Introduction

Adult readers encounter written words in a wide range of fonts, styles, colors, sizes, and letter cases (e.g., table, table, TABLE, table, etc.). Despite these variations, they can identify the correct lexical entry across perceptually similar competitors (e.g., slat and not salt; wine and not wire) in only a few hundred milliseconds. To explain this feat, all leading neurobiological models of visual word recognition propose that these perceptual features are filtered at the beginning of word processing1,2 so that only abstract representations fuel lexical access. After all, regardless of their different appearances, the words table, table, TABLE, and table refer to the same concept.

In the present paper, we investigate whether the same principles that apply to the identification of common words, in particular, letter position coding (i.e., a key element in orthographic processing3), can be extended to a category of words with unique characteristics: logotypes. As text-based graphical objects, logotypes comprise a perceptually consistent brand name presentation (i.e., with a characteristic font, color, and letter case configuration) embedded in a unique visual design (e.g., we can think of the logotypes of Coca-Cola, adidas, amazon, IKEA, etc.). These characteristics are especially helpful in creating visually appealing and memorable brand representations that are easy to identify (see4,5,6, for reviews). These unique properties of logotypes appear to set them apart from common words and suggest that their mental representation may contain not just orthographic representations (i.e., brand names are a special type of words), but also surface information (e.g., font, letter case, color)7.

Previous behavioral studies have shown that the identification of brand names and logotypes is, at a qualitative level, modulated by the same benchmark phenomena as common words. These include the key role of a word’s initial letter, the confusability produced by misspelling, or repetition priming effects, among others (e.g., see5,6,8 for recent research). Notably, as reviewed below, at a quantitative level, visual elements appear to play a greater role in brand names and logotypes than in common words, suggesting that the identification of logotypes may involve not only abstract orthographic information but also surface information (see9,10, for discussion; see also11 for a memory-based study).

The present experiment examined a key marker of orthographic processing (letter position encoding) during logotype identification by presenting intact and misspelled brand names embedded in logotypes while recording event-related potentials (ERPs). Specifically, we tracked the neural signature of a widely known orthographic manipulation: the transposed-letter effect. This effect is often measured by comparing the responses to misspelled items created by transposing two letters from a brand name (e.g., SASMUNG; original brand name: SAMSUNG) to replacement-letter controls (e.g., SARVUNG; see12 for early evidence)—note that this effect can also be examined by comparing the intact stimulus versus the transposed-letter misspelling (e.g., judge vs. jugde, see13, or SAMSUNG vs. SASMUNG in the context of logotypes). Previous research in the areas of visual word recognition and reading has consistently shown that, in tasks such as lexical decision, participants’ responses are slower and less accurate to transposed-letter pseudowords like JUGDE than to replaced-letter pseudowords like JUPTE12,14,15,16 (see also17 for evidence of transposed-letter effects in masked priming paradigms). That is, the transposed-letter pseudoword JUGDE is much more word-like (i.e., more perceptually similar to its base word JUDGE) than the replacement-letter pseudoword JUPTE. This finding has been typically explained by the combination of (1) a generic perceptual uncertainty when encoding the serial order of visual objects (letters, in this case) and (2) a coarse-grained orthographic process specific to letter strings—this latter mechanism explains why transposition effects are greater for letter strings than for other types of strings (e.g., digits, symbols; see15,18,19).

Several behavioral experiments have recently examined letter transposition effects with logotypes. In a task in which participants had to decide whether a given brand name was correctly spelled or not, Pathak et al.5 found faster and more accurate responses to misspelled logotypes when the letter transpositions occurred at the beginning than at the end of the brand name (e.g., afcebook vs. faceboko). Using the same task, Perea et al.9 found that transposed-letter misspelled logotypes (e.g., SASMUNG) produced longer and more errors than replacement-letter misspelled logotypes (e.g., SARVUNG) (i.e., a letter transposition effect). Perea et al.9 also found that the letter transposition effect was greater when the brand names were embedded in their corresponding logos than in plain text format or with a different font. The boost in letter transposition effects for the logotypes embedded in their graphical format is consistent with the idea that their mental representations include both orthographic and surface information.

