I took a stab at using google's n-grams library to look at potential sentences if the code was an initialism. It generated too many nonsense values at the value of 'n' that I picked (driven purely by download size concerns really) and any kind of names or codes or abbreviated language wouldn't have worked well with that approach.

I think at some students of the mentioned Prof. Abbot looked at the letter distributions and found the highest correlation with an english language initialism.