Introduction

There are minimal differences at a visual level between the words tight, light, and fight, and yet fluent readers are able to distinguish them with remarkable accuracy within a fraction of a second. Words like these, that look similar to many other words in the lexicon, are said to come from high-density (HD) orthographic neighborhoods. Words that have distinct letter combinations and do not resemble many other words (e.g., awful, kayak) belong to low-density (LD) orthographic neighborhoods. We asked whether the orthographic neighborhood density of a word influences the way in which it is represented in lexical memory, and how the nature of these representations impacts word recognition. More specifically, we compared the size of masked transposed-letter (TL) priming effects for HD and LD targets to examine the central tenet of the lexical tuning hypothesis (Forster & Taft, 1994): that words with more neighbors have more precise orthographic representations.

There is general consensus that priming effects—particularly those obtained in the masked priming paradigm—reflect the extent to which a prime stimulus preactivates the representations needed for subsequent target processing. In addition to facilitating target processing, primes activate lexical representations that then compete with the target for identification. By comparing the effects of different types of primes, we gain insight into the dimensions of similarity to which the orthographic processor is sensitive. For example, targets elicit faster lexical decision responses following TL primes formed by transposing two letters in the target word (e.g., tgiht-TIGHT) compared with substitution primes formed by replacing the letters in those same positions (e.g., tjoht-TIGHT; Comesaña et al., 2016; Ktori et al., 2014; Perea & Carreiras, 2006, 2008; Perea, Duñabeitia, & Carreiras, 2008; Perea & Lupker, 2004a). In the ERP waveform, there is some evidence to suggest that targets in the TL condition elicit smaller amplitude negativities (i.e., less effortful processing) than those in the substitution condition (Carreiras et al., 2009; Grainger, Kiyonaga, & Holcomb, 2006; Ktori et al., 2014). Across studies, the effect has been reported within the N250 and N400 windows, suggesting that processing at both the sublexical and lexical levels is facilitated (see Grainger & Holcomb, 2009, for a summary of the evidence that the N250 reflects sublexical processing and the N400 reflects lexical processing). Theoretically, these TL priming effects indicate that there must be some degree of flexibility or imprecision in the assignments between letters and their positions in the word. Indeed, misspelled words formed by transposing adjacent letters elicit similar N400s as correctly spelled words in the context of moderately constraining sentences, suggesting that the small differences in orthography do not hamper semantic access (Stites et al., 2016). If each letter was assigned an absolute position with perfect precision, then TL misspellings should not elicit lexico-semantic processing. In the context of masked priming, perfect precision would imply that TL and substitution primes should be considered equally similar to the target word and should have equal influence on target processing.

The finding that lexical representations are not encoded with perfect precision prompts a number of intriguing questions, not the least of which is how the human brain represents and processes orthographic information. In the present study, we focus on the precision with which orthographic information is encoded and how this precision varies as a function of orthographic neighborhood density. The lexical tuning hypothesis, which was first entertained by Forster (1987) and formulated by Forster and Taft (1994), posits that the orthographic representations of HD words are more narrowly tuned compared to LD words (see also, Castles et al., 2007; Castles et al., 1999, for a developmental version of the hypothesis). The proposal was formulated based on the neighborhood density constraint (Forster & Davis, 1991), reflecting the observation that masked priming effects with orthographically related nonword primes (e.g., drice-drive) are greater for LD target words than HD words (Andrews & Hersch, 2010; Castles et al., 1999; Forster et al., 1987; Forster & Davis, 1991; Forster & Taft, 1994; Meade, Grainger, Midgley, Emmorey, & Holcomb, 2018a; Perea & Rosa, 2000).

