Our main goal is to model real world problems, especially in the domain of medicine or biology. Thanks to large interaction with experts we have created Milord, a language and a tool for Expert Systems.

The main characteristics of Milord are:

• Modular language. Modules are subsystems specialised in parts of a whole problem. Modular decomposition facilitates users to program in the large dealing with the programming and validation of smaller problems. Each module contains its own object-level (facts, rules) and meta-level (meta-rules).
• Rule-based system. Propositional rules are a good knowledge representation for users without programming skills. This formalism facilitates the interaction among experts and knowledge engineers.
• Approximate reasoning capabilities. Uncertainty and imprecision are inherent to real world problems. The object-level of a module is based on a particular many-valued logic of linguistic terms.
• Complex reasoning. Control is based on reflection techniques over a meta-language (Horn-like rules) equipped with a declarative backtracking mechanism tackling the problem of knowledge incompleteness.

The applications that we have developed with Milord have constituted a precious feedback to improve our language, for instance in the domains of diagnosis and treatment of pneumonia, reumathology or classification of marina sponges.