Section: New Results

Scenario Description Language

Participants : Sabine Moisan, Annie Ressouche, Jean-Paul Rigault, Nazli Temur, François Brémond.

Last year, we developed a scenario recognition engine based on the Synchronous Model of reactive systems. We now need a scenario description language friendly to our end users who are not computer scientists in general. In fact, Stars has already defined such a language. However, it is a declarative language based on (temporal) constraints. This is certainly not the most natural and the simplest way for end users to express their domain specific scenarios.

Consequently, we started this year a comparative study of different means to express scenarios in various domains (video understanding but also games, movies, music, criminology, military strategy...). We investigated 16 formalisms covering these domains. We defined a comparison grid based on criteria relevant for our video understanding goals. We retained 9 such criteria: application domain scope, ease of use, representation of scenario basic elements (background, scene, roles...), modularity (possibility of scenario hierarchy), time representation (absolute, logical, multi-clocks, no clocks...), expression of temporal constraints, representation of repetitive patterns, support for concurrency and parallelism, and finally formal foundations.

To complete the study, we conducted an experiment: describing a case study scenario using some of these formalisms to concretely estimate their advantages and drawbacks, especially their ease of use.

At this time, none of the studied languages fulfills completely our needs. Many languages are graphical ones. While this may appear as user friendly, scalability and automatic analysis become an issue. Some languages lack formal semantics, which is not acceptable in our case; others are merely extensions of computer languages, hence dedicated to specialists.

We plan to define our own version, which will rely on solid semantic foundations. (see section  7.21 ). To enforce user-friendliness, we started to collaborate with ergonomists.