Computing efficiently the closeness of word sets in natural language texts

We consider some search problems which have applications in statistical text analysis and natural language processing. Given two sets of words A and B, we propose a statistical, corpus-based measure of the ``closeness'' between A and B in texts. Our proposed measure involves the search, throughout a text corpus, of the words in A and B, under the restriction that these words should co-occur within a given maximum distance n. We address the problem of efficiently computing this closeness measure and present algorithms for it.