Reading is an indispensable skill in our society and, thus, so much part of daily life that we do not realize how complex this skill is. The eyes fixate most words in the text being read, and typically only once. This implies that readers get a foveal glimpse of most words, and that an essential first stage of skilled reading behavior concerns the processing performed during that glimpse: the visual word recognition. When we read a single word, there are a series of complex cognitive processes involving computations at different levels of processing (e.g., perceptual, orthographic, phonological, morphological, syntactic, and semantic) during a few hundred milliseconds. Importantly, if there is one aspect of visual word recognition on which there is a general consensus among researchers, it is the crucial role of individual letters. It is generally assumed that the recognition of a printed word is mediated by the processing involving the letters that make up that word, at least in languages that use alphabetic orthography. To further our understanding about letter coding, this dissertation aimed examine how literacy characters the processes that build the bridge between the low-level stages of visual processing and the higher-level processing of words: orthographic processing. The orthographic processing refers to the encoding of information about the identities and position of the letters, thus allowing us to differentiate KISS from HISS or GOD from DOG. To that end, the thesis comprises three sections: (i) the emergence of orthographic processing, (ii) the resilience of letter detectors, and (iii) the methodological contributions to visual word recognition research. The section on the emergence of orthographic processing comprises three studies that examined the development of orthographic processing with literacy in the early years of reading acquisition. Importantly, with reading acquisition running its course, a flexible and powerful decoding system is built: readers can rapidly access the abstract representations of letters despite their format (a A a A a a a share the same abstract representation). In the section on the resilience of letter detectors, I present three studies that examined how visual distortion affects orthographic processing (i.e., the mapping from the visual input onto abstract word representations during normal reading). To that end, we focused on a single type of distortion: letter rotation. Finally, the last section includes a methodological contribution to the word recognition research. Specifically, it comprises two studies conducted to further comprehend the cognitive underpinnings of masked priming, the gold-standard tool for studying the very first moments of visual word recognition. 1.The Emergence of Orthographic Processing One of the central landmarks of learning to read is the emergence of orthographic processing. The encoding of letter identities, understood as the mapping of visual input to abstract letter representations, would occur in the first two years of reading acquisition (Jackson & Coltheart, 2001; Gomez & Perea, 2020). However, it is still not clearly understood when and how the encoding of letter order and position arises. To shed light on this issue, the present section focuses on the emergence of these processes via the examination of location-invariant processing (responsible for encoding the order of letters) and location-specific processing (responsible for encoding the position of letters), and the factors that may influence it. Pinpointing the factors that influence the emergence and development of orthographic processing is crucial, as it allows us to identify children in need of pre-literacy intervention and, consequently, to minimize eventual reading difficulties. With this purpose in mind, we conducted the study entitled Which factors modulate letter position coding in pre-literate children? Our goal was to scrutinize the precursors of letter order coding. Importantly, better readers encode letter order more accurately than the worse readers (see Gómez et al., 2021; Pagán et al., 2021). Thus, it is crucial to shed light on the roots of letter-order processing, defining its precursors. To that end, we examined the potential links between the encoding of letter order in pre-literate children (via the examination of their performance in differentiate transposed-letter pairs and identity pairs: TZ-ZT vs TZ-TZ) with the abilities related to early reading acquisition (basic cognitive processes: attention, sensation, perception, and memory). Results showed that the pre-literate children who best differentiated between transposed letter (TZ-ZT) and identity (TZ-TZ) pairs in a same-different task were those with higher scores on basic cognitive processes. Thus, basic cognitive skills shape the ability to encode serial order in letter strings. Importantly, research with skilled readers has shown that, once reading is mastered, the rank of the letters changes from being mere visual objects to enjoying a specific status. For instance, when presenting a string of symbols (£§?