Measuring Lexical Diversity in Texts: The Twofold Length Problem