In modern information and documentation systems, the thesaurus has the important function of serving as a link between the author of a document and the information searcher. For this reason, the language structure of thesauri should receive more attention than has been the case so far. In this study, the intermediate function of the thesaurus has led to an investiga¬tion of the principles for structuring and representing informa¬tion in such a way that it corresponds to the cognitive structure assumed to exist in the information searcher. The model of inves¬tigation of cognitive representation is based on overt manifesta¬tions of concepts and conceptual relations as they emerge in the abstract language of titles of scientific documents. On the basis of this language structure, an algorithm has been developed for coding the relevant concepts by means of prepositions. The relevance of the concepts is judged with respect to a schema model containing the main components Problem, Method and Goal, assumed to represent the basic components of research itself. When applied to scientific titles, the assumption is verified. The algorithm generates the concepts corresponding to the compo¬nents, and assigns them to different data registers. An important result of the analysis is thus that the intermediate language structure of scientific titles displays a degree of abstractness suitable for automatic concept extraction, provided that an in¬formation- or cognition-oriented approach is employed in order for the data registers to be properly interpreted.