%@) individuals fixate in the middle of the string. However, when a string of letters is presented (FGJGM), they fixate also in the exterior letters, especially in the first letter of the string (Tydgat & Grainger, 2009). This phenomenon is due to the location-specific processing, attained by adapting the mechanisms of visual object processing to the constraints of visual word processing. Moreover, the encoding of the order of letters becomes more flexible, leading to the location-invariant processing: FGJM is easily confounded with FJGM. This phenomenon is usually captured when using a same-different matching task in which participants have to decide if pairs of stimuli are the same or different. Results show that “no” responses to a transposed-letter pair (FGJM-FJGM) are slower and more error-prone than the responses to the replacement-letter control (FGJM-FPCM). This effect, known as the transposition letter (TL) effect, also occurs with strings composed of symbols, numbers or unknown letters, but the effect is are substantially larger for strings of letters. The presumed explanation for the dissociation letters vs. other visual objects is that there is an orthographic-specific mechanism used to encode location invariant letter-in-word-order in addition to general positional uncertainty. Crucially, this orthographic-specific mechanism has been posited to emerge with literacy acquisition (Dandurand et al., 2010; Duñabeitia et al., 2014, 2015). Nevertheless, the empirical data on the emergence of location-invariant processing is very scarce and inconclusive. The reason is that comparing results for pre-literate and literate children lead to interpretative issues (i.e., there are variability between groups). One option to overcome this limitation is to design a laboratory analog of learning to read with adult participants. This was the strategy followed in the second and third studies of the present section. In the second study, Does orthographic processing emerge rapidly after learning a new script? (Fernández-López et al., 2021), we tested whether orthographic processing, examined via location-invariant and location-specific processing, emerged quickly with literacy. To that end, we conducted a same-different task and a target-in-string identification task with two unfamiliar scripts (pre-training experiments). Across six training sessions, participants learned to fluently read and write one of these scripts. The post-training experiments were parallel to the pre-training experiments. Results showed that the magnitude of the transposed-letter effect in the same-different task and the serial function in the target-in-string identification tasks were remarkably similar for the trained and untrained scripts. We did find that responses to same trials in the same-different task were faster and more accurate in the post-trained phase for the trained script, but not for the control script. Thus, location-invariant and location-specific processing do not emerge rapidly after learning a new script. However, it seems that the learning-to-read training induced some rudimentary specialization for the letter strings that made easier the identification of same pairs. The emergence of orthographic processing probably needs much more time to emerge, or, alternatively it might require experience with specific orthographic structures. In the third study of the present section, A letter is a letter and its co-occurrences: Cracking the emergence of location-invariance processing, we tested whether location-invariant processing emerged rapidly after the exposure to orthographic regularities—bigrams—in a novel script. Results in a same-different task showed that responses to pairs with a letter transposition in a frequent bigram were less accurate than those with a letter transposition in an infrequent bigram. This finding suggests that location-invariance emerges rapidly after exposure to orthographic regularities. To sum up: Reading readiness emerges early in development with the adaptation of the mechanisms of visual object processing. For that reason, early detection of deficiencies in the visual analysis of the input is crucial to prevent later reading difficulties. Notably, we have added to the literature that orthographic processing does not emerge rapidly when learning to read, but, as a prelude, some non-specialized processes are fine-tuned to guide the later functional reading progress. 