Abnormal really means that it has a probability less than some threshold. So you might have a transition that has a probability of less then 0.1, depending on your situation, you might consider that abnormal even thought that transition does exist. Also a non existent transition would be considered "abnormal" because the probability is 0 in your model.

I've been thinking about doing a part 2 with text analysis and a cool use case would be to detect "abnormal" text from an author because he/she uses strings of words that have a low probability in the rest of their text.

Hope this helps and I'm glad you like the post.