In general, a power function like this, very well describes the relationship between word rank and word frequency: f(x) = predicted word frequency = constant * rank number^-1. (The constant varies from text to text and is a bigger number for a long text and a smaller number for a text with fewer unique words).
However, this typically does not hold for the 10-30 highest ranked words. Do you have a suggestion - apart from skipping the first occurrences as outliers? For your background: It seems that Zipf's Law does not cover the first part of the ranked vocabulary of a specific text.
