It has been suggested that the visual word identification system identifies recurrent letter clusters (n–grams) as a bridge between letters and words. Some evidence may suggest, rather indirectly, that n–gram coding is a more general visual mechanism, used beyond reading to extract statistical regularities in smaller visual units to build larger visual units. Here we test this hypothesis directly, and explore the boundaries of the phenomenon—what type of visual object, if any, fails n–gram based statistical learning?
In a non–linguistic version of a well established paradigm in reading, we asked adult participants to learn a set of novel objects made up of smaller parts. We show that, similar to what happens with (pseudo)reading material, participants have a hard time discarding objects that they’ve never seen, but comply with the statistical pattern of the smaller parts. This suggests that n-gram coding is a general mechanism used by the brain to learn about the visual environment. We test this mechanisms across different types of visual objects and consider the results in the context of general theories of visual learning, and learning to read in particular.