“…At the simplest level, if the algorithm is allowed to use a longer text segment around a seed word, a larger set of terms is likely to be measured. More interesting, in language usage, at least for English, the tendency is to avoid repeating a word in an adjacent sentence and to use a replacement term, such as a synonym (see, e.g., Beeferman et al, 1997). The relevancy scores of the words that are also seen in the one-sentence lists are, for the most part, higher in the three-sentence lists.…”