The topoi as a pragmatic paradigm of knowledge representation Extended Abstract Brigitte TROUSSE Daniel GALARRETA INRIA-Sophia Antipolis CNES SECOIA Group AI Group - TE/IS/MIS 2004, route des Lucioles 18 av Edouard Belin 06565 Valbonne Cedex FRANCE 31055 Toulouse Cedex FRANCE E-mail:
[email protected] E-mail:
[email protected]
Introduction
The aim of this paper is to set the question of knowledge representation in a pragmatic framework, i.e. related to action. However in this paper we restrained the scope of the pragmatic approach to linguistic representations of action - and knowledge - which fall within the domain of Pragmatics and Linguistics of Enounciation : namely the topoi [AD83]. In short topoi are common sense rules which deal with the ordering of quantities in their respective space. To be more explicit about our intentions :
First we shall give enlightenments about what we mean by a pragmatic view.
Second we shall criticize the knowledge level approach of A. Newell [New82] showing that it is a pragmatic view of an \intelligent agent".
Third we introduce the concept of topoi in relation with Pragmatics (seen in relation with linguistics).
Fourth we give examples of application to problem solving activities.
A pragmatic view of knowledge
Activity is a set of coordinated acts which makes sense among a given human organization. Activity is allowed and supported by the organization which gains in turn more legitimacy from it. Arti cial intelligence has tried to model and support problem solving activities among the most usual of them, e.g. interpretation of a situation or plani cation of an action. Activities must be distinguished from action : activity is derived from the observation of an agent who is ascribed to a task; it is descriptive in nature, action is not. Action usually goes further than one can predict and its eects are always irreversible. Action always takes place in a real system : e.g. a physical or social organization. Activities result from using tools which establish mediations between the agents and the reality. A tool is anything which can serve a peculiar purpose. It is related in every case to a de nite activity (however a task can use dierent tools to reach a de nite goal).
A criticism of the knowledge level of A. Newell
Let us call the pragmatic extension of a tool the set of all the eects throughout space and time (whatever the limitations we impose to them) of the use of this tool within a peculiar activity. Of course all the eects cannot be fortold and the set of eects can be seen as a 1
complex system in the sense of J.L. Lemoigne [Lem90] or H.A Simon [Sim82], i.e it cannot be reduced to a single (even complicated) model but must be described by several models : e.g economical, psychological, linguistic ones. It must be stressed that the concept of pragmatic extension of a tool is quite general and can be applied to any tool (used by a given activity) :
a pen, - for the writting activity -,
a computer with its software - for programming activity - ,
an intelligent agent in the sense of A. Newell - in the activity of solving a peculiar problem, diagnosis for instance.
It is worth noting that the pragmatic extension of a tool varies with the tool (e.g a crane or a personnal computer) and the context in which it is used (e.g - about a P.C - to make single calculations or to run a simulator). But the material form of a tool is not sucient to evaluate its pragmatic extension. It is the nality that the user will assign to it - in a given task - that can help in such an evaluation. In the case of an intelligent agent, a part of its pragmatic extension has been recognized and named the knowledge level of the agent, by A. Newell : \Knowledge is to be characterized entirely functionnaly in terms of what it does, not structurally in terms of physically objects with particular properties and relations"[New82]. However we do say that this extension still exists for any non-intelligent tool even if it is ignored.
Topoi in Pragmatics
Every activity need and is based upon an argumentation, i.e a sequence of dialogs within a group of speakers which can be the social body itself or a scienti c or technical community in order to explain, convince about, design a solution which will correspond to an activity. As G. Vignaux [Vig88] says : \The danger here is to consider argumentation as a poor proof, the clumsiness of which is to be explained at a technical level or to be search on the side of social practices which are, by nature far from any rigorous conduct of the discourse". A theoritical framework provided by O. Ducrot and J.C. Anscombre [Duc88, AD83] (see also P.Y. Raccah [Rac89, BR87]) establishes that in many cases argumentation is based upon gradual inference rules called topoi. These rules link gradual properties called topic elds. The linguistic predicates are seen as degrees of these topical eld. Given two topical elds P and Q a topos has one of the following forms : \the more x is P, the more y is Q" or \the more x is P, the less y is Q" or \the less x is P, the more y is Q" or \the less x is P, the less y is Q". where x and y are members of the elds P and Q respectively. Topoi can be seen as linguistic translation and descriptors of actions within argumentation.
AI applications using topoi
In this part, we examine the relation between the orientation of the topoi and the activities. This question is considered on the base of argumentative knowledge. Most of the given examples and the motivation come from practical experiences (in spatial engineering [Tro87], ...) or from a few works in cognitive psychology [Bon89]. 2
We rst analyse the type of activities which A.I. has dealt with and which are related to argumentative knowledge : knowledge acquisition [DT88], validation of knowledge bases [Dav89], user-assistance in redesign [Opp91, Tro92], etc. We will then propose to class the activities into two groups : activities where we need to analyse text to generate topoi, activities which consist rst in analysing the activity itself and then in generating topoi. The conclusion of this part aims at giving more insight within the use of argumentative knowledge such as topoi in accordance with the activities.
References
[AD83] J.C. Anscombre and O. Ducrot. L'agumentation dans la langue. 1983. Bruxelles, Mardaga. [Bon89] N. Bonnardel. L'evaluation de solutions dans la resolution de problemes de conception. Rapport de recherche RR-1072, INRIA, Ao^ut 1989. [BR87] S. Bruxelles and P.Y. Raccah. Argumentation and information: how to express consequences. In COGNITIVA, Paris France, May 1987. [Dav89] H. Davis. Using models of dynamic behavior in expert systems. In 9emes Journees Internationales sur les Systemes Experts et leurs Applications, pages 393{404, Avignon, France, Mai-Juin 1989. [DT88] R. Dieng and B. Trousse. 3DKAT, a dependency-driven dynamic-knowledge acquisition tool. In 3rd International Symposium on Knowledge engineering, Madrid, Spain, October 1988. [Duc88] O. Ducrot. Topoi et formes topiques. Bulletin d'etudes de linguistique fran caise, (22):1{14, 1988. [Lem90] J.L. Lemoigne. La modelisation des systemes complexes. Dunod- Afcet systemes, Paris, 1990. [New82] A. Newell. The knowledge level. AI Journal, 19(2), 1982. [Opp91] O. Oppizzi. Assistance logicielle a la conception : gageure pour les sciences cognitives. Technical report, UTC Sevenans, Fevrier-Juillet 1991. Rapport de stage. [Rac89] P. Y. Raccah. Modelling argumentation and modelling with argumentation. Argumentation, 1989. [Sim82] H.A. Simon. Models of bounded rationality. MIT Press, Boston, 1982. [Tro87] B. Trousse. EXSAT, systeme expert de conception de satellites de telecommunications. In EC2, editor, Journees internationale sur les systemes experts, Avignon, Juin 1987. [Tro92] B. Trousse. Vers une veritable assistance logicielle a la conception. In 01 DESIGN 922eme table ronde fran cophone sur la prise en compte des processus creatifs dans des systemes informatiques, Marrakech, Maroc, 25-27 Janvier 1992. EUROPIA et Team-Maroc. [Vig88] G Vignaux. Le discours acteur du monde . Enonciation, argumentation et cognition. Ophrys, 1988.
3