RESEARCH DOSSIER
Visual Word Recognition and Orthographic Processing
Abstract. Visual word recognition transforms visible print into access to learned linguistic forms through the identification and arrangement of letters or characters, graded activation of orthographic neighbors, interaction with phonology and meaning, and decisions shaped by task and reader experience. This fundamental research monograph critically synthesizes the historical development of interactive-activation, distributed learning, dual-route, overlap, and spatial-coding theories; experimental evidence from word identification, lexical decision, naming, masked priming, eye tracking, and electrophysiology; ventral occipitotemporal reading networks; and cross-language and bilingual findings. The principal synthesis distinguishes orthographic representation from lexical response criteria, flexible letter-position coding from indiscriminate visual similarity, and the organization of skilled recognition from its acquisition. The evidence is evaluated through a critical narrative review rather than a preregistered systematic review or meta-analysis. Open problems include representational generality across writing systems, temporal coordination of orthography and phonology, individual differences, and causal tests of competing computational models.
Keywords: visual word recognition; orthographic processing; letter position coding; orthographic neighborhood; lexical competition; masked priming; lexical decision; word naming; visual word form area; ventral occipitotemporal cortex; eye movements in reading; parafoveal preview; word frequency; contextual diversity; interactive activation; spatial coding model; overlap model; distributed reading models; dual route cascaded model; electrophysiology; bilingual visual word recognition; Artificial Era
Introduction: Printed Words as Learned Visual-Linguistic Objects
Visual word recognition is the process by which a reader extracts letter identities and their arrangement from a visible string, activates learned representations of written words, and makes associated linguistic information available for judgment, pronunciation, or interpretation. Orthographic processing denotes the transformations that support this achievement: sensitivity to graphemes, letter position, permissible sequences, lexical similarity, and the statistical structure of a writing system. The research problem is how readers recognize familiar printed forms despite changes in typeface, case, viewing position, neighboring strings, and task demands. The same letters can support distinct words when reordered, while approximate matches can activate the same lexical candidate. These observations require an account of perceptual tolerance constrained by discriminative precision Grainger & Jacobs (1996) Gomez et al. (2008) Carreiras et al. (2014).
This dossier focuses on the organization of skilled recognition at the printed-word level. It treats recognition as a cascade spanning visual feature extraction, orthographic coding, lexical competition, activation of phonology and meaning, and response selection. The main experimental evidence derives from isolated word identification, lexical decision, naming, masked priming, eye movements, and brain-based measures. Studies of learning to read are incorporated when they explain how mature orthographic representations arise; instructional intervention and developmental disorder classification remain distinct research objects. Similarly, sentence comprehension supplies context and parafoveal constraints but requires theories beyond isolated visual word identification Rayner (1998) Grainger & Holcomb (2009) Ziegler & Goswami (2005).
This work is a critical narrative synthesis of published experiments, computational proposals, and integrative reviews. It reports no original experiment, pooled effect estimate, or registered systematic-search procedure. Its analytic method is to compare the dependent variable and manipulation in each paradigm, identify assumptions that connect behavior to proposed mechanisms, and evaluate where convergent behavioral, neural, and cross-language evidence supports a mechanism. Particular attention is given to the difference between recognition accuracy, the dynamics of evidence accumulation, and the decision rule applied to a stimulus. Model agreement with a reaction-time pattern is informative only alongside performance on independently diagnostic manipulations Balota et al. (2007) Yap et al. (2012) Grainger & Jacobs (1996).
Historical Origins: From Word Superiority to Interactive Models
One founding problem was the word-superiority phenomenon: letters can be identified more accurately when embedded in familiar words than in comparable unstructured strings under selected brief-presentation conditions. The result challenged the assumption that recognition always proceeds from isolated, complete letter identifications toward a word as a final independent stage. McClelland and Rumelhart's interactive-activation model formalized feature, letter, and word levels linked by excitatory and inhibitory interactions. A partially activated word can feed back to support constituent letters while competing word candidates inhibit one another. The theory gave testable predictions about the relation between stimulus evidence and lexical context instead of assigning all contextual influence to deliberate guessing McClelland & Rumelhart (1981) Grainger & Jacobs (1996).
The next decades expanded the explanatory target beyond recognition of individual letters. Speeded naming, lexical decision, priming, and patterns of reading aloud raised questions about how orthographic knowledge interacts with pronunciation and lexical familiarity. Forster and Davis demonstrated repetition priming and an interaction with lexical frequency in experiments on lexical access, thereby showing why a single, invariant search-order account was insufficient for all response patterns. Seidenberg and McClelland constructed a distributed model in which mapping from orthographic input to phonological output emerged through learning across shared representational units. The research tradition became increasingly computational: claims about routes, representations, and decision thresholds could be evaluated against simulations of several tasks Forster & Davis (1984) Seidenberg & McClelland (1989) Coltheart et al. (2001).
Historical models remain central because each makes assumptions that can be differentiated experimentally. Localist architectures treat lexical candidates as identifiable processing units, while distributed accounts encode regularities across weighted representations. Later letter-position models loosened the rigid slot assignments assumed in early interactive schemes, and neural investigations connected behavioral regularities with left ventral occipitotemporal cortex. These developments did not produce one final model that subsumes every language and task. They shifted the central scientific question from whether words influence letter perception to how visual, orthographic, phonological, semantic, and decision processes are coordinated in time and learned over experience Whitney (2001) Davis (2010) Price & Devlin (2011).
Definitions, Representations, and Operational Boundaries
A grapheme is a writing-system unit that can correspond to a sound or other linguistic distinction; a printed character is its visible realization, while a letter is one kind of orthographic unit. Orthographic identity can therefore be considered at several levels: visual features, letter or character category, relative position, sublexical sequences, and whole-word form. A lexical representation connects familiar written forms to linguistic knowledge. Orthographic neighborhood often refers to words differing by one substituted letter in the same position, although contemporary research also analyzes addition, deletion, and transposition similarity. Operational definitions matter because different neighborhood metrics can rank the very same target differently and can yield different predictions for interference or facilitation Grainger & Jacobs (1996) Gomez et al. (2008) Davis (2010).
Recognition is typically operationalized through correct identification, word-versus-nonword responses, naming latency, eye-movement measures, or electrophysiological changes. Lexical decision asks whether a displayed string belongs to the language's word inventory, but the observed response incorporates criterion setting and response execution. Naming measures access to pronunciation and articulatory initiation; perceptual identification more directly constrains identification of briefly presented alternatives; eye movements index allocation of visual sampling during reading. These tasks share representational substrates without isolating identical computations. Evidence for a lexical effect in one task therefore requires a process-level explanation before it is generalized to another Grainger & Jacobs (1996) Coltheart et al. (2001) Rayner (1998).
