On the Proper Treatment of Tokenization in Psycholinguistics