Words and Echoes: Assessing and Mitigating the Non-Randomness Problem in Word Frequency Distribution Modeling