The lexical tuning hypothesis makes the key prediction that LD words should be more tolerant to orthographic changes across prime and target stimuli in a masked priming paradigm and, therefore, that TL priming effects should be greater for LD words than HD words. However, the two behavioral studies that have investigated this issue have yielded inconsistent results (Kinoshita et al., 2009, Experiment 1; Perea & Lupker, 2004b). Perea and Lupker reported that TL priming effects were significantly reduced for HD word targets compared with LD word targets. Although the effects went in the same direction numerically in the study reported by Kinoshita et al., the interaction between the size of TL priming and neighborhood density failed to reached significance. Especially considering that null effects like the one reported by Kinoshita and colleagues are notoriously difficult to interpret, the issue of whether the size of TL priming is modulated by orthographic neighborhood density remains unresolved. The present study therefore provides a further investigation of TL priming with HD and LD words, this time with the added sensitivity of ERPs. Any facilitatory TL priming effects seen in RTs should be accompanied by smaller amplitude N250s and N400s for word targets preceded by TL primes (Carreiras et al., 2009; Grainger et al., 2006; Ktori et al., 2014). Crucially, according to the lexical tuning hypothesis, neighborhood density should modulate the size of these electrophysiological TL priming effects; LD words should elicit larger TL priming effects than HD words.

Methods

Participants

Participants included 48 young adults (34 F; mean age 21.6, SD 3.1) who were right-handed and had normal or corrected-to-normal vision. By self-report, all participants were native speakers of English and were not fluent in any other language. Participants had no history of neurological dysfunction and had not been diagnosed with a language or reading disorder. An additional 11 participants took part in the experiment; however, their data were excluded from analyses due to high artifact rejection rates (>20% of all trials) or experimenter error. All participants were recruited and provided written informed consent in accordance with the Institutional Review Board at San Diego State University.

Stimuli

Each trial consisted of a lowercase prime followed by an uppercase target, both of which were five letters long. There were 100 word targets and 100 pseudoword targets. Half of the targets in each condition came from HD neighborhoods and the other half came from LD neighborhoods (Table 1). Neighborhood density was determined using OLD20, which reflects the number of additions, deletions, or substitutions required to obtain the twenty closest orthographic neighbors. All HD targets had an OLD20 of 1.75 or less; all LD targets had an OLD20 of 1.85 or greater. These restrictions and the resulting mean OLD20 values for each condition are comparable to previous ERP studies of neighborhood density in which this measure was used (Meade et al., 2019; Meade, Midgley, Dijkstra, & Holcomb, 2018b; Vergara-Martínez & Swaab, 2012). HD words had a significantly smaller OLD20 than LD words, t(98) = 11.44, p < 0.001, and HD pseudowords had a significantly smaller OLD20 than LD pseudowords, t(98) = 12.81, p < 0.001. However, HD word and pseudoword targets had a comparable OLD20, t(98) = 0.00, p = 1.00, as did LD word and pseudoword targets, t(98) = 0.00, p = 1.00. Due to high correlations between neighborhood density and constrained bigram frequency, the two variables followed the same patterns.Footnote 1 HD words had a higher bigram frequency than LD words, t(98) = 7.44, p < 0.001, HD pseudowords had a higher bigram frequency than LD pseudowords, t(98) = 3.90, p < 0.001, but the two HD and LD conditions did not significantly differ from one another, both ps > 0.35 (Medler & Binder, 2005). HD and LD word targets were balanced for SUBTLEX-US frequency (Brysbaert & New, 2009), including on the Zipf scale (see van Heuven et al., 2014), and concreteness (Brysbaert et al., 2014), which are known to affect N400 amplitude (e.g., Dufau et al., 2015; Kounios & Holcomb, 1994), ps > 0.43.

Table 1 Target characteristics [mean (SD)]
Full size table

Each target was presented twice to each participant, preceded by both a TL prime and a substitution prime. TL primes were formed by inversing two adjacent word-internal letters in the target (e.g., leomn-LEMON, vgiht-VIGHT). All transpositions were between a consonant and a vowel. Substitution primes were formed by replacing the letters in those same positions with other letters (e.g., leuzn-LEMON, vpoht-VIGHT) such that the CV structure and visual shape (i.e., ascenders and descenders) of the TL primes were maintained. TL and substitution primes preceding word targets were closely matched for OLD20, both ps = 1.00, and constrained bigram frequency, both ps > 0.68, within each neighborhood condition (Table 2).

