story
If thought (as expressed in language) is just probabilistic pattern matching, then how did we develop our own training data from scratch?