This study aims to explore the new opportunities opened by the application of sophisticated tools to indefinitely large text collections and by the new empirical perspective brought by the quantitatively based and qualitative approaches used in corpus linguistics. Word co-occurrences and collocations are focused in the study. The motivation for my attention to collocations is twofold. On the one hand, collocations are a key phenomenon in linguistics in their own right. On the other, they provide empirical evidence for background conceptual knowledge and their extraction can give an important contribution to the description of frame elements and in the disambiguation of word sense. The paper refers to the Firthian approach and to Minsky's and Fillmore's studies, observing the presence of persisting and non-persisting elements in collocations as well as in frames configurations. The relation between corpus data and frames elements is discussed and briefly illustrated in some case studies. (Rema Favretti Rossin)