Table 2 Prime Neighborhood Densities [mean (SD)]
Full size table

Procedure

The main experiment consisted of 400 trials. Each trial began with a white fixation cross presented at the center of the screen for 500 ms. A forward mask (#######) then appeared for 300 ms, followed by a lowercase prime for 50 ms, a backward mask (#######) for 20 ms, and an uppercase target for 300 ms. This trial structure was identical irrespective of response latencies. Participants were instructed to respond as quickly and accurately as possible, pressing a button with one hand if the word that appeared on the screen was a real word and a button with the other hand if it was not (i.e., no mention was made of the prime). Response hand was counterbalanced across participants. Responses were made with a Logitech F310 and were required on every trial. The screen remained blank for 750 ms after the response before a purple fixation cross appeared for 1,500 ms. We asked participants to try to blink during this purple fixation cross in between trials and during occasional longer breaks.

Stimuli were displayed in Courier font such that they subtended a horizontal visual angle of 1.7 degrees. Trials were presented in one of two pseudorandomized orders. Half of the participants saw any given target preceded by a TL prime in the first half of the experimental list and by a substitution prime in the second half of the list and the other half of participants received the opposite order. TL and substitution primes were equally distributed between the first and second halves of the experimental lists. Orthographic and semantic relatedness of the targets was minimized between consecutive trials and no more than three consecutive trials had the same lexical status or belonged to the same neighborhood density condition. The experiment began with a practice that consisted of 20 trials, half of which had word targets.

EEG Recording

Participants wore an elastic cap (Electro-Cap) with a standard montage of 29 electrodes. Impedances for all electrodes were maintained below 2.5 kΩ. EEG was amplified with SynAmps RT amplifiers (Neuroscan-Compumedics) with a bandpass of DC to 100 Hz and was sampled continuously at 500 Hz. One electrode was placed on each mastoid bone; the electrode on the left mastoid was used as a reference during recording and for subsequent analyses. An electrode next to the right eye was used to detect horizontal eye movements and another electrode below the left eye was used to identify blinks in conjunction with the electrodes on the forehead.

Data Analysis

RT analyses were limited to trials with correct responses between 200 and 2,000 ms (49 trials, or 0.3%, were excluded as outliers). Word and pseudoword target trials were analyzed separately. The RT and error data were analyzed using linear and logistic mixed-effects regression modeling, respectively (Baayen et al., 2008; Jaeger, 2008). Each model included Neighborhood, Prime, and the interaction thereof as fixed effects and a maximal random structure (Barr et al., 2013).

A similar linear mixed-effects (LME) regression approach was adopted for the EEG analyses. Epochs time-locked to target onset extended 1,000 ms, including a 100 ms pre-target-onset baseline, and were low-pass filtered at 15 Hz. Only artifact-free epochs with correct responses between 200 and 2000 ms were included in analyses. Artifact rejection was accomplished in two steps. The first pass was automatic with thresholds for horizontal eye movements (95 μV), blinks (65 μV), and drift at each of the recorded electrodes (190 μV). The manual pass involved verifying that these thresholds were appropriate for each participant and that all artifacts were accurately identified. On average, 19 trials, or 5%, were excluded for artifacts. The average number of trials excluded per condition ranged from 2.17 to 2.69. An ANOVA with factors Target Lexicality (word, pseudoword), Prime (related, unrelated), and Neighborhood (HD, LD) confirmed that there were no significant differences in the number of trials excluded per condition of interest, all ps > 0.09. Given that there is no agreed upon way to incorporate scalp distribution in LME models (cf. Payne et al., 2015), we chose to focus on representative anterior and posterior regions of interest (ROIs) that were analyzed separately (but see Supplementary Materials for grand mean waveforms at all electrodes). The anterior ROI was calculated by averaging the amplitudes at FC1, Fz, and FC2; the posterior ROI was comprised of CP1, Pz, and CP2. Based on the previous masked priming literature, we calculated mean amplitude for each ROI, trial, and participant between 150 and 275 ms (Grainger et al., 2012) after target onset for N250 analyses and between 350 and 550 ms after word target onset for N400 analyses (Ktori et al., 2015; Massol et al., 2010; Meade et al., 2020, 2018a). We expected that the N250 effects would be predominantly anterior relative to the N400 effects, which we expected would be predominantly posterior. As with the behavioral models, each model included Neighborhood, Prime, and the interaction thereof as fixed effects and a maximal random structure (Barr et al., 2013). The full models for word targets at the anterior ROI in both the N250 and N400 windows failed to converge, so we used a simplified model without the random by-participant slope for the main effect of neighborhood. The key predictions revolve around the interaction between Prime and Neighborhood, which indicates that the size of the TL priming effect differs for HD versus LD words. We expected follow-up analyses to indicate that the TL priming effects were significant for both types of words, but larger for LD words than for HD words.