The account also distinguishes orthographic familiarity from literal visual memorization. A skilled reader can recognize a word in different fonts and letter cases because learned abstract mappings preserve relevant identity across many visual appearances. Yet familiarity can be graded: words encountered in diverse contexts, rare orthographic combinations, and unfamiliar letter sequences differ in ease of processing. The goal of explanation is to specify when the system relies on generalized letter and sequence knowledge, when a particular lexical form is decisive, and how accumulated experience sets the strength of these representations. Such distinctions guide the interpretation of models trained on corpora and of experimental measures drawn from lexical databases Adelman et al. (2006) Balota et al. (2007) Grainger et al. (2008).
Visual Feature Extraction and Abstract Letter Identity
Printed-word processing begins with a spatially distributed visual signal whose precision varies with contrast, eccentricity, crowding, exposure duration, and attention. An identifying system must relate those signals to stable letter or character categories and then preserve information about their order. Case invariance demonstrates one form of abstraction: a written word can be recognized across uppercase and lowercase forms despite changes in visible contours. This property is best understood as learned tolerance within a visual-linguistic network rather than evidence that initial processing contains no visual detail. The relative contributions of retinal acuity, display typography, and learned character familiarity should be manipulated separately Grainger et al. (2008) Dehaene & Cohen (2011) Price & Devlin (2011).
Feature-based accounts predict that visual similarity can influence confusability before an exact lexical form is fully activated. Interactive-activation architectures allow partially supported letter categories to gain or lose activation through their compatibility with whole words. A crucial methodological challenge is disentangling physical resemblance from orthographic resemblance: visually similar letters may share contours without forming similar lexical strings, and two strings may be orthographically related despite differences of typeface. Experiments using case changes, symbol controls, different exposure durations, or carefully matched character sets can test which level supports an observed advantage. Language-specific grapheme inventories further complicate any assumption that the same visual unit inventory is shared across scripts McClelland & Rumelhart (1981) Grainger et al. (2008) Carreiras et al. (2014).
At the cortical level, specialization for print is acquired through reading experience. The left ventral occipitotemporal system responds reliably to written strings, although its responses participate in interactions with other visual and language regions. Cohen and colleagues' functional-anatomical analyses helped establish the visual word form area as a relevant node, while subsequent accounts debated how selective its activity is to orthographic features and how much it depends on feedback. A measurement of regional activation identifies participating tissue under particular conditions; it does not by itself demonstrate a sequence of computations or the existence of a dedicated word-image store. The strongest accounts connect anatomical responses to behavioral contrasts and patterns of reading impairment Cohen et al. (2000) Dehaene & Cohen (2011) Price & Devlin (2011).
Coding Letter Order: Slots, Gradients, and Flexible Positions
Word identity depends on order, yet recognition tolerates some location uncertainty. Transposed-letter strings can be perceptually close to their base words, showing that early processing cannot be fully described by perfect, independent letter-to-slot assignments. Andrews demonstrated interference from familiar transposition competitors in naming and lexical decisions, and Perea and Lupker used masked associative priming to probe the activation produced by transposed-letter nonwords. Such effects constrain the representational structure of a model because adjacent transpositions and substitutions can have different consequences even when they affect similar numbers of characters Andrews (1996) Perea & Lupker (2003).
Whitney's SERIOL model proposed serial-order coding through graded activation and temporal structure, whereas Gomez, Ratcliff, and Perea's overlap model treated a letter's position as a probability distribution extending into nearby positions. Davis's spatial-coding model offered another account of relative position and graded overlap while retaining mechanisms for lexical identification. These theories disagree in architecture, yet they share a commitment to explaining perceptual similarity through flexible coding rather than exact visual identity alone. The overlap model was evaluated against forced-choice identification data and showed that its small set of position parameters could capture numerous observed conditions. Fit to those conditions supports explanatory utility, while novel item types and languages remain necessary tests of generalization Whitney (2001) Gomez et al. (2008) Davis (2010).
Position coding is shaped by word length, repeated letters, position in the string, and the distribution of orthographic patterns. A model that explains adjacent transpositions may still struggle with insertions, omissions, or words containing repeated characters; one optimized for alphabetic left-to-right print may need modification for other scripts. Experimental designs should contrast the same number of shared letters under different positional arrangements, match lexical frequency and neighborhood, and separate prime-target similarity from response bias. Flexible order representations provide a compelling organizing principle precisely because they produce predictions across multiple manipulations, rather than a single intuitive demonstration that readers can recognize scrambled words Gomez et al. (2008) Davis (2010) Carreiras et al. (2014).
Orthographic Neighborhoods and Lexical Competition
A printed word activates candidates that share some of its orthographic structure. The common substitution-neighborhood measure counts words obtainable by changing one letter at the same position, whereas other measures admit transpositions, insertion, or deletion. Neighborhood density can therefore be related to familiarity with shared orthographic patterns as well as to competition among specific lexical candidates. A dense neighborhood may make a string appear wordlike, yet simultaneously create several plausible lexical identities that must be discriminated. The net behavioral effect depends on candidate frequency, decision demands, and timing. Models of multiple lexical read-out criteria were developed partly to explain why related strings facilitate identification under some conditions and interfere under others Grainger & Jacobs (1996) Davis (2010).
Orthographic competition is especially diagnostic when a higher-frequency neighbor competes with a lower-frequency target. The target may gain activation from shared sublexical information while losing discriminative advantage at the word level. The logic can be tested by holding target frequency and length constant and manipulating competitor frequency, neighborhood density, or the type of letter perturbation. Lexical decision responses also depend on whether an experiment's nonwords resemble real words, because a strategy based on overall familiarity will misclassify increasingly plausible distractors. Consequently, a neighborhood effect is never interpreted without the stimulus distribution and the read-out process that produces the response Grainger & Jacobs (1996) Andrews (1996) Gomez et al. (2008).
Competing accounts differ over whether orthographic similarity is implemented through whole-word nodes, distributed patterns, or an interaction of sublexical and lexical codes. Graded letter-position representations can explain why neighbors differing by transposition are unusually confusable, but a complete account also needs learning and frequency-sensitive lexical selection. A decisive experiment would hold the visual overlap approximately constant while manipulating lexical status and the strength of the competing form. Combining error types, latency distributions, and neural measures is preferable to treating a single average response difference as proof of one computational architecture Davis (2010) Seidenberg & McClelland (1989) Grainger & Holcomb (2009).
