We have designed the MICO model (Observer-based Constructional Interpretation Model) in order to provide an answer to certain needs of pratical dialogue systems, such as genericity and adequacy with cognitive theories, which have to be fulfilled in order to produce evolutive and expressive dialogue systems. The constructional interpretation model provides a homogeneous way to describe treatments of different inputs of the dialogue system, such as textual, visual or other kinds of inputs.The system is based on a single kind of operation: s-constructions (situated constructions), an advanced version of Fillmore's constructions. S-constructions offer the ability to describe the way information such as temporal of visual information can be processed. In MICO, the structures containing the information are called contexts. The implementation of the contexts is let to an external part, the model only helps to describe the interface of the structure, through relations and operations on the locations of the structure.We introduced the notion of observer for the implementation of the model. Observers are the computational counterpart of s-constructions, and are able to recognize patterns in context, then to produce new information. In order to make the model computationally tractable, we also propose to combine several mechanisms for execution control.As an illustration of our MICO's expressiveness, we conceived a concrete reference resolution model handling vague predicates and using a continuous version of domains of reference (Corblin, Reboul). We propose a representation for one of the possible roles of referential predicates, based on several functions acting on domains of reference, and we adapt it to the constructional interpretation model.
