We present in this paper a new approach for the automatic generation of lexical structures from texts. This task is based on the strong hypothesis that simple statistical observations on textual usages can provide pieces of semantics about the lexicon. Using such "naive" observations only, we propose a (pre)-topological framework to formalize and combine various hypotheses on textual data usages and then to derive a structure similar to usual lexical knowledge databases such as WordNet. In addition, we also consider the evaluation problem for the resulting lexical structures; our evaluation strategy is organized as a multi-level evaluation process, measuring the fitting between a given reference structure and an automatically generated structure from two different points of view: intrinsic/structural and application-based. The evaluation strategy is then used to quantify the contribution of the new structuring approach with respect to the corresponding solution proposed by (Sanderson et al., 2000) on two case studies that differ on the domain and the size of the lexicon.