Masked Priming and the Time Course of Form Activation
Masked priming briefly presents a prime before a target, often with masks or changes of case that reduce the prime's availability for explicit report. When related primes facilitate target performance, the effect shows that processing of the preceding string influenced the target response under the experiment's conditions. Forster and Davis established important repetition-priming phenomena and their dependence on word frequency. Subsequent transposed-letter priming studies probed how imperfect orthographic matches activate representations of known words. A masked prime is therefore an experimental intervention into processing history, although masking quality, prime visibility, target expectations, and response strategy must be assessed rather than presumed Forster & Davis (1984) Perea & Lupker (2003).
Repetition, form similarity, associative relatedness, and phonological overlap produce distinguishable predictions. Exact repetition can facilitate multiple stages, from early visual coding to later lexical processing and decision. A transposed-letter prime narrows the inquiry toward tolerance of letter-position uncertainty, while a pseudohomophone prime can reveal the contribution of phonological representations. Differences in task demands can change whether a prime helps or hinders, particularly when the prime activates a strong competitor. The relevant evidence is the interaction between prime condition and target attributes under controlled temporal presentation; a prime-related difference by itself does not locate the effect at a unique neural or cognitive stage Perea & Lupker (2003) Grainger & Holcomb (2009) Gomez et al. (2008).
Grainger and Holcomb integrated behavioral masked priming with electrophysiological observations to propose a temporal architecture in which early form-related processes interact with phonology and feed later semantic access. This research gives an empirical basis for distinguishing temporally unfolding components without assuming one fixed processing sequence under every script and task. A controlled priming experiment is strongest when it reports stimulus onset asynchrony, mask format, item norms, whether primes are consciously discriminable, and separate accuracy and latency analyses. Cross-task comparisons between masked priming and gaze-contingent preview are especially valuable because both involve prior information that is available before full processing of a target Grainger & Holcomb (2009) Huang et al. (2025).
Lexical Frequency, Contextual Diversity, and Experience
Words encountered often are, on average, recognized more efficiently than unfamiliar words. This robust relationship was historically interpreted as evidence that frequently encountered lexical entries possess stronger representations or more favorable activation thresholds. But raw word frequency bundles several forms of experience: number of encounters, distribution of contexts, orthographic typicality, age of acquisition, and association with familiar meanings. A theory that assigns all latency differences to one input frequency count is vulnerable when those variables are correlated. Models should therefore specify the learning history represented by a corpus measure and test whether the relation survives adjustment for correlated lexical attributes Adelman et al. (2006) Balota et al. (2007).
Adelman, Brown, and Quesada compared corpus frequency with contextual diversity, defined by the number of distinct contexts in which a word occurs. Their analyses found that contextual diversity accounted for naming and lexical-decision patterns that were conventionally attributed to frequency, under the tested corpora and data sets. This is an important empirical challenge to simple count-based explanations of lexical familiarity. It does not establish that repetition never matters or that contextual diversity is a unique underlying causal variable: observational corpus properties are interdependent and may proxy a reader's actual history imperfectly. Randomized exposure experiments and better measures of individual vocabulary can help distinguish frequency of experience from its distribution Adelman et al. (2006) Yap et al. (2012).
The English Lexicon Project made large-scale lexical comparisons possible by collecting behavioral and descriptive measures across many words and nonwords. Its design avoids the extreme restriction of a short hand-selected stimulus list, allowing simultaneous estimation of frequency, length, neighborhood, and other attributes. Yet mega-study findings remain linked to sample composition, word inventory, nonword construction, and task. More recent work on individual differences shows that the same item property can have different consequences for readers with different skills and response tendencies. Item-level regularities should thus be interpreted through hierarchical models that allow both person and word variability rather than through an imagined average reader Balota et al. (2007) Yap et al. (2012).
Phonological Activation During Printed-Word Recognition
Written words generally provide learned routes to phonological representations, and these links can become active even when pronunciation is not the overt task. This creates a fundamental modeling question: when a participant decides that a string is a word, how much of the response is supported by orthographic form and how much by automatically activated speech-related information? A dual-route theory permits direct lexical access alongside procedures translating graphemes to phonology. Distributed models instead learn regularities in mappings through weighted connections and can produce graded effects of consistency. Both families connect behavioral performance to print–sound structure, but they differ in the way mechanisms are represented and learned Coltheart et al. (2001) Seidenberg & McClelland (1989).
The Dual Route Cascaded model formally simulates lexical decision and reading aloud using interacting lexical and nonlexical components. Its explanatory force lies in generating predictions about regular words, exceptions, and novel letter strings rather than merely naming two routes. Seidenberg and McClelland's distributed developmental model offered a contrasting implementation that learned correspondences without explicit stored pronunciation rules of the same sort. Evidence from naming latency, exceptions, and nonword pronunciation constrains these explanations; lexical decision alone cannot determine how a system would pronounce an unfamiliar word. A mechanistic comparison must distinguish the task's intended response and the latent processing that can influence it Coltheart et al. (2001) Seidenberg & McClelland (1989).
Phonological activation can have different functional importance in different writing systems and levels of reading expertise. Frost argued that orthography must be analyzed as a representation of phonological, semantic, and morphological information within a particular system, while Ziegler and Goswami traced developmental variation to the consistency and grain size of print–speech correspondences. Those frameworks make cross-language predictions: readers may rely on different sublexical units, and the same priming manipulation may tap distinct representational relations. Current findings from Korean masked-priming EEG show how early orthographic and later phonological effects can differ in a particular alpha-syllabic script; such evidence refines generalized claims about timing without implying that all alphabetic or nonalphabetic writing systems share an identical cascade Frost (2012) Ziegler & Goswami (2005) Kim et al. (2026).
Morphology, Meaning, and the Boundary of Orthographic Access
Orthographic form enables access to meaningful words, yet form-based recognition must be distinguished from morphological parsing and semantic interpretation. A reader encountering a familiar inflected or derived word may activate the whole printed form, its stem, related word-family members, and meaning-relevant associations. The timing and representational level of such activation depend on task and language. In a visual lexical-decision task, morphological relatedness is often partly correlated with letter overlap, so a priming advantage for related forms must be tested against carefully matched orthographic controls. A complete theory requires an account of both genuine morphological structure and superficial resemblance Grainger & Holcomb (2009) Frost (2012).
Semantic activation is likewise connected to orthographic recognition without being identical to word identification. A word may be classified successfully before its context-dependent interpretation has been resolved, and semantic expectation can feed back toward interpretation of ambiguous visual evidence. Electrophysiological investigations provide a temporal perspective, particularly when form-related changes occur earlier than context-sensitive components. Laszlo and Federmeier illustrated how regression of item-level properties against event-related potentials can assess overlapping influences rather than assign lexical and semantic processing to rigidly isolated stages. The interpretation of a neural component must remain conditional on experimental materials and statistical design Laszlo & Federmeier (2014) Grainger & Holcomb (2009).
