The roles played by the context are different if the object is mobile or not. The reason is that the environment itself has a dynamic particularly important for a mobile object. For example, changes in the environment can transform an optimal solution in an inadequate solution in another context [Br zillon, 1999]. Moreover, two concepts can be close in a context and distant in another one. A mobile object must be able to revise all its beliefs at a given moment, even during the course of a plan execution. This supposes that the object can follow users' actions and watch for an eventual derive of the observed behaviour with respect to a predicted behaviour (e.g. see [Br zillon, 2000] on case-based intelligent assistant systems). The objective is to ensure that the local user 's needs respect global constraints at any time.
Up to date, the dynamic of the environment is taken into account through the evolution of physical factors as user's location, time of the request. However, this dynamic should also take into account knowledge, not only data, on the environment and the user.
A system using contextual knowledge can develop a user's model increasingly elaborated along user-system interaction. Note that we do not speak of a model drawn from a library but an online modelling of the user as the system can view him, i.e. through their interaction. Thus, the system can provide relevant answers to users' questions, and even helping first the user in the formulation of his questions. The experience acquired by the system with a user accomplishing a task then can be reused for helping the same user in other tasks .
A system can also reuse the experience acquired with a user for helping other users with the same task. This is realised directly with either a stand-alone system or by interaction among agents. In the latter situation, each agent helps a user (e.g. see the works of Maes at MIT for the last approach), the agents exchanging their experience with their user to support other agents .
A system can support a collaborative work between humans by intervening in all the phases of the collaboration (cooperation, negotiation, etc.). This situation is increasingly important when manufactured objects are more and more complex and require the collaboration of different specialists (think of the design of a spacecraft). The system can then take in charge the adjustment of individual contexts of the users in order to make compatible their interpretation on a given event [Karsenty, 1995]. For example, a TV, as a communicating object, would have to make compatible the interests (eventually diverging interests) of the father, the mother and the children (say, a boy and a girl).
The granularity of the context can be compared to a distance measure from a contextual element to the focus of attention. The closer the contextual element is to the focus, more detailed the context. For example, for sending a letter, you need to know the way from where you are to the nearest letter box, when you only need to know that the location of Scotland (from Paris) is North.
Practically, context granularity is restricted now to the distinction between a local context and a global context. Van Dijk (1998) gives a good example in the analysis of political discourses. In computer-aware applications, the Fisheye system does a similar operation [Pook, 2000]. However, this approach is not new: Conceptual graphs already proposes mechanisms of aggregation and expansion [Sowa, 2000]). Nevertheless, context must be represented in a machine in an efficient way for modelling knowledge and reasoning, from the programming point of view as well as the viewpoint of its effective use.
An explicit use of the context may bring some insights on known problems in information technology, such as the management of information presentations in response to a query, a support in the formulation of a query, information exchange between heterogeneous databases. Figure 21.4 presents where different contexts intervene in the interaction between a user and a system (as a communicating object).
Moreover, Goh proposes a context manager to make compatible the contexts of the emitter ontology and the context of the receiver ontology [Goh, 1995]. Indeed, the main observation here is that context would permit a dynamic organisation of the data, the information, and the knowledge in memory for the extraction of the elements of an answer as well as the acquisition of new items.