There are several hits for 'Internet' in 1853, and hundreds for 'email' in the 1890s. To avoid that, maybe they could weight the OCR recognition based on a word's first occurrence in dictionaries (backdated a decade or two).