This dossier therefore treats morphological and semantic activation as critical interactions at the boundary of its object. The dedicated mental-lexicon dossier addresses the larger organization of morphological and lexical knowledge; sentence-processing research addresses interpretation in discourse. Within visual-word research, the central test is whether letter-order and orthographic-competition effects survive when pronunciation, morphology, and meaning are controlled or explicitly manipulated. Such factorial designs provide firmer conclusions than attributing every lexicality effect to a pure orthographic store Carreiras et al. (2014) Frost (2012) Price & Devlin (2011).
Experimental Identification and Lexical-Decision Methods
The word-identification paradigm varies exposure duration, degradation, alternatives, and contextual constraints to ask how accurately an orthographic form can be recovered from incomplete evidence. Lexical decision, by contrast, demands a judgment about membership in a learned word inventory. The distinction is consequential because the same printed word may be identified correctly yet receive a delayed lexicality response when the nonword alternatives are unusually wordlike. Multiple read-out models accommodate this by permitting decisions from several dimensions of accumulated activation and by allowing criteria to change across task environments. A careful study reports the exact distractor construction, display conditions, instructions, and response deadlines, because these properties can change the apparent psychological effect of a lexical variable Grainger & Jacobs (1996) Balota et al. (2007).
Response-time means summarize performance but can conceal shifts in distribution, speed–accuracy tradeoffs, and differences among participants. A condition with longer responses may reflect delayed perceptual encoding, slower evidence accumulation, a more conservative decision threshold, or motor latency. Experimental comparisons are stronger when response-time distributions, accuracy, and participant- and item-level variability are modeled jointly. The overlap model's forced-choice tests illustrate how a computational representation can be evaluated in a task designed to separate identification from broad lexicality judgments. English Lexicon Project analyses further show the value of large item samples and participant-level differences Gomez et al. (2008) Yap et al. (2012).
For isolated-word experiments, the appropriate unit of inference is often the participant-by-item response, since each reader brings different experience and each target carries multiple correlated attributes. Mixed-effects or suitably hierarchical analyses can accommodate this crossed structure, although the causal interpretation continues to depend on the manipulation. Frequency, orthographic neighborhood, and letter length cannot be assumed independent simply because they appear in separate regression columns. A rigorous comparison between models should specify the predicted sign and functional shape of each effect in advance and test out-of-sample generalization. Descriptive megastudies expand coverage, whereas controlled manipulations remain important for identifying underlying mechanisms Balota et al. (2007) Adelman et al. (2006) Yap et al. (2012).
Word Naming, Orthographic Consistency, and Reading Routes
In speeded naming, the participant pronounces a displayed letter string, which introduces demands beyond visual identification: access to a plausible pronunciation, preparation of articulation, and execution of the response. Regularity and consistency effects provide evidence about how printed units connect to speech. The Dual Route Cascaded model explicitly separates lexical and grapheme-to-phoneme procedures, allowing a known exception word to be read through stored lexical knowledge while a nonword receives a constructed pronunciation. Distributed learning models assign these patterns to the learned structure of mappings across shared units. The competition between accounts concerns precise generalization to stimulus classes, not whether phonology is relevant to reading Coltheart et al. (2001) Seidenberg & McClelland (1989).
The nature of naming difficulty depends on the orthography. English contains inconsistencies at multiple grain sizes; a model trained only on transparent correspondences cannot fully explain its exception-word behavior. Across languages, spelling-to-sound consistency changes the relative predictiveness of graphemes, larger units, and entire word forms. Ziegler and Goswami's psycholinguistic grain-size theory predicts consequences for both acquisition and skilled recognition. Frost's broader framework likewise emphasizes that orthographic systems represent multiple kinds of linguistic structure, which alters what a shared computational principle must explain. Cross-language comparison demands equivalence of stimuli and learning conditions, not simply an English-centered model evaluated on translated word lists Ziegler & Goswami (2005) Frost (2012).
Speeded naming contributes evidence about the time needed to prepare a spoken response, but an articulation effect should not be attributed automatically to the visual encoding stage. Pronunciation ambiguity, vocal onset measurement, and variation in familiarity with particular spoken forms can change response times independently of orthographic discrimination. Analyses that compare naming with lexical decision are informative because a variable's effects may differ according to the response required. A plausible architecture must explain such divergences while preserving common representational constraints, including the graded letter-position similarity observed in identification experiments Coltheart et al. (2001) Gomez et al. (2008) Balota et al. (2007).
Eye Movements, Parafoveal Preview, and Word Position
Ordinary reading proceeds through sequences of fixations and saccades. During a fixation, letters near the center of gaze are resolved more precisely than those at greater eccentricities; the reader's next eye movement changes which word is sampled at high acuity. Rayner's synthesis established how fixation duration, perceptual span, and integration across movements can inform cognitive theories of reading. A word may be partly processed while it lies in parafoveal vision, producing a subsequent benefit when it becomes the fixation target. This means that word recognition in natural reading depends on earlier sampling history as well as the visual evidence available at the moment of fixation Rayner (1998) Rayner (2009).
In gaze-contingent boundary experiments, a target can be changed during a saccade, separating the identity of the preview from that of the subsequently fixated item. Shorter fixations after informative previews support the claim that parafoveal information contributes to later processing. Yet the effect depends on visual position, attention, context, and the kind of information previewed. Preview benefit must not be equated with completed word recognition before fixation; it can arise through partial orthographic, phonological, or contextual preparation. Further, sentence contexts bring predictability and syntactic constraints that are absent from isolated lexical-decision trials Rayner (2009) Grainger & Holcomb (2009) Huang et al. (2025).
A direct comparison of masked foveal priming and parafoveal preview offers a way to test whether the same processing stages are facilitated by distinct sources of prior visual information. Huang and colleagues combined eye tracking and EEG while presenting Chinese character materials in matched conditions. Their findings included qualitatively comparable early repetition-related activity and differences in effect magnitude across the two paradigms. The result strengthens the argument for common components while also showing that experimental route and viewing conditions matter. Integration across paradigms requires explicit treatment of gaze position, preview validity, visual exposure, and time-locking of physiological signals Huang et al. (2025) Rayner (1998).
