In the author's opinion, language and clinical facts in psychiatry are misused and it is hypothesized that many malfunctions of psychiatric practise are symptomatic. It is indeed very hard to modelise mental illness : what terms should be chosen and how should they be articulated? Considering the complexity of the relationship between language, ideas and memory, an empirical and experimental approach has been chosen; working out structures of concepts isomorphic to psychiatric reality, either from observed phenomena or from words. In our view, language skills are sufficient to build these sophisticated models, and language skills can be trained as can be trained sensorimotor skills. Experimenting psychiatric language in using their encyclopedic knowledge will allow staff members to teach themselves and will have a heuristic value in changing their beliefs about reality. These experiments have been undertaken for a year and an half in a general psychiatry ward, in a ward for retarded and autistic admits and in a rehabilitation ward. Rigour and perseverance are required but these experiments seem particularly promising.