A drawback of the above-cited behavioral experiments is that they only compared the two misspelled conditions (transposed-letter versus replacement conditions) so that the question of how perceptually close was the transposed-letter misspelled word to the original logotype was left unanswered—note that the intact and misspelled transposed-letter logotypes could not be directly compared because they required different responses (i.e., whether the brand name was correctly spelled or not). In the present experiment, we employed a go/no-go semantic categorization task (i.e., whether the brand corresponded to a means of transportation), where both the intact and misspelled logotypes required the same response. Furthermore, previous  experiments were limited to a single time point towards the end of processing (i.e., decisional stage), thus not providing information on the processing stages where those behavioral effects originate. To obtain a more detailed and time-sensitive measure of the stages underlying the identification of logotypes, we measured event-related potentials (ERPs). ERPs serve as key markers of neural processing that allow us to differentiate the time course and the processing stages affected by the experimental manipulations20. Thus, this technique allows for a more precise characterization of orthographic encoding of (misspelled) logotypes than behavioral experiments. In this way, the methodology applied in the present experiment enhances our understanding of how letter order information is processed in words with unique characteristics, such as logotypes. This advances our knowledge not only of the mental representations of logotypes but also of the broader question of how serial order is encoded in printed stimuli21.

One key ERP component for studying visual word recognition in a semantic context (as done in the present experiment: participants decided whether a given logotype was related to a means of transportation) is the N400. The N400 is a negative deflection starting around 300 ms, with an amplitude peak at around 400 ms after stimulus onset at a centro-parietal scalp distribution. This component is sensitive to lexical-semantic processing of potentially meaningful stimuli and is considered to reflect the processing cost of retrieving properties associated with a word form stored in memory. For instance, pseudowords and low-frequency words evoke larger N400 amplitudes than high-frequency words22,23,24,25. More important for the present study is that not all pseudowords behave similarly. As shown by Vergara-Martínez et al.13, replacement-letter pseudowords produce a larger N400 amplitude than transposed-letter pseudowords, which is consistent with the idea that replacement-letter pseudowords share fewer features with their corresponding words. With high-frequency base words (the more comparable scenario to popular brand names), Vergara-Martínez et al.13 found a similar early N400 (360–470 ms after stimulus onset) amplitude for transposed-letter pseudowords and their intact base words. In contrast, replacement-letter pseudowords showed much larger amplitudes than their original counterparts. Only in the last processing stages (470–580 ms after stimulus onset; late N400) did the electrophysiological brain response for transposed-letter pseudowords differ from base words, with no difference between transposed-letter and replacement-letter pseudowords. To explain this pattern, Vergara-Martínez et al.13 proposed that during the initial stages of lexical-semantic activation (i.e., in the early N400 time window), there was some uncertainty regarding letter position encoding, explaining why the pseudoword jugde produced ERP waves that did not differ significantly from those of the word judge. This uncertainty would be reduced later (late N400), and both jugde and the control jupte would produce ERP waves that are different from the intact word judge (see14,26, for models of word recognition with a progressively reduced uncertainty concerning letter position encoding). Similar findings have been reported by Carreiras et al.27 in another ERP study—note that they only compared transposed-letter and replacement-letter pseudowords without the baseline (word) condition.

Thus, the ERP findings with common words have not only shown that transposed-letter pseudowords activate their lexical representations in memory to a larger degree than replacement-letter pseudowords but also that the presumably fine-grained processing at later verification stages of visual word recognition (around 470–580 ms after stimulus onset) counteracts and reduces this activation. In the present experiment, we examined the neural correlates of the letter transposition effects with logotypes. As stated earlier, previous behavioral research has shown that transposed-letter effects are stronger when the brand names are presented with their graphical information9, suggesting some interplay of surface and orthographic information in the identification of logotypes. In the present experiment, we employed a go/no-go semantic categorization task (“Is the logotype related to a means of transportation?”). The target stimuli were logotypes not related to transportation. We chose this task not only because it allowed for a more direct comparison with the Vergara-Martínez et al.13 ERP semantic categorization experiment but also because the task relies on lexical-semantic access of the brand names rather than on their spelling (i.e., unlike the previous experiments asking whether brand names were correctly spelled, e.g.,5,9), thus increasing the ecological validity of the findings10,11. To ensure that participants did not make their decisions on graphical elements alone, we included a small proportion of mismatched logotype filler stimuli with incongruent visual features—this included non-transportation brand names embedded in transportation logos and transportation brand names embedded in non-transportation logos (examples are provided in the OSF Link at the Data Availability Statement).