Results

Word targets

Behavior

Behavioral results are presented in Table 3. In RT analyses, a significant main effect of Prime indicated that target words elicited faster responses following TL primes compared with substitution primes (Table 4). A significant main effect of Neighborhood further indicated that HD words elicited slower responses than LD words. Although the TL priming effect was numerically smaller for HD words (25 ms) compared with LD words (33 ms), the Prime × Neighborhood interaction failed to reach significance.

Table 3 Behavioral results [mean (SD)]
Full size table
Table 4 b-, t-values, and standard errors of the reaction time analyses
Full size table

Error analyses revealed a significant main effect of Prime, such that word targets preceded by TL primes elicited fewer errors than those preceded by substitution primes (Table 5). The main effect of Neighborhood was also significant, indicating that LD words elicited fewer errors than HD words. However, the Prime × Neighborhood interaction was not significant.

Table 5 b-, z-values, and standard errors of the error analyses
Full size table

N250. Mean amplitudes for all ERP analyses are reported in Table 6. Significant main effects of Prime at both the anterior and posterior ROIs indicated that word targets preceded by TL primes elicited smaller amplitude N250s than those preceded by substitution primes (Figs. 1 and 2, Table 7). A significant main effect of Neighborhood at the anterior ROI further indicated that HD word targets elicited larger amplitude N250s than LD word targets (Fig. 3). Finally, a significant interaction at the posterior ROI indicated that the TL priming effect was smaller for HD word targets compared with LD word targets (Figs. 1 and 2). In follow-up analyses for each type of word target separately, the TL priming effect at the posterior ROI was significant for both HD words, ß = 0.73, SE = 0.28, t = 2.61, and LD words, ß = 1.65, SE = 0.26, t = 6.37.

Table 6 Mean amplitude and standard deviation of measured windows in μV per condition
Full size table
Fig. 1
figure 1

Grand average ERP waveforms showing the main effect of TL priming for word targets at the anterior and posterior ROIs. Targets preceded by TL primes (dotted line) elicited smaller amplitude negativities than those preceded by substitution lines (solid line). Each vertical tick marks 100 ms and negative is plotted up. The black vertical line marks target onset and the calibration bar marks 2 μV. The grey boxes illustrate the two windows that were analyzed

Full size image
Fig. 2
figure 2

Scalp voltage maps showing the distribution of the TL priming effects (substitution-TL) for HD and LD words within the N250 and N400 windows that were analyzed

Full size image
Table 7 b-, t-values, and standard errors of the word target analyses
Full size table
Fig. 3
figure 3

The left part of the figure illustrates the main effect of neighborhood density over time for word targets at the anterior and posterior ROIs. HD words (red line) elicited larger amplitude negativities than LD words (blue line). Each vertical tick marks 100 ms and negative is plotted up. The vertical line marks target onset and the calibration bar marks 2 μV. The scalp voltage maps to the right show the distribution of the effects (HD-LD) within the N250 and N400 windows that were analyzed

