Figure 3.1 Orthographic processing as the central interface of the reading process, enabling mappings between vision and language for words, morphemes, and sublexical spelling‐sound correspondences.

The present chapter is concerned uniquely with the leftmost pathway in Figure 3.1. Bysbaert, this volume deals with phonology (the rightmost pathway), and Rastle (this volume) deals with morphology (the center pathway). Also note that the work to be covered here involves silent reading for meaning¹ and largely ignores the extensive theoretical (e.g., Coltheart et al., 2001; Perry et al., 2007; Seidenberg & McClelland, 1989) and empirical work on reading aloud. That said, within the framework of dual‐route theories of reading (Coltheart et al., 2001; Perry et al., 2007), adding a connection between orthographic words (also termed whole‐word orthographic representations) and phonological word forms in Figure 3.1 provides the necessary architecture for reading aloud and for phonological influences on silent reading (cf. the bimodal interactive‐activation model: Diependaele et al., 2010; Grainger & Holcomb, 2009). The processing of letters and words along the leftmost pathway in Figure 3.1 is thought to be performed by neural structures located in the left fusiform gyrus, dubbed the Visual Word Form Area (Yeatman, this volume), with these structures connecting to brain areas dedicated to the processing of phonology as well as higher‐level syntactic and semantic processing.

      So, what is orthographic processing? I argue that orthographic processing is best defined as the processing of letter identities and letter positions. This definition of orthographic processing, and the hypothesized fundamental role it plays during reading, is based on the premise that written words are primarily recognized via their component letters.

Letter‐Based Word Recognition

      Historically, the “word shape” hypothesis (i.e., words are recognized holistically rather than via their component letters) was once dominant in theories of word recognition and reading – a mistaken view that had a major influence on educational practice for the better part of the twentieth century. The popularity of this hypothesis was founded in Cattell’s (1886) observation that words are read aloud more easily than single letters (the “word superiority effect”), which lead to the following reasoning: How can we possibly read words via their constituent letters if it is harder to read individual letters than to read words? The solution to this conundrum was provided by theoretical advances (e.g., McClelland & Rumelhart, 1981) showing how a word can be identified from the combination of partial information available at the level of each of its constituent letters (see Grainger, 2018, for further discussion of the word superiority effect and its interpretation).

      Although there is some evidence that word shape information might influence skilled word reading in certain situations (e.g., Perea & Rosa, 2002; Perea et al., 2015), there is abundant evidence that skilled readers use information about abstract letter identities to identify words, along with information about letter positions. Together, letter identity and letter position comprise orthographic information. We know from experiments using masked repetition priming (Forster & Davis, 1984) that letter identities are abstract: The processing of word targets is not affected by whether or not the prime word is presented in the same case or different case (e.g., TABLE – TABLE vs. table – TABLE, Grainger & Jacobs, 1993; Perea et al., 2014; Vergara‐Martinez et al., 2015). Likewise, although overt reading is handicapped by mixed case presentation (e.g., tAbLE), an effect that is probably due to the perceptual grouping of same‐case letters (Mayall et al., 1997), mixed‐case primes are just as effective as same case primes (Perea et al., 2015). Another example here is our ability to read CAPTCHAs (Completely Automated Public Turing test to tell Computers and Humans Apart) – those highly distorted words we are asked to read when an internet site is checking whether we are a human or a robot. Hannagan et al. (2012) demonstrated masked repetition priming effects using CAPTCHA primes. This demonstrates that our ability to solve even extreme cases of shape distortion is achieved automatically, without resort to slow inferential processes (for an analogous finding with handwritten words, see Gil‐López et al., 2011).

      Having reviewed the evidence for letter‐based word recognition, we now need to understand how letters are identified and their positions encoded during reading.

       Letter perception

The seminal work of Oliver Selfridge (Selfridge & Neisser, 1960) laid the foundations for a cognitive theory of letter perception. In Selfridge’s “pandemonium” model, letter identification is achieved by hierarchically organized layers of feature and letter detectors. Support for such a hierarchical organization was provided at that time by neurophysiological studies of the cat visual cortex, and over the years, a general consensus has developed in favor of a generic feature‐based approach to letter perception. One key guiding principle here is that isolated letter perception is just a simplified case of visual object recognition (e.g., Pelli et al., 2006). Our knowledge of visual object perception, much of which has been derived from neurophysiological studies of nonhuman primates, should therefore inform our knowledge of letter perception in humans, as exemplified in the model presented in Figure 3.2. This is a blueprint for a model of letter perception (Grainger et al., 2008) adapted from a classic account of object recognition (Riesenhuber & Poggio, 1999; see Dehaene et al., 2005, for an extension of this approach to visual word recognition).

Evidence concerning the nature of letter features comes from studies using a method known as the confusion matrix. In a typical experiment used to generate a confusion matrix, isolated letters are presented in data‐limited conditions (brief exposures and/or low luminance and/or masking). Participants are asked to name the letters, and erroneous letter reports are noted. Error rates (e.g., reporting F when E was presented) are hypothesized to reflect visual similarity driven by shared features, and therefore an analysis of the pattern of letter confusions is expected to reveal the set of features used to identify letters. There are more than 70 published studies on letter confusability (Mueller & Weidemann, 2012). These have formed the basis of concrete proposals of lists of features for letters of the Roman alphabet, mainly consisting of lines of different orientation and curvature (see Fiset et al., 2008, for an alternative method for defining letter features, and Grainger et al., 2008, for a discussion of this technique).