The overall objective of the FERMI project is:
The logic developed within the project will constitute the theoretical basis of information systems supporting the storage and retrieval of multimedia documents, that is, complex documents possibly including text, graphics, images, video, sound and the like.
We propose to develop a theory of the representation and reasoning on the structure and content of complex documents and queries, to develop a theory of the imprecision inherent in the information retrieval process, and to develop a theory of the semantic content of multimedia documents; the first will adopt features of Terminological Logics, Relevance Logics, Modal Logics and Fuzzy Logics, the second on Probability Theory and the Dempster-Shafer Theory of Evidence, while the third will be based mainly on the theory of Conceptual Graphs. We also propose to integrate the three theories we will have developed into a unified Multimedia Information Retrieval Logic (hereafter referred to as the MIRLOG logic), whose expressive power will allow an adequate representation of complex multimedia documents and queries, and whose relation of logical consequence will model a dynamic and effective retrieval of those, and only those, documents that are relevant to the user's information needs.
The central idea of the project is that the retrieval of a multimedia document in response to a user request can, and indeed should, be seen as the result of checking the validity of the formula of the MIRLOG logic, where is the representation of the document, is the representation of the user request, and ``'' is the notion of implication formalized by MIRLOG.
In order to adequately model the retrieval of multimedia documents, the logic we are seeking will have to satisfy a number of requirements. First, the logic must be able to support the representation and reasoning about a number of different aspects of multimedia documents (e.g. those that relate to the structure, the layout and the content of the document, respectively). Second, the logical consequence relation of the MIR logic must capture the notion of ``likely relevance'' typical of information retrieval, a discipline that inherently deals with imprecision: that is, is valid in the logic if and only if the document represented by is likely to be relevant to the information needs expressed by the user through the request represented by . Third, mechanisms have to be enforced in order for the inferential behaviour of the MIRLOG logic to be adaptive, i.e. sensitive to the dynamic nature of relevance. Finally, the problem of deciding whether in MIRLOG must be computationally tractable, that is, solvable with a limited amount of resources.