Full size image

N400

As in the N250 window, significant main effects of Prime at both the anterior and posterior ROIs indicated that word targets preceded by TL primes elicited smaller amplitude N400s than those preceded by substitution primes (Figs. 1 and 2). Significant main effects of Neighborhood also indicated that HD word targets elicited larger amplitude N400s than LD word targets at both the anterior and posterior ROIs (Fig. 3). However, there were no significant interactions between the two factors of interest at either ROI. Footnote 2

Pseudoword Targets

Behavior

In the pseudoword target analyses, only the main effects of Neighborhood were significant, such that HD pseudowords elicited slower and less accurate responses than LD pseudowords (Tables 4 and 5).

N250

Significant main effects of Prime at both the anterior and posterior ROIs indicated that pseudoword targets preceded by TL primes elicited smaller amplitude N250s than those preceded by substitution primes (Fig. 4, Table 8). Contrary to the word target analyses, there were no significant interactions between the size of the TL priming effects and neighborhood density.

Fig. 4
figure 4

Grand average ERP waveforms showing the main effect of TL priming for pseudoword targets at the anterior and posterior ROIs. Targets preceded by TL primes (dotted line) elicited smaller amplitude negativities than those preceded by substitution lines (solid line). Each vertical tick marks 100 ms and negative is plotted up. The vertical line marks target onset and the calibration bar marks 2 μV. The scalp voltage maps to the right show the distribution of the effects (substitution-TL) within the N250 and N400 windows that were analyzed

Full size image
Table 8 b-, t-values, and standard errors of the pseudoword target analyses
Full size table

N400

The pattern of results was identical to the word targets in this window. There were significant main effects of Prime and Neighborhood at both the anterior and posterior ROIs, but the two factors did not significantly interact (Figs. 4 and 5).

Fig. 5
figure 5

The left part of the figure illustrates the main effect of neighborhood density over time for pseudoword targets at the anterior and posterior ROIs. HD pseudowords (red line) elicited larger amplitude negativities than LD pseudowords (blue line). Each vertical tick marks 100 ms and negative is plotted up. The vertical line marks target onset and the calibration bar marks 2 μV. The scalp voltage maps to the right show the distribution of the effects (HD-LD) within the N250 and N400 windows that were analyzed

Full size image

Discussion

In the present study, we used TL priming to investigate potential differences in the precision with which HD and LD words are accessed and represented. Following the lexical tuning hypothesis (Forster & Taft, 1994), we expected that HD words would have more narrowly tuned orthographic representations that help differentiate them from surrounding neighbors (see also, Andrews & Hersch, 2010; Grainger, 2008; Meade et al., 2018a). In contrast, LD words that do not resemble many other words may not require such a high level of orthographic precision. Consistent with this line of reasoning, we found evidence that the N250 TL priming effect was significantly larger for LD words compared to HD words, especially over more posterior electrodes. Target words preceded by TL primes continued to elicit smaller amplitude negativities than those preceded by substitution primes into the N400 window, as well as faster and more accurate responses; however, the size of the N400 and behavioral effects did not significantly differ as a function of neighborhood density. Moreover, the size of TL priming effects for pseudoword targets was never significantly modulated by neighborhood density. Taken together, these results demonstrate that TL priming is sensitive to differences in the precision of tuning for HD and LD words, but that ERPs might be required to reliably measure these differences. We use these results as a catalyst to discuss how lexical tuning might be achieved in current models of orthographic processing.