2.The Resilience of Letter Detectors Importantly, with reading acquisition running its course, a flexible and powerful decoding system is built: readers can rapidly access the abstract representations of letters despite their format. However, the empirical evidence on the resilience of the word recognition system to perceptual distortion is still scarce, probably because it entails many different elements (see Vergara-Martínez et al., 2021, for discussion). In this section, I present three studies that aimed to examine the tolerance of the word recognition system to visual distortion in a single dimension: the rotations of letters. An added value of studying rotation is that leading models of visual word recognition make explicit predictions about how letter rotations affect word reading. The SERIOL model of word recognition (Whitney, 2001, 2002) assumes that “input levels to letter units are reduced for rotated stimuli” (p. 119, Whitney, 2002), increasing gradually the time required to encode a letter string. Instead, the Local Combination Detectors model (Dehaene et al., 2005) offers an all-or-none account on how letter rotation affects letter and word processing. It postulates a boundary of 40°-45° beyond which letter detectors would not be able to efficiently cope with rotations—but there would be little cost for letter rotations below that boundary. Nevertheless, the assumption of a letter rotation cost has been recently put into question by a number of studies that have shown that readers can cope with rotations up to 180° from the initial access to orthographic representations when the word is wholly rotated (Perea et al., 2018; Yang and Lupker, 2019; see also Benyhe & Csibri, 2021). Nevertheless, when a word is entirely rotated, readers can mentally rotate the stimulus as a whole, with one movement of mental rotation. By contrast, if we rotate the individual letters within words, individuals would need more than one rotation movement. Moreover, the rotation of the individual letters involves the disruption of trans-letter features. Thus, this manipulation allows us to strictly examine the models predictions. For this reason, in the three studies of this section, we rotated the words in a letter-by-letter basis. Our main aim was to examine the resilience of letter detectors to different rotation angles and, hence, test the predictions of the SERIOL (Whitney, 2001, 2002) and Local Combination Detectors model (Dehaene et al., 2005). In the first study, How resilient is reading to letter rotations? A parafoveal preview investigation, we tested the resilient of letter detectors to rotations while reading. We conducted a reading experiment with sentences in uppercase using Rayner’s (1975) gaze-contingent boundary change paradigm in which we examined whether identity parafoveal previews were more effective than unrelated parafoveal previews. The previews were rotated at 15°, 30°, 45°, or 60°, but the rest of the text was presented with letters in their canonical orientation. In the second study, On the time course of the resilience of letter detectors to rotations: A masked priming ERP investigation, we examined the temporal course of the effect of rotation of the words’ constituent letters in a masked identity priming paradigm. We focused on testing the prediction of the LCD model: “letter detectors should be disrupted by rotation (> 40°)” (Dehaene et al., 2005, p. 340). We conducted a masked repetition priming lexical decision experiment in which the individual letters of the identity/unrelated primes we rotated 0°, 45°, or 90°. We also recorded Behavioral and Event-Related-Potentials (ERP). In the third study, Letter rotations: Through the Magnifying Glass and What Evidence Found There, we conducted three experiments that scrutinized the cost of letter rotation on lexical access and tested the predictions of the LCD and SERIOL models concerning the resilience of letter detectors to relatively moderate angles of letter rotation. Experiments 1 and 2 employed four rotation angles (0°, 22.5°, 45°, 67.5°) in lexical decision and semantic categorization tasks. In Experiment 3, we focused on four moderate rotation angles at or below 45° (22.5°, 30°, 37.5°, 45°). Considering the outcomes of the three studies and building up a comprehensive picture, the main ideas that can be drawn are summarized as follows: (i) the cost of letter rotation originates very early, in the first stages of word recognition. Importantly, (ii) even moderate rotations lead to a cost. Nevertheless, (iii) this cost is easily overcome, and the system can handle rotations up to 45° in the initial moments of processing. Importantly, (iv) to cope with stepper rotations, individuals seem to engage extra top-down (lexical) feedback. In sum (v), we found that the cost elicited by rotation angle increases smoothly at angles below 45°, as proposed by the SERIOL model. The letter detectors reach a resilience limit at around 45° (consistent with the LCD model), and the cost becomes dramatic at steeper angles. Therefore, (vi) the most suitable option is combining the assumptions from SERIOL and LCD models to build a more comprehensive explanation of how rotation angle, and more generally, visual distortion, hampers word recognition. 