Formulaic speech has been notoriously difficult to define and identify despite its crucial importance to native-like fluency and idiomaticity. In this article, I introduce a way of identifying phraseological units in a running text. I am interested in recurrent fragments like "charged with crimes against humanity" in texts which involve multiple word collocations in a "fuzzily fixed" lexico-syntactic frame. I suspect these kind of phraseological fragments not only add to the fluency and idiomaticity of texts, save text production time, but actually constitute milestones in the generation of text and sentences. No currently well known corpus seems large enough to provide adequate instances of prefabricated chunks like this for closer investigation. It is proposed here that Internet as a gigantic corpus and a search engine like Google can help identify and retrieve these phraseological units for linguistic research and language teaching and learning. (Contains 12 figures and 12 tables.)