The most general implication of these results is that the size of electrophysiological indices of TL priming effects differ as a function of lexical-level characteristics. Until now, TL manipulations have centered around how the effect changes depending on the letters that are transposed (e.g., vowels versus consonants; Carreiras et al., 2009; Lupker et al., 2008; Perea & Acha, 2009; Perea & Lupker, 2004a; Vergara-Martínez et al., 2011) and their positions within words (e.g., internal versus boundary; Perea & Lupker, 2004c) but not how these processes might vary for different types of words. The finding that neighborhood density modulates the size of TL priming effects confirms that this is a valuable approach for tapping into differences in orthographic tuning. However, the temporal sensitivity of ERPs appears to be necessary for measuring these differences given that we failed to find a significant interaction between prime type and neighborhood density in the behavioral data. Moreover, ERPs provide a means to distinguish between sublexical and lexical processing. Extensive prior work combining masked priming and EEG recordings has systematically revealed effects of sublexical processing on the N250 component, and lexically driven effects on the N400 component. Particularly relevant for the present findings is that there is evidence that the N250 component is sensitive to the mapping of sublexical representations onto lexical representations during visual word recognition (see Grainger & Holcomb, 2009 for a review).

How might current models of orthographic processing accommodate these findings and, more generally, how could they implement a lexical tuning mechanism? The dual-route model proposed by Grainger and Ziegler (2011) is particularly well-suited to implement lexical tuning. Indeed, Forster and Taft (1994) originally envisaged a dual-route structure in order to account for the differences in form priming (e.g., drice-drive) that they found between HD and LD words. Forster and Taft’s initial proposal involved single letter units for LD words and multiletter subsyllabic representations for HD words. In contrast, the starting point of Grainger and Ziegler’s model is the open-bigram coding scheme (Grainger, 2008; Grainger & van Heuven, 2004; Grainger & Whitney, 2004). Open bigrams are formed by taking combinations of adjacent and non-adjacent letters in the correct order (e.g., f-i, f-g, f-h, f-t, i-g, i-h, etc. for the word fight). Such a coding scheme can readily account for TL priming effects given that TL primes share more open bigrams with their targets than substitution primes. In the dual-route model, the open-bigram coding scheme is implemented along the so-called coarse-grained route, which is hypothesized to provide a fast means of mapping sublexical orthographic representations onto lexical representations. In terms of lexical tuning, the approximation generated by this approach is befitting for LD words that do not often share a high proportion of open bigrams with other lexical entries. The fine-grained route is comparatively more precise and might be better suited for HD words. Along this route, letters and commonly occurring multi-letter graphemes (e.g., th, ch) are assigned to specific positions. This precise tuning decreases the effectiveness with which TL primes activate the target lexical representations and should decrease TL priming effects.

Note that in the original instantiation of the model, Grainger and Ziegler (2011) proposed that task demands modulate the relative weight assigned to the two routes, with the coarse-grained route optimized for silent reading for meaning and the fine-grained route optimized for generating sublexical phonology during reading aloud. Here, rather than an all-or-none separation between the two routes, we propose a division of labor implemented along the lines of the lexical tuning hypothesis. More specifically, a common pool of fine-grained and coarse-grained sublexical orthographic representations feed activation forward to a common set of lexical representations. Lexical tuning is implemented by an adjustment of the relative strength of these feedforward connections, most likely achieved during word learning. When a new word with many orthographic neighbors is learned, the connections between fine-grained orthographic representations and lexical representations are strengthened relative to the coarse-grained connections in order to improve discrimination of the new word with respect to known words. The result is that HD words are mainly activated via sublexical representations that are finely tuned with respect to letter position, whereas LD words are mainly activated via coarse-grained representations. Previously established connections are updated as new words are learned; the first words within any given HD neighborhood are first processed using more coarse-grained representations, but eventually processed using finely tuned letter position information as more neighbors enter the lexicon. This not only predicts that TL effects should be greater in LD words compared with HD words, as found in the present study, but also accounts for the pattern of effects seen with word neighbor primes in prior research (Andrews & Hersch, 2010; Meade et al., 2018a). That is, the interfering effect of neighbor primes (e.g., blue-blur) is significantly greater for HD words than for LD words, and particularly so for the participants with the highest spelling skills. As argued by Meade et al. (2018a), the combination of high neighborhood density and good spelling ability would increase the use of fine-grained orthographic representations when processing prime words and increase their capacity to interfere with target word identification.

