This is not a "homework", is a project in language generation. There are not specific requirements, we have to find ways to generate natural phrases, idioms, expressions etc. This was one of the ideas: train a large file and find those sentences where the co-occurrence of n words occur in n specific gap. Find the most frequent offset and retrieve the sentence to check in which context they appear. The insistence of hashes is dumb, I just wanted to prove my point... somebody that validate my idea that hash was the only solution..:-(
Sorry that I did not provide more information before, I only wanted to keep it simple. Thanks