It should be an improvement to use CountMinSketch as an index stat for text indicies and then for terms that are longer than the "n" for the n-gram we can choose the n-gram within the term that has the smallest estimated cardinality.