One extreme approach would be the “bag of words” model, where every word is entirely independent. Reading a word in this representation will have no bearing on any other. However, this approach seems lacking, or else teaching students to read would simply consist of memorization of a significant subset of the English language, word by word. At the other extreme, one might suggest that every pair of words has some dependence. This implies that every time a student reads a word, he becomes (marginally) better at every other word in the language. This alternative approach also seems counterintuitive, as the authors feel that a student, both before and after reading the words ‘dog’ and ‘cat’ for the first time, will fail equally quickly on a word like ‘supercalifragilisticexpialidocious’. However, we could reasonably assume that the student would perform more accurately on ‘dogs’ and ‘cats’.
— ❐ What's in a word﹖ Extending learning factors analysis to model reading transfer