Three factors have been proposed to account for this asymmetry (see Brysbaert & Nazir, 2005, for a detailed overview). The first is perceptual learning. When reading text, the eye lands preferentially at certain positions in a word, referred to as the preferred landing position (Rayner, 1979). For long words in languages read from left‐to‐right, this position is somewhat to the left of the middle of the word (Ducrot & Pynte, 2002). Noting the similarity between the preferred landing position seen in text reading and the optimal viewing position observed for isolated words, Nazir (2000) proposed that the processing of isolated words is determined by how these words are typically fixated in normal (i.e., text) reading and that this is achieved via a form of perceptual learning. That is, a frequency‐sensitive learning mechanism operating on visuo‐orthographic representations of words. According to this account, the preferred landing position is not a consequence of readers aiming to optimize word identification processes (e.g., Legge et al., 1997). Rather, it is due to low‐level oculomotor constraints, and it is these same constraints that subsequently determine the optimal viewing position function that characterizes the processing of isolated words.
A rather different account of the asymmetric nature of the optimal viewing position function is hemispheric specialization. On this account, part of the observed asymmetry may be caused by the location of brain structures involved in processing printed words. Information falling on the right visual field of both eyes is initially projected onto the left hemisphere of the brain, whereas information falling on the left visual field is projected onto the right hemisphere. Lexical processing typically takes place in the left hemisphere. This means that information about letters left of fixation and initially sent to the right hemisphere must be transmitted through the corpus callosum to the brain regions dedicated to lexical processing and located in left ventral occipital cortex (Cohen et al., 2000). This detour takes time, and could therefore produce the observed asymmetry in the viewing position function since with initial fixations toward word beginnings more letters will be directly processed by the left hemisphere (Brysbaert, 1994).
Finally, the distribution of information within a word might result in the asymmetric function. O’Regan et al. (1984) suggested that the optimal viewing position might arise simply from a combination of changes in visual acuity and the distribution of information across a word, with word beginnings being more informative with respect to word identity than are word endings (see Clark & O’Regan, 1999, for a more detailed exploration of this hypothesis).
Brysbaert and Nazir (2005) argued that all three factors play a role in determining the shape of the viewing position function, along with visual acuity. Key evidence in favor of this stance is that the asymmetry does not completely reverse in languages read from right‐to‐left such as Arabic (Farid & Grainger, 1996) and Hebrew (Deutsch & Rayner, 1999). Instead, the viewing position function is more symmetrical in these languages. This suggests that hemispheric specialization is relevant, in combination with other factors. There is good evidence that the distribution of information within words is key and might even remove the need to appeal to a perceptual learning account. It is possible to shift the viewing position function by varying the way information is distributed across a word, for example, by comparing prefixed versus suffixed words (Farid & Grainger, 1996; Deutsch & Rayner, 1999) or considering the location of the most informative letters with respect to word identity (O’Regan et al., 1984) as well as the visibility of these letters (Stevens & Grainger, 2003). The fact that the most informative letters tend to be located toward word beginnings might explain why readers’ initial fixations tend to be located toward word beginnings (Legge et al., 1997).
Other visual factors are known to have an impact on reading fluency. Two such factors are print size and inter‐letter spacing. There is a critical print size for maximal reading speed beyond which there is no further gain (Chung et al., 1998): increasing print size in peripheral vision beyond this does not counteract the drop‐off in acuity. This is partly because increasing the size of text is accompanied by an increase in eccentricity. Similarly, there is evidence that small increases in inter‐letter spacing are beneficial for reading (e.g., Perea & Gomez, 2012; Zorzi et al., 2012), but larger increases in inter‐letter spacing eventually interfere with reading (e.g., Chung, 2002; Legge et al., 1985), as does smaller than normal inter‐letter spacing (Montani et al., 2015).
In sum, the position in a word where readers first fixate that word has a strong impact on ease of word identification, along with the effects of other more obvious visual factors such as print size and inter‐letter spacing.
Encoding letter‐order for word identification
The presence of anagrams in alphabetic languages forces attention to be paid to letter‐in‐word order. The issue here is just how much “attention” to letter‐order information is needed. At one extreme is the length‐dependent, position‐specific slot‐coding used in the interactive‐activation model (McClelland & Rumelhart, 1981) according to which the reader knows precisely which letter is at which position in a word of a given length. Although probably only applied for computational convenience, this coding scheme had the advantage of generating precise predictions with respect to effects of orthographic similarity on visual word recognition (see previous section). Nevertheless, a number of empirical findings challenge this view and point to the need for a more flexible letter‐position coding scheme. Much of this evidence comes from experiments using the masked priming paradigm (Forster & Davis, 1984; see Adelman et al., 2014, for a mega‐study) whereby target words are preceded by various types of word or nonwords primes. Taken together, findings from these experiments indicate a certain flexibility in the way an orthographic description of the stimulus (letter identities and letter positions) is matched with whole‐word orthographic representations in long‐term memory.
One key piece of evidence that forced a re‐consideration of how letter position information is encoded during orthographic processing was provided by experiments demonstrating effects of orthographic overlap between two stimuli of different length. One important finding, first reported by Humphreys et al. (1990), is that facilitatory masked priming can be obtained from orthographically related nonword primes that are not the same length as the target word (e.g., bvk as a prime for the target word black). Referred to as the relative‐position priming constraint by Grainger et al. (2006), this finding is important because it falsifies the simple length‐dependent, slot‐based letter position coding implemented in the original interactive‐activation model. The basic effect is well replicated (Peressotti & Grainger; 1999; Schoonbaert & Grainger, 2004) and extends to superset primes where primes (e.g., garbdfen) facilitate target word processing (e.g., garden) albeit with a cost of approximately 10 ms per inserted letter (e.g., Adelman et al., 2014; van Assche & Grainger, 2006). These priming effects are related to another phenomenon that illustrates the flexibility of letter position encoding. Participants in Bowers et al. (2005) found it hard to reject the word that as an item of clothing, arguably because the embedded word hat provided evidence that the stimulus did refer to an item of clothing. Once again, this finding is incompatible with length‐dependent slot‐coding of letter position information, and points to a greater degree of flexibility in the manner in which letter‐position information or letter‐order information is encoded during visual word recognition.
Perhaps the single key finding that drew attention to the fact that letter identities are not tied to a strictly length‐dependent position‐in‐word is the effect of small changes in the order of the letters in a word – so‐called transposed‐letter effects. Following earlier reports (e.g., Bruner & O’Dowd, 1958; Chambers, 1979), key findings using the masked priming paradigm (Perea & Lupker, 2003, 2004) are among the most replicated effects in experimental psychology. The standard finding is that primes formed by transposing two letters of a target word facilitate processing of the target words compared with primes formed by substituting the same letters (e.g., caniso – CASINO vs. carivo – CASINO, Perea & Lupker, 2004). It is now established that priming is greater when at least one of the transposed letters is a consonant (Lupker et al., 2008; Perea & Lupker, 2004); that priming diminishes as the distance between the two transposed letters increases