Electrophysiological Dynamics of Orthographic Access
Event-related potentials provide millisecond-scale measurements of stimulus-related electrical activity. In visual-word experiments, early components are influenced by visual and orthographic features, while later changes can reflect converging lexical and semantic processing. Grainger and Holcomb reviewed masked-priming studies in which form-related activity appears in temporal relation to phonological and semantic contributions. The N250 is commonly examined as a candidate index of orthographic-to-lexical form mapping, and N400 modulation is frequently connected to later meaning-related integration. These labels describe observable waveform families; they do not identify single, anatomically isolated computations. Scalp distributions depend on reference montage, task, language, and stimulus history Grainger & Holcomb (2009) Laszlo & Federmeier (2014).
Laszlo and Federmeier's single-item regression approach illustrates why temporal claims should consider continuous variation in item properties rather than only subtracting two arbitrary word lists. An effect tied to orthographic neighborhood may overlap with frequency, length, or semantic associations; electrophysiological timing by itself cannot assign causal direction among these factors. Researchers can increase explanatory power by jointly analyzing graded stimulus properties, repetition histories, and behavioral decisions while controlling for multiple comparisons and correlated predictors. Replication with different item sets tests whether a reported latency window generalizes beyond a particular stimulus sample Laszlo & Federmeier (2014) Grainger & Holcomb (2009).
The 2026 Korean masked-priming study offers a contemporary writing-system test. Its reported design contrasted orthographically overlapping and phonologically overlapping syllable primes using lexical decisions, EEG event-related potentials, and oscillatory measures in thirty participants. The authors reported early orthographic modulation and later phonological effects under the tested conditions. These observations are particularly useful because the orthographic and phonological overlap could be manipulated separately within Hangul. The sample and task support an interpretation about this experimental configuration, while larger independent replications and convergent behavioral paradigms are needed to characterize broader universality. The result challenges an assumption that early orthographic and phonological contributions are always temporally indistinguishable Kim et al. (2026) Frost (2012).
The Visual Word Form Area and Distributed Reading Networks
The left ventral occipitotemporal cortex is regularly implicated in skilled reading, and a region commonly termed the visual word form area has become central to neuroscience of written-word recognition. Cohen and colleagues characterized its anatomical and temporal involvement using human imaging and neurological observations. Dehaene and Cohen later developed the neuronal-recycling account, proposing that literacy reorganizes visual cortical resources whose prior properties constrain their adaptation for print. This framework links cultural learning with a reproducible functional profile, but its interpretation depends on how selectivity is measured and how other cortical areas participate in the task Cohen et al. (2000) Dehaene & Cohen (2011).
Price and Devlin's interactive account places ventral occipitotemporal activity within a network integrating bottom-up visual evidence with feedback from speech and meaning systems. Their proposal differs from a sharply encapsulated orthographic module by predicting that the same region can be sensitive to experience and higher-level associations. The two influential accounts partly converge on acquired specialization yet prioritize different mechanisms in its emergence and operation. Functional imaging responses alone cannot decide whether a local signal represents letters, task difficulty, prediction error, phonological access, or their interaction. Strong adjudication combines stimulus contrasts with connectivity, lesion data, and temporal measurements Price & Devlin (2011) Dehaene & Cohen (2011).
Acquired reading impairment provides causal constraints unavailable from imaging correlations alone. Lesions involving left occipitotemporal pathways can disrupt fluent recognition of printed words, sometimes producing letter-by-letter reading. Such patterns demonstrate the importance of affected pathways for efficient processing, but heterogeneous damage and compensatory behavior require detailed anatomical and behavioral characterization. A complete neural architecture must account for how features are transformed into orthographic patterns and how those patterns exchange information with pronunciation and meaning. The strongest current conclusion is that fluent visual-word recognition depends on a learned, distributed, interactive brain network with a prominent ventral occipitotemporal contribution Cohen et al. (2000) Dehaene & Cohen (2011) Price & Devlin (2011).
Cross-Linguistic Orthography and Universality
A theory of visual-word recognition must account for writing systems with different mappings among visual forms, speech, morphology, and meaning. Alphabetic orthographies vary in spelling-to-sound consistency; abjads make consonantal and morphological structure salient; syllabic and morphosyllabic systems organize orthographic units differently. It is therefore an empirical question which components of a recognition model generalize at the level of visual discrimination, letter or character identity, relative position, and learned linguistic correspondences. Frost argued that universal principles should be derived while respecting the functional structure of each orthography. The challenge is to distinguish genuinely general processing operations from the representation-specific phenomena produced by a writing system Frost (2012) Ziegler & Goswami (2005).
Cross-language masked-priming evidence challenges simple accounts in which the same positional and temporal mechanisms operate identically everywhere. Boudelaa, Perea, and Carreiras examined early orthographic processing with Semitic words and placed universal and language-dependent accounts in direct experimental tension. Findings of this kind matter because patterns of morphological organization, letter arrangement, and lexical competition may influence effects often treated as primitive visual similarities. A cross-language study should match orthographic complexity, exposure, expertise, and task demands as closely as possible. Literal translation of stimuli seldom preserves all these variables; script-specific computational representations and comparable behavioral outcomes offer a stronger basis for inference Boudelaa et al. (2025) Frost (2012) Carreiras et al. (2014).
Research on bilingual recognition adds another test of the scope of an orthographic system. Biedermann and colleagues studied cross-language orthographic neighborhood density in Dutch–English and Spanish–English bilinguals, making the influence of a reader's combined lexical experience experimentally relevant. The existence of shared or partly competing form information across languages does not establish one uniform bilingual lexicon; it points to measurable interactions conditioned by proficiency, task language, and language history. Comparisons among monolingual and bilingual readers should identify the participant's learned scripts, age of acquisition, and relative usage. General conclusions about early orthographic activation must be weighed against these individual and cross-script differences Biedermann et al. (2024) Lecerf et al. (2024).
Learning the Orthographic Lexicon and Skilled Expertise
Skilled recognition depends on histories of learning: visual letter categories become stable, print–speech correspondences become familiar, and particular written forms accumulate lexical associations. Share's self-teaching hypothesis assigns a critical role to successful phonological recoding during reading acquisition because it allows a learner to acquire word-specific orthographic knowledge from new print encounters. This framework connects early decoding to later fluent recognition without treating fluent adult identification as the same process as initial decoding. It also makes testable claims about how exposure to unfamiliar words changes later recognition and spelling. Orthographic learning is consequently part of the causal history of skilled word recognition even when immediate experimental questions concern adult lexical access Share (1995) Ziegler & Goswami (2005).
