The current approaches to the analysis of natural language text are not viable for documents of unrestricted scope. A global text analysis system is proposed designed to identify homogeneous text environments in which the meaning of text words and phrases remains unambiguous, and useful term relationships may be automatically determined. The proposed methods include document clustering methods, as well as comparisons of local document excerpts in specified global contexts, leading to structured text representations in which similar texts, or text excerpts, are appropriately linked.