMICK'94, Compiègne, France, pp. 223-235.
Van Engers, T. V., Mathies H., Leget J., and Dekker C.C.C. (1995). Knowledge Management in the Dutch Tax and Customs Administration: Professionalisation within a Knowledge Intensive Organization. ISMICK'95, Compiègne, France, pp. 71-80.
Vandenberghe, L., and de Azevedo, H. (1995). Multi-Agent Systems & Knowlege Capitalisation: an Overview. Proc. of ISMICK'95, Compiègne, France, pp. 61-68.