The psycholinguistic grain-size theory links writing-system consistency with units that readers learn to use productively. In a more inconsistent system, relatively large chunks or whole-word representations may acquire particular utility; in other systems smaller correspondences support broad generalization. The development of a reader's orthographic lexicon also depends on vocabulary, instruction, amount of print exposure, and the distribution of encountered words. These influences yield individual differences in the precision and accessibility of the forms used during adult recognition. A developmental explanation should measure acquisition and transfer through repeated encounters rather than infer childhood learning solely from adult word-frequency effects Ziegler & Goswami (2005) Share (1995) Yap et al. (2012).
The distinction between acquisition and mature operation also matters for educational translation. Evidence that a skilled adult can rapidly identify a word under masked priming does not show that teaching an inexperienced reader whole visual shapes would be an effective route to literacy. Conversely, the importance of decoding during acquisition does not imply that every adult recognition decision entails overt phonological recoding. The mechanism of learning, the form of mature representation, and the procedure used for a particular response are separate empirical targets. Interventions need their own controlled learning outcomes, retention tests, and measures of transfer; predictions from adult recognition models are hypotheses, not treatment validation Share (1995) Ziegler & Goswami (2005) Coltheart et al. (2001).
Individual Differences, Bilingual Experience, and Reader Variability
The conventional literature often reports average response times aggregated over experienced adult readers. Such averages are useful for identifying robust item effects but conceal variation in lexical skill, processing speed, language background, and decision thresholds. Yap and colleagues analyzed English Lexicon Project data to examine differences among individuals, demonstrating that word-recognition research benefits from modeling people as well as words. Effects associated with frequency or length can be moderated by reader skill, so a successful model must predict not only which stimuli are relatively difficult but also how that difficulty changes with experience. The relevant generalization is statistical and conditional rather than the behavior of a single idealized participant Yap et al. (2012) Balota et al. (2007).
Bilingual readers provide additional leverage because the same visual string can have different lexical status and phonological associations across languages. Studies of orthographic markedness and cross-language neighborhood effects show that prior language knowledge is an active part of recognition rather than incidental participant metadata. Lecerf, Casalis, and Commissaire synthesized how markedness can shape bilingual processing, while Biedermann and colleagues examined neighborhood relationships across bilingual groups. These literatures require explicit distinctions between script similarity, shared letter inventory, lexical similarity, and phonological overlap. Differences between bilingual conditions may reflect exposure and control demands as much as the architecture of representational access Lecerf et al. (2024) Biedermann et al. (2024).
A mature account should also encompass variation across age and reading proficiency without assigning a diagnostic label from a single laboratory effect. Naming speed, lexical-decision accuracy, and electrophysiological amplitudes are measures of performance under defined conditions; they are not interchangeable with reading comprehension, educational attainment, or clinical diagnosis. Assessment of developmental reading differences requires broader language, literacy, and history measures. The contribution of visual-word recognition research is to provide mechanisms and targeted measures that can inform such investigation while preserving the distinction between cognitive task performance and the functional outcome of reading Ziegler & Goswami (2005) Yap et al. (2012) Share (1995).
Computational Model Comparison and Explanatory Scope
The main theoretical families make distinct commitments about the representation of written-word information. Interactive-activation and multiple read-out models emphasize word candidates, sublexical support, competition, and task-sensitive criteria. Distributed developmental models explain regularities through learned mappings, while spatial coding and overlap models target the positional uncertainty of letters. Dual-route models emphasize distinct procedures relevant to lexical access and pronunciation. No taxonomy by itself determines which model is preferable: an explanatory comparison requires simulation of the same stimuli, response conditions, errors, and distributions. A model that reproduces one classic effect but fails basic transposition tests remains incomplete for the intended scientific object McClelland & Rumelhart (1981) Grainger & Jacobs (1996) Seidenberg & McClelland (1989) Coltheart et al. (2001) Gomez et al. (2008).
Model comparison is most informative when the alternatives differ on an intervention rather than fit a heavily aggregated mean. A rigid-slot account and an overlap account may yield divergent predictions when letter identity is preserved but position varies. A fixed-criterion and a flexible-decision account may diverge when nonword similarity changes between blocks. A purely orthographic and an interacting phonological explanation may differ under controlled pseudohomophone primes or cross-script conditions. Such contrasts should be defined before inspecting the critical outcomes. Parameter flexibility, corpus representativeness, and training experience must be reported, since unconstrained tuning can reproduce experimental averages without producing a stable theory of recognition Grainger & Jacobs (1996) Gomez et al. (2008) Davis (2010) Frost (2012).
Neural and behavioral accounts are linked at different levels. Behavioral models describe computational transformations and choice rules; brain imaging identifies engaged regions and their dynamics; lesion and stimulation research can constrain causal necessity in specific pathways. A successful multilevel account does not require a one-to-one mapping between every model unit and an anatomical population. It requires explicit bridging predictions about timing, representational similarity, and intervention outcomes. Price and Devlin's interactive proposal and Dehaene and Cohen's neuronal-recycling framework illustrate the importance of distinguishing functional specialization from the network mechanisms that generate it Price & Devlin (2011) Dehaene & Cohen (2011) Carreiras et al. (2014).
Evidence Appraisal, Reproducibility, and Measurement Bias
Evidence about printed-word recognition is strongest when different procedures converge on the same representational constraint. Masked priming isolates brief preceding exposure; forced-choice identification assesses similarity-based confusability; lexical decision adds criterion setting; naming samples print–sound access; eye tracking records information acquisition during reading; EEG estimates the timing of physiological correlates; imaging contributes spatial information. Each also creates characteristic limitations. A reaction-time contrast can combine perceptual and decision components, a masked prime can be partly visible, and an ERP amplitude can arise from several interacting neural sources. A claim about a distinct processing stage should therefore rest on factorial controls and convergent evidence rather than one measurement label Grainger & Jacobs (1996) Grainger & Holcomb (2009) Huang et al. (2025).
Studies differ in participant recruitment, item selection, treatment of error trials, control of correlated lexical variables, and reporting of exclusion criteria. Small factorial experiments can establish closely controlled causal contrasts but may have narrow linguistic coverage; megastudies offer broad item-level estimation but often depend on observational lexical predictors. Neither design eliminates the need for replication. Reanalysis using shared materials, distribution-sensitive models, and independent participant samples can reveal whether an effect is an invariant property of orthographic processing or contingent on a particular stimulus population and decision environment. The scientific target is a reproducible relation between a manipulation and a proposed computation Balota et al. (2007) Yap et al. (2012) Adelman et al. (2006).