The interaction between the size of the TL priming effect and neighborhood density here only held within the N250 window, which is generally characterized as reflecting the transition from sublexical processing to lexical processing (Grainger & Holcomb, 2009). This timing would appear to be consistent with our proposed implementation of the lexical tuning hypothesis within the framework of Grainger and Ziegler’s (2011) model. That is, variations in the weights assigned to fine-grained and coarse-grained representations as a function of neighborhood density would impact the way that sublexical orthographic information is mapped onto whole-word orthographic representations. To some degree, the scalp topographies within the N250 window might also support this argument of the two types of words being processed differently. The interaction between Prime and Neighborhood was only significant in this time window at the posterior ROI. Previous work has indicated that more anterior N250 scalp distributions are associated with the sublexical mapping of orthography-to-phonology (e.g., pseudohomophone priming), whereas more posterior N250 scalp distributions are associated with orthographic processing (Grainger et al., 2006). The posterior priming effect that was larger for LD words than for HD words might have been driven by representations along the coarse-grained route that are orthographic, rather than phonological, in nature. These dissociations remain speculative for the time being. What is important to highlight is that TL priming effects can be obtained irrespective of the type of sublexical representations that are utilized, with differences in both the size and the nature of the effects.

The other main class of models of orthographic processing are noisy slot-coding models such as the Overlap model (Gómez et al., 2008) and the Bayesian Reader model (Norris, 2006). In these models, the level of precision of orthographic processing is determined by the amount of noise applied to position-specific letter detectors. TL priming effects occur when this positional noise is great enough to generate evidence that a given letter at a given position is in fact located at a neighboring position. Any given letter detector can be tuned to be more or less tolerant to these changes in position. However, this mechanism would operate independently of the lexical characteristics of stimuli, such as neighborhood density, and therefore cannot accommodate the present findings.

One way that these alternative accounts of orthographic processing might be able accommodate the present findings is if there were another factor at play sublexically. There tends to be a strong correlation between neighborhood density and bigram frequency, and this was true of the stimuli in the present experiment. This relationship prompts the question of whether the N250 results could be explained in terms of these sublexical features or whether they are informative with respect to the sublexical codes that are used to access lexical representations (i.e., the connectivity between sublexical and lexical representations), as predicted by the dual-route model (Grainger & Ziegler, 2011). The pseudoword target data support the latter interpretation. If the interaction was driven solely by sublexical features, then we should have found a similar pattern in the N250 window for pseudoword targets. Although we did find significant TL priming effects on mean amplitude within the N250 and N400 windows for pseudoword targets, neighborhood density never significantly modulated the size of that effect. Thus, the best explanation for the interaction between neighborhood density and TL priming that we found for word targets in the N250 window is in terms of the interface between sublexical and lexical processing.

The absence of a significant interaction in the N400 window suggests that these qualitative differences in the nature of this interface led to similar levels of lexical activation. This was also reflected in the finding that behavioral indices of TL priming were similar for HD and LD words. This pattern of results follows from the principle that the relative weight assigned to different types of sublexical orthographic representation is driven by the goal to optimize orthographic processing and word identification. In other words, the word identification processes reflected in lexical decision responses and the N400 can be equally optimal and independent of the means used to achieve identification. Different types of sublexical orthographic representation can give rise to similar levels of lexical activity since the mapping of sublexical onto lexical representations is assumed to be optimal in all cases.

Conclusions

The present study used TL priming in order to test the hypothesis based on the lexical tuning hypothesis that HD and LD words differ in terms of the precision of their orthographic representations. Differences in TL priming for HD versus LD words were especially prominent during the N250 window, suggesting that it is the way in which lexical representations are accessed that differs between them. We have interpreted this pattern of results within the framework of the dual-route model of orthographic processing (Grainger & Ziegler, 2011). More specifically, we suggest that HD words are more likely to be accessed via fine-grained orthographic representations in order to be differentiated from their neighbors, whereas LD words are more likely to be accessed using coarse-grained orthographic representations. Future studies might extend this approach to examining how other lexical-level factors influence lexical access, as well as how this process differs across adult readers of different skill levels.