The predictions were as follows. If orthographic processing during the identification of a logotype mimics that of common words, we expect to find some flexible position coding from the brand names’ constituent letters in the initial stages of processing that would be eventually resolved (as assumed by the overlap model14 or the noisy Bayesian reader model26). If this is the case, during initial semantic activation stages (early N400), one would expect that transposed-letter misspelled logotypes (e.g., SASMUNG) and original logotypes (e.g., SAMSUNG) elicit a similar amplitude, whereas replacement-letter misspelled logotypes (e.g., SARVUNG) would elicit a larger amplitude (i.e., SAMSUNG = SASMUNG < SARVUNG). Critically, in later processing stages (late N400), we would expect that the uncertainty regarding letter position encoding is reduced so that the ERP waves of transposed-letter misspelled logotypes would resemble those of replacement-letter misspelled logotypes (i.e., SAMSUNG < SASMUNG = SARVUNG). This outcome would extend the ERP findings of Vergara-Martínez et al.13 obtained with common words to logotypes.

Alternatively, one might argue that the identification of logotypes involves not only the processing of orthographic information but also surface information9,10. The logic here is that the graphical information from the logotypes could normalize relatively minor deviations in orthographic processing (e.g., misspellings such as letter transpositions)—note that customers often inadvertently buy counterfeit branded products that look like the original ones28. In this scenario, we would expect to find similar amplitudes for transposed-letter misspelled and original logotypes not only during early semantic access (early N400, as in the Vergara-Martínez et al.13 study with common words), but also in later processing stages (i.e., late N400). As orthographic deviations such as replacing two letters would be more difficult to normalize, we would expect a difference in amplitude between the replacement-letter misspellings and the other two conditions in both time windows (SAMSUNG = SASMUNG < SARVUNG). This outcome would not only have relevance at a theoretical level (i.e., how the mixed nature of logotypes makes their mental representations richer than those of common words) but would also have an applied component, for instance, when deciding if a logotype is too similar to the one from another product, as it is done in trademark infringement cases with (potential) counterfeit brands.

Methods

Participants

We recruited 28 native Spanish individuals (22 self-identified as women). Their ages ranged from 18 to 26 (mean age = 21.4 years, SD = 2.18 years). This sample size nearly doubled the number of responses analyzed in the transposed-letter ERP study of Vergara-Martínez et al.15 with common words, and follows the recommendation from Jensen and MacDonald29 regarding sample size in N400 experiments. All participants had normal or corrected-to-normal vision, were right-handed as assessed with an abridged Spanish version of the Edinburgh Handedness Inventory30, and had no history of reading disorders. The participants gave informed consent before the beginning of the study, which followed the Helsinki Convention. The Research Ethics Committee of the University of Valencia approved this experiment.

Materials

The target stimuli (non-transportation brands; “no-go” responses) were 15 unique brand names (embedded in their logos) that are commonly known in Spain and unrelated to means of transportation (SAMSUNG, amazon, Colgate, MERCADONA, Estrella Galicia, FANTA, STARBUCKS, Google, intel, Disney, LACOSTE, Levi’s, movistar, BURGER KING, NESCAFÉ). These stimuli were the same as in the behavioral lexical decision experiment of Perea et al.9. We created three critical manipulations for the brand names embedded in logotypes: (1) intact version (i.e., original logotype; e.g., SAMSUNG); (2) transposition of two adjacent middle letters of the brand names (i.e., transposed-letter misspelling; e.g., SASMUNG); and (3) replacement of two adjacent middle letters of the brand names (i.e., replacement-letter misspelling; e.g., SARVUNG) (examples and materials are provided in the OSF Link in the Data Availability Statement). As probe items for the go/no-go semantic categorization task, we selected three well-known brand names belonging to a means of transportation (“go” responses; metrovalencia, renfe, RYANAIR). These items had a similar length and logotype complexity as the non-transportation brand names. To ensure that participants were paying attention to the written information from the logotypes, we included a small percentage of filler trials (20% of stimuli) where the non-transportation brand names were also presented in the logo of a transportation brand (e.g., SAMSUNG could be written within the logo of the European airline company RYANAIR; 45 observations) and transportation brand names were presented embedded in the logo of a non-transportation brand (e.g., the Spanish railway company renfe was presented within the Google logo; 45 observations). Following previous experimental designs with a repeated presentation of the brand names5,6,9,10, each version of the stimuli was presented three times throughout the experiment across three blocks—note that these behavioral studies found a comparable pattern of effects across blocks (e.g., see Fig. 3 in10). This resulted in 45 observations per critical condition.

