Yea there may be a shift in usage hidden in those numbers. As this article laments, we can't use ngrams to measure the develpment of usage between said, replied, and retorted.