The evidence reviewed here includes foundational original experiments, formal models, methodological syntheses, and recent cross-language studies. Its scope is deliberately narrative and mechanism-centered: no exhaustive database-search log, preregistered eligibility rules, risk-of-bias coding, or pooled meta-analytic result was produced. Accordingly, the review can evaluate the coherence of convergent findings and specify unresolved contrasts but cannot provide an estimated global prevalence of any effect or a quantitative certainty grade across the field. The appropriate next step for a disputed effect is a registered multi-laboratory comparison with shared stimuli, transparent analysis, and sufficiently varied writing systems Carreiras et al. (2014) Frost (2012) Boudelaa et al. (2025).
Artificial Vision Models and Orthographic Representation
Artificial recognition systems supply computational comparators for the scientific problem of print. A vision model can learn to classify letters, transform sequences into embeddings, and map those embeddings onto lexical or phonological outputs. Such systems allow controlled ablation of the visual encoder, position scheme, training distribution, and decision rule, thereby turning architectural assumptions into directly testable manipulations. Yet high classification accuracy does not demonstrate that an artificial model uses the same representational organization as a human reader. Behavioral alignment requires tests on transpositions, substitution neighbors, visual changes, novel strings, and generalization beyond the training corpus Gomez et al. (2008) Davis (2010) Seidenberg & McClelland (1989).
The most promising comparisons are diagnostic rather than metaphorical. A model with explicit positional distributions can be tested against overlap effects, a recurrent model against time-sensitive priming, and a learned distributed code against frequency and consistency gradients. Training on multilingual orthographies may help establish which regularities arise from generic visual sequence learning and which reflect language-specific structure. Evaluating multiple trained instances is important because an observed performance difference may be attributable to optimization or dataset idiosyncrasy rather than the intended representational mechanism. Work on artificial systems should report objective functions and exposure histories, enabling meaningful comparison with the educational and perceptual history of readers Frost (2012) Carreiras et al. (2014) Share (1995).
Within the Artificial Era, the comparative research question concerns the functional organization of orthographic knowledge across biological and artificial information-processing systems. A scientific explanation remains anchored in operational measures, controlled generalization, and reproducible model comparisons. An artificial word recognizer can serve as an experimentally manipulable model of spatial coding without inheriting the biological neural substrates of reading, and human behavior can constrain such a model without requiring that its internal state be read as a psychological experience. The shared object of analysis is the structure and transformation of information about visible linguistic forms Price & Devlin (2011) Davis (2010) Frost (2012).
Open Problems and a Research Agenda
The principal unresolved theoretical problem is how orthographic codes preserve identity across visual variation while remaining sensitive to informative letter order and lexical competition. Flexible position theories explain transposition phenomena, but the contributions of relative order, absolute position, and language-specific sublexical statistics remain difficult to separate across all scripts. A decisive program would compare carefully matched position perturbations under identical visual conditions in readers of different writing systems, then test whether a single trained architecture reproduces both response distributions and individual differences. The desired output is a set of quantitative predictions rather than the post hoc naming of a representational unit Gomez et al. (2008) Davis (2010) Boudelaa et al. (2025).
A second problem concerns temporal coordination. Orthographic, phonological, and semantic activation can overlap, and the ordering inferred from ERP components can change with stimulus materials, task, and measurement reference. Controlled masked-priming experiments, naturalistic eye movements, and simultaneous EEG offer complementary evidence. The within-participant comparison used by Huang and colleagues illustrates the value of bringing previously separate paradigms into direct experimental contact. Future studies should preregister predictions about the timing, location, and direction of condition effects and assess whether the predicted temporal relation survives changes in script, familiarity, and lexical decision criterion Grainger & Holcomb (2009) Huang et al. (2025) Kim et al. (2026).
A third problem is learning and population generalization. Laboratory adults differ in reading history and linguistic knowledge, while novel artificial systems can be trained with precisely known exposure distributions. Linking longitudinal human learning, corpus-derived measures, and simulated acquisition may resolve why frequency, contextual diversity, and orthographic typicality predict recognition in partly overlapping ways. Such work should distinguish development of representational knowledge from strategic response learning and from general processing speed. It should also make the relation to literacy instruction precise through independent training studies rather than extrapolation from adult lexical decision. The scientific objective is a causal and computational explanation of durable, flexible recognition across people, scripts, and tasks Adelman et al. (2006) Yap et al. (2012) Share (1995).
Conclusion
Visual word recognition is a learned, dynamically organized process through which visible linguistic forms become identifiable despite variability in visual appearance and incomplete positional information. Orthographic processing extracts letter or character identities, represents their sequence with useful flexibility, activates and discriminates lexical candidates, and coordinates with phonology and meaning. Research from word-superiority experiments to interactive activation, distributed learning, flexible position coding, and modern electrophysiology has progressively constrained these operations. The strongest findings concern the measurable effects of orthographic similarity, lexical experience, timing, and task structure; no single experiment establishes the entirety of the recognition mechanism McClelland & Rumelhart (1981) Grainger & Jacobs (1996) Gomez et al. (2008) Carreiras et al. (2014).
The evidence also establishes why methods must be interpreted in relation to their specific dependent variables. Identification, lexical decision, naming, masked priming, gaze-contingent preview, and brain recordings disclose overlapping but distinct aspects of the same system. Modern cortical theories converge on an important role for learned ventral occipitotemporal processing embedded in broader language networks, while leaving open the organization of feedback and the division of labor among components. Cross-linguistic research makes clear that an adequate account must explain both general visual processing principles and the orthographic structure that readers have actually learned Rayner (1998) Dehaene & Cohen (2011) Price & Devlin (2011) Frost (2012).
A scientifically mature explanation requires models that generate testable predictions across words, nonwords, scripts, readers, and experimental procedures. Future progress will depend on computational contrasts with explicit learning histories, appropriately powered interventions, shared stimulus resources, and convergent behavioral and physiological measures. This monograph consolidates that research program as an account of a specific psychological function: reliable access to learned written-word forms under the constraints of vision, language, experience, and decision. Its conclusions are a synthesis of published research and a guide to questions that can be resolved by further experiments Davis (2010) Yap et al. (2012) Huang et al. (2025).