Procedure

Participants were seated in a dimly lit and quiet room in front of a high-resolution computer screen, positioned at eye level at 1-m distance from the participants. As is common in semantic categorization tasks, participants were instructed to press a response button if the brand name belonged to a means of transportation (“go” trials) and refrain from responding if not (“no-go” trials). The stimuli were presented in the following setup: first, a fixation cross was presented on the screen for 800 ms, followed by a blank screen for 100 ms. Then, the stimulus (an intact or misspelled logotype) was presented until the response, with a maximum of 1500 ms. This was followed by a blank screen of random duration between 500 and 900 ms. The stimuli were presented in a random order for all participants. Participants were asked not to blink or move their eyes during the presentation of the stimuli to avoid subject-generated artifacts in the EEG signal. Several 10-s breaks were interspersed approximately every 30 trials for participants to rest their eyes. The experimental session began with 12 practice trials, followed by the 270 experimental trials with three longer breaks in between. The whole experimental session lasted approximately 30 min, excluding the EEG setup.

EEG recording and analyses

The electroencephalogram (EEG) was recorded from 29 Ag/AgCl electrodes mounted in an elastic cap (EASYCAP GmbH, Herrsching, Germany) according to the 10/20 system. The electrodes were referenced to the right mastoid and then re-referenced offline to the right and left mastoid average. Three external electrodes, each placed on the right lower orbital ridge and the left and right external canthi, were used to measure potential blinks and eye movements. The EEG signals were continuously recorded with a sampling rate of 250 Hz by a BrainAmp amplifier (Brain Products, GmbH, Gilching, Germany), and bandpass filtered offline between 0.01 and 30 Hz. Impedances were kept below 5 KΩ. For the EEG data analysis, we used BrainVision Analyzer (Brain Products, GmbH, Gilching, Germany). We first screened all single-trial waveforms for eye movements, muscle artifacts, blinks, amplifier blocking, and drift. The average ERPs or statistical analyses did not include trials containing artifacts or incorrect responses. This resulted in an average rejection rate of 10.4% of all trials with no significant difference in the number of rejections across conditions (p = 0.19). We averaged the ERPs separately for each of the experimental conditions, subjects, and electrode sites.

We performed the statistical analyses on the mean ERP values in three epochs (150–225 ms; 350–500 ms; 500–600 ms) for the three experimental conditions (original logotype, transposed-letter [TL] misspelled logotype, replacement-letter [RL] misspelled logotype). We selected these time windows based on the results of repeated measures t-tests (between conditions) of 4-ms intervals between 1 and 600 ms at all 27 scalp sites. We set the statistical significance level at p < 0.05 for a minimum of 15 continuous data points (60 ms) to minimize false positive rates31 (see also32,33 for a similar approach).

The topographical distribution of the ERP results was analyzed by including the averaged amplitude values across three electrodes of four representative scalp areas (see13,27, for a similar approach): (1) left-anterior (F3, FC5, FC1), (2) left-posterior (CP5, CP1, P3), (3) right-anterior (F4, FC2, FC6), and (4) right-posterior (CP2, CP6, P4). In the analysis, these areas were encoded as the Factors Hemisphere (right, left) and Anterior–Posterior (A/P) distribution (anterior, posterior). We conducted repeated-measures analysis of variance (ANOVA) of the mean voltage amplitudes of each time window with the factors Type of stimulus (original logotype, TL misspelled logotype, RL misspelled logotype), Hemisphere (left, right), and A/P distribution (anterior, posterior). We adjusted critical values using the Greenhouse–Geisser correction, where appropriate. Effects were reported when they interacted with the experimental manipulations. Interactions between factors were followed up with simple-effects tests.

