We present here an approach that automatically extracts semantic relations between concepts/words. To this end we rely on the semantics of words (lexicon) and their roles, that is, syntax. The extracted lexical and the syntactic information is based on the so-called distri-butional theory (Harris, 1954 ; Firth, 1957) : words occuring in the same contexts tend to have similar meanings. The theory is nowadays widely accepted and used, among other things, for computing the similarity betwen text fragements (similarity approach). We extended this intuition to extract semantic relations using lexical and syntactic distribution, the former being quite common. We have also extended this approach to make it feasable for the extraction of syntactic features : word pairs being linked by the same kind of semantic relation are used in similar (if not identical) syntactic patterns, or, put differently, word pairs linked via similar semantic relations are expressed via similar syntactic patterns. rnWe used lexical semantics, i.e. the semantics of the constituent words, linked via a given syntactic structure for two reasons: (a) the extraction of semantic relations based exclusively on the distribution of words requires less resources than other state-of-the-art approaches, (b) it allows to disambiguate extracted ambiguous words via the distribution of syntactic features. Our approach is weakly supervised, hardly language dependent and completely independent of huge resources like WordNet (henceforth WN), Framenet, or else. rnWe illustrate our approach for two tasks of language production: (a) lexical access and (b) discourse structuring. In the first case the computer is meant to help authors to overcome the â€—tip of the tongue problemâ€˜, while in the second case the goal is to assist authors to organize their thoughts (or messages) to produce coherent (i.e. connected) discourse. In both cases the system relies on a semantic network automatically created. To allow for this we extracted from a text (corpora) its core elements (words), inferring then automatically the link(s) holding between them. Since this is a very complex task, we have focused here only on a subset of links: synonymes, meronyms, (i.e. part-of relations : X is part of Y, X has Y, etc.), and hypernyms (X is more general than Y). rnTo evaluate our results we used standard data sets, comparing our work against state of the art approaches. The results show that our approach is at least as good as the major competitors.