References
- Adelman, J. S., Brown, G. D. A., & Quesada, J. F. (2006). Contextual diversity, not word frequency, determines word-naming and lexical decision times. Psychological Science, 17(9), 814–823. https://doi.org/10.1111/j.1467-9280.2006.01787.x
- Andrews, S. (1996). Lexical retrieval and selection processes: Effects of transposed-letter confusability. Journal of Memory and Language, 35(6), 775–800. https://doi.org/10.1006/jmla.1996.0040
- Balota, D. A., Yap, M. J., Cortese, M. J., Hutchison, K. A., Kessler, B., Loftis, B., Neely, J. H., Nelson, D. L., Simpson, G. B., & Treiman, R. (2007). The English Lexicon Project. Behavior Research Methods, 39(3), 445–459. https://doi.org/10.3758/BF03193014
- Biedermann, B., Beyersmann, E., Blosfelds, M., Macapagal, C., Rosevear, A., & Marinovic, W. (2024). Cross-language orthographic neighborhood density effects in Dutch–English and Spanish–English bilinguals. Frontiers in Language Sciences, 3, Article 1482861. https://doi.org/10.3389/flang.2024.1482861
- Boudelaa, S., Perea, M., & Carreiras, M. (2025). Are the early stages of orthographic processing universal? Insights from masked priming with Semitic words. Psychonomic Bulletin & Review, 32(2), 770–778. https://doi.org/10.3758/s13423-024-02563-8
- Carreiras, M., Armstrong, B. C., Perea, M., & Frost, R. (2014). The what, when, where, and how of visual word recognition. Trends in Cognitive Sciences, 18(2), 90–98. https://doi.org/10.1016/j.tics.2013.11.005
- Cohen, L., Dehaene, S., Naccache, L., Lehéricy, S., Dehaene-Lambertz, G., Hénaff, M.-A., & Michel, F. (2000). The visual word form area: Spatial and temporal characterization of an initial stage of reading in normal subjects and posterior split-brain patients. Brain, 123(2), 291–307. https://doi.org/10.1093/brain/123.2.291
- Coltheart, M., Rastle, K., Perry, C., Langdon, R., & Ziegler, J. (2001). DRC: A dual route cascaded model of visual word recognition and reading aloud. Psychological Review, 108(1), 204–256. https://doi.org/10.1037/0033-295X.108.1.204
- Davis, C. J. (2010). The spatial coding model of visual word identification. Psychological Review, 117(3), 713–758. https://doi.org/10.1037/a0019738
- Dehaene, S., & Cohen, L. (2011). The unique role of the visual word form area in reading. Trends in Cognitive Sciences, 15(6), 254–262. https://doi.org/10.1016/j.tics.2011.04.003
- Forster, K. I., & Davis, C. (1984). Repetition priming and frequency attenuation in lexical access. Journal of Experimental Psychology: Learning, Memory, and Cognition, 10(4), 680–698. https://doi.org/10.1037/0278-7393.10.4.680
- Frost, R. (2012). Towards a universal model of reading. Behavioral and Brain Sciences, 35(5), 263–279. https://doi.org/10.1017/S0140525X11001841
- Gomez, P., Ratcliff, R., & Perea, M. (2008). The overlap model: A model of letter position coding. Psychological Review, 115(3), 577–600. https://doi.org/10.1037/a0012667
- Grainger, J., & Holcomb, P. J. (2009). Watching the word go by: On the time-course of component processes in visual word recognition. Language and Linguistics Compass, 3(1), 128–156. https://doi.org/10.1111/j.1749-818X.2008.00121.x
- Grainger, J., & Jacobs, A. M. (1996). Orthographic processing in visual word recognition: A multiple read-out model. Psychological Review, 103(3), 518–565. https://doi.org/10.1037/0033-295X.103.3.518
- Grainger, J., Rey, A., & Dufau, S. (2008). Letter perception: From pixels to pandemonium. Trends in Cognitive Sciences, 12(10), 381–387. https://doi.org/10.1016/j.tics.2008.06.006
- Huang, X., Wong, B. W. L., Ng, H. T.-Y., Sommer, W., Dimigen, O., & Maurer, U. (2025). Neural mechanism underlying preview effects and masked priming effects in visual word processing. Attention, Perception, & Psychophysics, 87(1), 5–24. https://doi.org/10.3758/s13414-024-02904-8
- Kim, J., Lee, S., & Nam, K. (2026). Rapid orthographic and delayed phonological processing: ERP and oscillatory evidence from masked priming in Korean. Cortex, 200, 74–91. https://doi.org/10.1016/j.cortex.2026.04.008
- Laszlo, S., & Federmeier, K. D. (2014). Never seem to find the time: Evaluating the physiological time course of visual word recognition with regression analysis of single item ERPs. Language and Cognitive Processes, 29(5), 642–661. https://doi.org/10.1080/01690965.2013.866259
- Lecerf, M.-A., Casalis, S., & Commissaire, E. (2024). New insights into bilingual visual word recognition: State of the art on the role of orthographic markedness, its theoretical implications, and future research directions. Psychonomic Bulletin & Review, 31(3), 1032–1056. https://doi.org/10.3758/s13423-023-02347-6
- McClelland, J. L., & Rumelhart, D. E. (1981). An interactive activation model of context effects in letter perception: I. An account of basic findings. Psychological Review, 88(5), 375–407. https://doi.org/10.1037/0033-295X.88.5.375
- Perea, M., & Lupker, S. J. (2003). Does jugde activate COURT? Transposed-letter similarity effects in masked associative priming. Memory & Cognition, 31(6), 829–841. https://doi.org/10.3758/BF03196438
- Price, C. J., & Devlin, J. T. (2011). The interactive account of ventral occipitotemporal contributions to reading. Trends in Cognitive Sciences, 15(6), 246–253. https://doi.org/10.1016/j.tics.2011.04.001
- Rayner, K. (1998). Eye movements in reading and information processing: 20 years of research. Psychological Bulletin, 124(3), 372–422. https://doi.org/10.1037/0033-2909.124.3.372
- Rayner, K. (2009). Eye movements and attention in reading, scene perception, and visual search. Quarterly Journal of Experimental Psychology, 62(8), 1457–1506. https://doi.org/10.1080/17470210902816461
- Seidenberg, M. S., & McClelland, J. L. (1989). A distributed, developmental model of word recognition and naming. Psychological Review, 96(4), 523–568. https://doi.org/10.1037/0033-295X.96.4.523
- https://doi.org/10.1016/0010-0277(94)00645-2
- Whitney, C. (2001). How the brain encodes the order of letters in a printed word: The SERIOL model and selective literature review. Psychonomic Bulletin & Review, 8(2), 221–243. https://doi.org/10.3758/BF03196158
- Yap, M. J., Balota, D. A., Sibley, D. E., & Ratcliff, R. (2012). Individual differences in visual word recognition: Insights from the English Lexicon Project. Journal of Experimental Psychology: Human Perception and Performance, 38(1), 53–79. https://doi.org/10.1037/a0024177
- Ziegler, J. C., & Goswami, U. (2005). Reading acquisition, developmental dyslexia, and skilled reading across languages: A psycholinguistic grain size theory. Psychological Bulletin, 131(1), 3–29. https://doi.org/10.1037/0033-2909.131.1.3