Results

Participants correctly identified transportation brand names (“go” response) in 96.0% of the trials and correctly refrained from responding to non-transportation brand names (“no-go” response) in 97.25% of the trials.

The scalp map and the ERP waves of the three conditions are displayed in Figs. 1 and 2 (ERP waves are represented in black for original logotypes, red for transposed-letter misspelled logotypes, and blue for replacement-letter misspelled logotypes). Visual inspection shows an overall negative potential with a peak around 100 ms post-stimulus onset, a positive frontal P200 between 150 and 200 ms, and a negative deflection starting around 350 ms until 600 ms (N400). Here, the intact and the transposed-letter misspelled logotypes elicit smaller negativities than the replacement-letter misspelled logotypes, a difference that seems to increase with time. The results of the ANOVA for each time epoch are described below. For the pairwise contrasts on Stimulus Category (i.e., original versus RL, original versus TL, RL versus TL), we present t-values with p-values adjusted using the Holm-Bonferroni correction, and Cohen’s d for effect sizes.

Figure 1
figure 1

ERP waves in selected electrodes of the original brand names (black), transposed-letter brand names (red), and replacement-letter brand names (blue).

Full size image
Figure 2
figure 2

Scalp maps displaying the topographic distribution of the effects in the three time-windows of the analysis. Highlighted circles/points correspond to the electrodes F3, F4, P3, and P4, whose ERPs are shown in Fig. 1.

Full size image

150–225 ms epoch

While the ANOVA did not reveal a main effect of stimulus category, F(2,54) = 2.113, p = 0.131, η2 = 0.073, the interaction between stimulus category and A/P distribution was significant [F(2,54) = 3.647, p = 0.03, η2 = 0.119]. At the frontal scalp areas, amplitudes were smaller for replacement-letter misspelled than for intact logotypes (t(27) = 2.602, d = 0.219, p = 0.036), but neither between transposed-letter misspelled and intact logotypes (t(27) = 0.999, d = 0.084, p = . 322), see also Fig. 3), nor between transposed-letter misspelled and replaced-letter misspelled logotypes (t(27) = 1.602, d = 0.135, p = 0.230) (see left panels of Fig. 2).

Figure 3
figure 3

Average voltages (per subject) in raincloud plots and box plots of the intact logotypes (green) and the transposed-letter (TL) misspelled logotypes (orange) for the three time windows.

Full size image

350–500 ms epoch

The ANOVA revealed a main effect of stimulus category [F(2,54) = 7.54, p = 0.002, η2 = 0.218]: amplitudes were larger for replacement-letter misspelled logotypes than for intact logotypes (t(27) = 3.747, d = 0.309, p = 0.001) and transposed-letter misspelled logotypes (t(27) = 2.763, d = 0.228, p = 0.016). Again, the amplitudes of the transposed-letter misspelled and intact logotypes did not differ significantly (t(27) = 0.984, d = 0.081. p = 0.330, see also Fig. 3). No other effects or interactions were found (see middle panels of Fig. 2).

500–600 ms epoch

The ANOVA revealed a main effect of stimulus category [F(2,54) = 14.517, p < 0.001, η2 = 0.350]: amplitudes were larger for replacement-letter misspelled logotypes than for intact logotypes (t(27) = 4.750, d = 0.444, p < 0.001) and transposed-letter misspelled logotypes (t(27) = 4.578, d = 0.428, p < 0.001). The amplitudes of the transposed-letter misspelled logotypes and intact logotypes did not differ significantly (t(27) = 0.172, d = 0.016. p = 0.864, see also Fig. 3); indeed, this contrast showed evidence towards the null hypothesis in a Bayesian ANOVA (BF01 = 8.81) with the default prior Cauchy (0, r = 1/√2), indicating that these data were 8.81 times more likely given the null hypothesis than the alternative hypothesis. We found no other interactions (see right panels of Fig. 2).

