A Crucial Parameter for Rank-Frequency Relation in Natural Languages