This paper reviews the problem of characterizing documents and
suggests a technique based on the systemic functional theory to select the
relevant information for indexing and categorizing. The approach is called
the systemic feature selection (SFS) and differs from the widely used
techniques in the information retrieval and classification domains. We
provide an overall overview of the processing steps applied to the document
content before its categorization. This later is based on adaptive resonance
theory. The paper concludes with a report on the first experiments using SFS.