As suggested by one reviewer and based on previous results34,35, one might argue that the presentation of each of the 15 brand names across three blocks might have elicited an N400 repetition effect that could have modulated the present findings in unexpected ways. To address this, we conducted an exploratory post-hoc ANOVA including the factor block for each epoch. Results revealed main effects of repetition in the 350–500 ms and 500–600 ms epochs. As in the study by Laszlo et al.35, we observed a decrease in the N400 elicited by all experimental conditions across the blocks. Importantly, we did not find an interaction between repetition and stimulus category, which is consistent with previous behavioral experiments with logotypes (e.g.,9,10). We acknowledge that further research with a larger number of brand names is necessary to fully address this issue. These exploratory statistical analyses are presented in the Supplementary Material S1 (Appendix A).

Discussion

Unlike common words, brand names (e.g., OREO, IKEA, adidas, etc.) are usually presented in a visually uniform format and embedded in graphical objects, as logotypes. Thus, not only the printed (abstract) letters of the word, but also surface information may be relevant in the processing of logotypes7. Recent behavioral word recognition experiments using logotypes as stimuli8,9,10,36 have questioned the universality of the common assumption that, during lexical access, the visual input is mapped onto progressively abstract representations after filtering out perceptual information1. For instance, the magnitude of transposed-letter effects with misspelled logotypes (e.g., SAMSUNG vs. SARVUNG) is greater when the items were presented with their logotypes than when presented with another font or in plain text9, suggesting that, in logotypes, surface information may compensate some minor orthographic deviations.

To better characterize an orthographic marker, such as letter position coding, in logotypes, the present go/no-go ERP semantic categorization experiment examined the time course of orthographic processing in the intact logotypes (e.g., SAMSUNG) against their transposed-letter misspellings (e.g., SASMUNG) and replacement-letter misspellings (e.g., SARVUNG). This procedure allowed us to track the electrophysiological markers of different orthographic processes during brand name semantic processing. Early in processing (around 150–225 ms post-stimulus onset), the electrophysiological response for the replacement-letter misspelled logotypes differed from the intact version. While larger P200 amplitudes were obtained for the intact than for the replacement-letter misspelled logotypes (see Fig. 1), the P200 amplitudes elicited by transposed-letter misspelled logotypes rest somewhere in between, with no statistical differences from the other two conditions. This pattern fits well with previous research showing larger P200 amplitudes as a function of perceptual matching between a visual stimulus and its template stored in memory37,38,39,40 and smaller P200 amplitudes for a more efficient figure/ground segregation mechanism in decision making around 200 ms post-stimulus onset41. As the logotypes misspelled by substituting letters share fewer characteristics with their intact templates stored in memory, we observe a reduced P200 amplitude. Notably, the logotypes misspelled by transposing letters and their correct versions showed no significant differences in their P200 amplitudes, suggesting that, at this point in time, the encoding of letter positions has not yet been completed.

More relevant for our present study are the findings during later semantic processing stages: the amplitudes of transposed-letter misspelled logotypes and intact logotypes did not significantly differ during initial semantic access (early N400), whereas the amplitudes elicited by replacement-letter misspelled logotypes were larger (e.g., SAMSUNG = SASMUNG < SARVUNG)—this again parallels the ERP findings of Vergara-Martínez et al.13 with common words in a semantic categorization experiment. Critically, we found that during the later stages of semantic access (late N400), transposed-letter logotypes did not significantly differ from intact logotypes but differed from replacement-letter logotypes (e.g., SAMSUNG ≈ SASMUNG < SARVUNG). This pattern differs from the one reported by Vergara-Martínez et al.13 with common words, who found a larger amplitude for the transposed-letter pseudowords than for the intact words in this time window (e.g., judge < jugde = jupte). Thus, the present ERP findings reveal that the neural signature elicited by transposed-letter misspelled logotypes is close to the one elicited by correctly spelled logotypes, not only in an early N400 window but also in processing stages as late as 600 ms after stimulus onset (see right panel of Fig. 3). Consequently, the present results reveal that letter position coding in logotypes is highly flexible. Thus, the cognitive system does not seem particularly sensitive to small orthographic changes in transposed-letter misspelled logotypes.

