Automatic extraction of meaningful units from corpora

In this article, we will reconsider the notion of a word as the basic unit of analysis in language and propose that in an information and meaning carrying system the unit of analysis should be a unit of meaning (UM). Such a UM may consist of one or more words. A method will be promoted that attempts to automatically retrieve UMs from corpora. To illustrate the results that may be obtained by this method, the node word ‘stroke’ will be used in a small study. The results will be discussed, with implications considered for both monolingual and multilingual use. The monolingual study will benefit from using the British National Corpus, while the multilingual study introduces a parallel corpus consisting of Swedish novels and their translations into English.