3.The methodological contribution to the word recognition research The more significant breakthrough in studying the reading process started around the late 1950s, coinciding with what is often called the cognitive revolution (see Pollatsek & Treiman, 2015, for review). This revolution broke with the behaviorist approach and led to many ingenious new techniques for studying reading and word identification processes, including the lexical decision task (Rubinstein et al., 1970, 1971a, 1971b) and the masked priming procedure (Forster & Davis, 1984), considered “the major methodological innovation of the last two decades” (Grainger, 2008). Notably, the combination of the lexical decision and the masked priming has led to much empirical work investigating letter and word processing. Nevertheless, the technique requires further examination to build firm conclusions on the outcomes obtained with masked priming. Here, I present two studies that attempted to understand the cognitive underpinnings of masked priming and one of its variants, the sandwich priming. In the first study, entitled Can response congruency effects be obtained in masked priming lexical decision? (Fernández-López et al., 2019), we aimed to test whether participants apply the same instructions to the primes and targets in the masked priming lexical decision task by examining the response congruency effect with unrelated primes. The response congruency effect in lexical decision can be defined as the difference in response times between a congruent trial (i.e., when the prime elicits the same response as the target (bright-WINDOW: “word”; geuzel-POLCIR: “nonword”) and an incongruent trial (geuzel-WINDOW: “word”; bright-POLCIR: “nonword”). Importantly, the examination of this effect allows us to scrutinize “the state of the lexicon after the prime stimulus has been processed up to an early stage” (Loth & Davis, 2010b, p. 174). Specifically, we examined whether there were differences in the influence of a masked unrelated prime that varied in wordlikeness: primes could be a high-frequency word (radio), a low-frequency word (beret), a pseudoword (geuze), or a consonant string (xfgtp). We developed three scenarios that varied in the difficulty of the word/nonword discrimination: orthographically legal nonwords matched in subsyllabic elements with the target words (Experiment 1), orthographically illegal nonword foils (Experiment 2), and orthographically legal nonwords with no neighbors (Experiment 3). We found that when the foils were orthographically legal, lexical decision responses were not affected by the lexical status or wordlikeness of the unrelated primes, as predicted by the Bayesian Reader model and the selective inhibition hypothesis in interactive-activation models. When the foils were illegal, consonant-string primes produced the slowest responses for word targets and the fastest responses for nonword targets. The Bayesian Reader model can capture this pattern assuming that participants in Experiment 2 were making an orthographic legality decision (i.e., anything legal must be a word) rather than a lexical decision. Despite the numerous advantages of the masked priming technique, it is not sensitive enough to detect subtle manipulations (e.g., lght-LIGHT vs. liht-LIGHT; Grainger, 2008) or examine the letter and word recognition processes in special populations. The reason is that, because of the short stimulus-onset asynchrony between prime and target, the magnitude of masked priming effects is small, in the range of 10-50 ms in behavioral studies. To overcome this limitation, Lupker and Davis (2009) introduced the sandwich technique. This technique is a variant of the standard masked priming in which the addition of a pre-prime (identical to the target) magnifies the size of priming effects. Nevertheless, the specific mechanisms at play and how they differ from the standard technique, have not been fully elucidated (Lupker & Davis, 2009 vs. Trifonova & Adelman, 2018). In the second study of this section, Unveiling the boost in the sandwich priming technique (Fernández-López et al., 2021), we aimed to fully characterize the sandwich technique by examining whether the priming effects in the sandwich technique are simply magnified or, instead, the underlying processes are fundamentally altered. To that end, we compared the sandwich and standard techniques by examining the RTs and their distributional features (delta plots; conditional-accuracy functions), comparing identity vs. unrelated primes. Results showed that the locus of the boost in the sandwich technique was two-fold: faster responses in the identity condition (via a shift in the RT distributions) and slower responses in the unrelated condition. The shift in the RT distributions for identity pairs in the sandwich and standard technique strongly suggests that the extra time from the pre-prime is not altering the underlying processes beyond encoding. Nevertheless, for unrelated primes, the pre-prime in the sandwich technique hampered the processing of unrelated prime-target pairs. This finding warns researchers to be cautious in interpreting some of the across condition priming effects in subtle manipulations in the sandwich technique, as these effects may derive not from facilitation in the related condition but rather from inhibition in the unrelated condition. Three highlights can be drawn from this section: when choosing the baseline, it does not matter the wordlikeness of the unrelated primes (as long as the task employs orthographically legal foils); the nature of the decision required by the task modulates masked priming effects; when employing the sandwich technique, researchers should use both the identity and unrelated conditions as controls.