These findings extend previous behavioral findings showing a large transposed-letter effect for brand names embedded in their graphical setting9. They are also consistent with recent memory experiments showing that transposed-letter misspelled logotypes are remembered as if they were presented in their original (correctly written) format11 and with those surveys that show that customers have often unintentionally bought a look-alike product28 (see also5). In the following paragraphs, we discuss the theoretical implications of the boost in the flexibility of letter position coding in logotypes.

A feature that sets brand names apart from common words is that they typically occur in a perceptually consistent format (i.e., with a characteristic font, color, and letter case configuration) and are embedded in a unique visual design. Consequently, the mental representations of logotypes may include both abstract orthographic information and perceptual information (see9,10 for discussion). Critically, most current models of visual word recognition1,2 do not account for the interplay between abstract and surface information in the first moments of visual word recognition. Future models of visual word recognition should consider the processing mechanisms of those words that keep some visual consistency. While this is mainly the case for logotypes, other types of words show some regularities, such as the initial capitalization on proper nouns in Latin-based languages42,43. These ideas can be modeled as exemplar-based models of memory by assuming that words are stored and accessed through memory traces based on previous encounters with those words. Those printed stimuli that are presented in many formats—as occurs with common words—would be fused in “functionally abstract” lexical entries. In contrast, printed stimuli constantly presented in a specific visual format may also contain links to surface information in their lexical entries44. That is, in the case of logotypes, their mental representation would include not only abstract information but also perceptual information (i.e., we can think of the characteristic font and the red color in Coca-Cola). As a result, when accessing the mental representations of logotypes, the combination of abstract and surface information may induce the transposed-letter misspelling SASMUNG to evoke a brain signature that could be comparable to its intact counterpart SAMSUNG. Further simulation work is required to test this hypothesis by manipulating the training regime for printed words45.

Our findings also have an applied perspective. Logotypes are designed to be easily recognizable46; for this reason, designers select specific fonts, colors, and graphical formats to make them unique and more memorable. For instance, as indicated above, when thinking about Coca-Cola, who does not think about the iconic red and white colors and its renowned font in the logotype? While this homogeneity can be quite successful for marketing purposes5,36,46, it can also pose unintended problems: scammers may use small changes in the spelling of the brand names to produce counterfeits (or look-alike brand names) with a similar name to the original one (for instance, adiads instead of adidas). The present experiment provides compelling evidence that consumers may not realize that the brand name is a counterfeit look-alike: the ERP waves of the original brand names do not differ from transposed-letter misspellings (e.g., adidas, adiads), even after 600 ms post-stimulus onset. Thus, in the pursuit of unique perceptual information, the logotypes bear their own burden: it is precisely their distinctiveness and uniqueness that render them more susceptible to counterfeiting. Therefore, our findings are valuable for companies in their efforts to combat counterfeit or look-alike products. An illustrative case involved GODIVA, the renowned chocolate brand, suing DOGIVA, a dog biscuit brand, over intentional imitation (i.e., a letter transposition, where D and G were transposed) of its brand name47. However, in other cases, similarly named competitors have won the trials48,49. These legal battles reveal the complexities within trademark law, demonstrating the inconsistency in adjudicating brand imitation cases. We believe that empirical evidence on the perception of brand similarity, like that provided by our study, will help shape more precise legal guidelines to resolve disputes over brand names and trademarks.

Altogether, the present experiment has shown that logotypes are remarkably immune to small changes in letter position. We found no significant differences in ERP amplitudes between intact and transposed-letter misspelled logotypes throughout all processing stages, whereas the amplitudes for replacement-letter misspelled logotypes deviated from those of the other two conditions. These findings suggest that the mental representations of logotypes are composed of both orthographic elements (i.e., letter position, letter order) and surface elements (e.g., color, font, letter case, graphic design). The interplay between these elements may help normalize the visual input, making minor orthographic differences, such as those produced by transposed-letter misspellings, less noticeable. Further research is necessary to fully characterize the role of orthographic vs. surface elements in logotypes by manipulating orthographic components (e.g., letter similarity, as in SANSUNG vs. SARSUNG, a marker of letter identity encoding) or surface components (e.g., the font, letter-case, or graphic design of the logotype).