Foreword



TROEPS is an object based knowledge representation system. It thus allows to model particular domains under the form of objects. This means that the individual entities of the domain are seen as discrete entities with fields denoting their characteristics and their relationship with other such entities. These objects are also grouped into families of objects called concepts or classes which, in turn, can be related to each others by specialisation or attributive relationship. Each construct of the system has a precise semantics which allows to check the models for internal consistency.
TROEPS knowledge bases can be explored in order to find the objects and their relationships. To that extent, turning a knowledge base into a Web knowledge server, either for browsing, editing or exploiting the model, has been made easy. Moreover, lexicons and annotations enable to document the model straightforwardly.
However, the interpretation and exploitation of the model is left to external programs (generally through the TROEPS API). The system provides basic tools for easing the use of models such as sophisticated field value computation, instance classification and conceptual clustering.
This Reference manual provides an in depth presentation of the TROEPS system. It can be found on-line at http://co4.inrialpes.fr/docs/troeps-manual.html.

About TROEPS

TROEPS can be considered as the successor of SHIRKA. It thus shares with SHIRKA the notions of objects, classes, taxonomy, hypertext links and classification facilities. It also adds the possibility of having several taxonomies over the same set of objects, extendibility through abstract data types and constraint satisfaction.
TROEPS is a result of the Sherpa project headed by François Rechenmann from 1988 to 1999. The system stems from the work of the whole Sherpa team and the previous research on SHIRKA and continuous efforts made in order to enhance it. Meanwhile, the overall design of TROEPS has been the early initiative of Olga Mariño, the development of TROEPS has been carried out at various level of involvement by Christophe Alemany, Cécile Capponi, Isabelle Crampé, Pierre Fontanille, Pierre Girard, Jérôme Gensel, Claire Lecourtier, Florence Lemaire, Olivier Schmeltzer, Loïc Tricand De La Goutte, Nina Tayar and Petko Valtchev under the supervision of Jérôme Euzenat.
The actual implemented system is not faithful to the design of Olga Mariño's papers. Some notions (like annotations, lexicon, categorisation) have been added to the system, others (like the semantics of taxonomies) have been changed but the main advantage of the current design is that it works and is implemented. Nevertheless, the implemented system is not the final one. The implementation has been carried out in such a modular way that it will be possible to replace a straightforward implementation by another one (concerning, for example, the memory storage or the basic classification mechanisms) or to add new modules (for instance, clustering facilities or a complete revision mechanism).
The extensions of the system such as the knowledge server are built on top of an API (Application Programming Interface). That API which represents the core of the system is implemented and described in the present manual.
Beware: although other documents, anterior or posterior, can help to understand particular aspects of the system, this reference manual is the only trustworthy documentation of TROEPS 1.3 made available from december 1999 .

Organisation and notation

As said above, TROEPS is made of a restricted set of basic notions. These notions are described in this manual in a constant frame consisting of:
Among the particular notations used in this manual, functions of the API are presented in the following format:
(primitive-name argument1... argumentn) primitive-type in [library-name]

-> returned-type, argument1 argument 1-typeerror-types..., argumentn argumentn-typeerror-types
Description of the primitive behaviour.
Type expressions are expressed in a kind of Backus-Naur form. The error names can be found in Appendix 7. Sometimes, the mention " NIY" and a "--" sign after the closing parenthesis signals that the primitive is not available in the current version. There are very few non available primitives.
Interface are displayed through particular figures called Screendumps. They provide a picture of the HTTP browser screen together with the indications of the next screens when clicking on an active element (link or button). Some screens (mainly dialogues) are omitted since they do not carry a lot of information and other ones are reduced through white stripes across them.
Screendump <number>: <Description of the screen function>. <Additional information>

The examples are given in their TALK form (i.e. not in a C form, however, the use of TALK constructions is limited to the well-known list operations: cons, car, cdr, list and eq). They constitute the creation of a knowledge base extending an already existing base ( real-estate available in the examples, which, by opposition, is the subject of the examples given in the text). So, each example session can be played only after having played the former sessions. A primitive is presented in both the section corresponding to its API description and/or in the section corresponding to its arguments. Finally, the format of TALK examples is made of a form to evaluate (preceded by a question mark) and the results of the evaluation (preceded by an equal sign). The result is printed by the TROEPS printer which displays a simple readable form for TROEPS entities. For instance,
? (tr-find-concept "house")
= (tr-concept "house")
corresponds to the evaluation of (tr-find-concept "house") which returns the "house" concept printed as (tr-concept "house") . This is useful since the result can be cut and pasted in order to provide a reference to the object.
Hereafter, the "designer" will not be considered as different from the "user": a designer has, a priori , no additional facilities to use a knowledge base. So, "user" will be the general term used, and "designer" will be reserved for discussions on strict design problems, if any.

Plan

The manual is made of three main chapters. The first one, Data model , describes the static model used in TROEPS in order to represent knowledge in term of objects. The second one, Computing field values , is dedicated to the dynamic aspect of inferring field values. The third one, Classification principles , concerns a very important aspect of TROEPS: the possibility to find the most suited class for an object with regard to the overall knowledge base. A fourth chapter concerns the setting of a Knowledge server taking advantage of the World wide web.
A set of appendices is added to these four main chapters. They concern:
The novice reader is invited to read the manual through the three main sections; it is possible to reproduce the examples from the first to the last. The installer and/or newcomer should consult the Appendix 1 for a quick start. The programmer will access directly the manual from the function index at the end of the manual or by browsing the on-line manual through a WWW client (see Appendix 1).
Montbonnot, le 20 décembre 1999