>That probability is n_word/total_words the corpus.

Thanks. It's been 15 years since I implemented it. Guess I failed the interview.