Variable Vulnerability of Words to Visual Impairment: Exploring Grain-Size Effects in Reading

Languages vary in the grain size used to map orthography onto phonology, but the use of different grain sizes varies also within a language. A core question is what kind of information the reading system is using to map orthography onto phonology. A unit that has received particular attention in recent years is the grapheme. Graphemes are defined as written representations of phonemes, and they can be composed of multiple letters. If graphemes are functional reading units, then a word containing a multi-letter grapheme (digraph) should be read more accurately than a word without, given the same kind of perceptual noise to impair the orthographic input. We trained a neural network model to map English orthography onto phonology. Impairing the model revealed a distinction in performance with words with digraphs less affected by damage than control words. A behavioral experiment confirmed the prediction of the model. We discuss these results in light of a psycholinguistic grain size theory of reading.