Collocations (typical word associations like "to meet a condition", "to believe firmly", "highly controversial", "a deep concern") are pervasive in language. Due to their encoding idiomaticity, collocations are of paramount importance for text production tasks, for Foreign Language Learning as well as for Natural Language Processing applications such as machine translation. This thesis tackles the problem of automatic acquisition of collocations from text corpora, and proposes a methodological framework for their identification based on syntactic criteria. It is shown that (1) the results obtained are more reliable than those of standard methods based on linear proximity constraints, and (2) the syntax-based approach enables a more advanced treatment of these expressions, which is now largely absent in the related computational work: extraction of n-ary collocations (n>2), induction of syntactic patterns for collocations, Web-based extraction of collocations, and translation of collocations.