Content area
Reading engages complex neural networks integrating visual, phonological, and semantic information. The dual-stream model posits ventral and dorsal pathways for lexical and sublexical processing in the left hemisphere and is well-supported in alphabetic languages. However, its applicability to non-alphabetic scripts remains unclear. The Japanese writing system, comprising Kana (syllabograms) and Kanji (morphograms) with distinct orthographic, phonological, and semantic properties, provides a unique framework to investigate neural dissociation between phonological and orthographic-semantic processing. Previous studies suggest that Kanji relies on the ventral route for whole-word recognition and semantic processing, whereas Kana depends mainly on the dorsal route for phonological decoding via grapheme-to-phoneme conversion; however, their spatiotemporal dynamics remain unknown. Using high-gamma power analysis from electrocorticography recordings in 14 patients with epilepsy and subdural implants, we examined the spatiotemporal neural dynamics of Kana and Kanji reading. Participants completed a visual lexical decision task with Kana and Kanji words and pseudowords. Across 912 electrodes, differential high-gamma power analysis showed that Kanji activated bilateral occipitotemporal fusiform regions early (120–550 ms) and the left inferior temporal gyrus (150–240 ms). Conversely, Kana showed prolonged late activation (270–750 ms) in the left-lateralised superior temporal, supramarginal, and inferior frontal gyri, especially during pseudoword processing. These findings indicate that Kanji relies on bilateral ventral stream earlier, while Kana depends on the left dorsal stream, with slower processing reflecting the extra grapheme-to-phoneme conversion. This underscores the value of non-alphabetic languages in elucidating both universal and script-specific neural mechanisms, advancing a cross-linguistic understanding of the reading network.
1 Introduction
Reading, a fundamental skill for communication in contemporary society, involves a complex neural network that dynamically processes visual, phonological, and semantic information ( Dehaene and Cohen, 2011). Over the past few decades, substantial efforts have been made to elucidate the neurobiological mechanisms underlying reading in alphabetic languages. One influential framework is the dual-route cascaded (DRC) model ( Marshall and Newcombe, 1973; Coltheart, 2006), which proposes two distinct pathways for reading: the lexical/ventral and sublexical/dorsal routes. The ventral route retrieves whole-word representations, which are essential for reading irregular words, while the dorsal route sequentially converts graphemes into phonemes, facilitating the reading of regular words and pseudowords. Neuroimaging studies in alphabetic languages have shown that the ventral occipitotemporal cortex plays a key role in whole-word recognition, whereas the dorsal temporoparietal region supports phonological decoding ( Jobard et al., 2003; Price, 2012; Taylor et al., 2013).
Despite advances in our understanding of the neurocognitive mechanisms of reading in Western languages, the applicability of these models to non-alphabetic scripts remains uncertain. Neuroimaging studies on Chinese reading indicate that while the DRC model is consistent with non-alphabetic scripts, the neural processing of logographic characters follows distinct patterns. Chinese reading elicits bilateral activation of the ventral occipitotemporal cortex in contrast to the left-lateralised activation observed in alphabetic languages ( Bolger et al., 2005; Perfetti and Tan, 1998), suggesting that Chinese reading relies more on perceptual word identification, reflecting the visuospatial demands of logographic characters.
The Japanese writing system, which includes two distinct types of characters—Kana (syllabograms) and Kanji (morphograms)—presents a unique opportunity to study the neural basis of reading. A single Kana character represents a single syllable or mora without inherent meaning, whereas a single Kanji character conveys both sound and meaning. Kanji, comprising >2000 commonly used characters, is often more graphically complex than Kana’s 46 simpler characters. Kana symbols exhibit a one-to-one character-to-sound correspondence, making Kana words/pseudowords regular because of their unambiguous pronunciation, similar to European consonant-vowel syllables. In contrast, Kanji characters and words are generally irregular with subtle regularities, lacking consistent character-to-sound mapping, thus resembling irregular words in European languages ( Sakurai, 2004). While English often uses contrasts between irregular words and pseudowords to explore differences in the two routes of the dual-route model, Japanese script differences within a single language provide a natural framework to examine the neural dissociation between phonological and orthographic-semantic processing.
Previous neuropsychological and neuroimaging studies have identified partially separable neural mechanisms underlying Kana and Kanji reading ( Sakurai, 2019; Suzuki, 2022). Case studies of Japanese patients with alexia and focal brain lesions show dissociative impairments in reading Kana and Kanji. Kana-dominant alexia arises from lesions in the left parietal lobe ( Yamadori, 1975; Sakurai et al., 1998), lateral occipital gyrus ( Sakurai et al., 2001a), and posterior fusiform gyrus (FG) ( Sakurai et al., 2008), primarily affecting the reading of Kana pseudowords. In contrast, Kanji-dominant alexia results from lesions in the left posterior inferior temporal gyrus (ITG) and middle FG ( Iwata, 1984; Kawahata et al., 1988; Kawamura et al., 1987), severely impairing Kanji reading, sometimes with mild Kana alexia characterised by letter-by-letter reading ( Sakurai et al., 2006). A dual-route model for Japanese reading, similar to the DRC model, has been proposed to explain these dissociations, which are primarily associated with left intrahemispheric processes ( Iwata, 1984; Sakurai, 2004). The ventral route, which extends from the visual cortex to the left ITG via the FG, supports orthographic and semantic processing, which is critical for Kanji reading. In contrast, the dorsal route, which extends from the visual cortex to the left posterior superior temporal gyrus (STG) via the lateral occipital gyri and inferior parietal lobule, is responsible for the sequential conversion of graphemes to phonemes based on phonological rules, primarily during Kana reading and especially when processing Kana pseudowords. Kana and Kanji reading engage both routes; however, dissociation may occur when one route is impaired, reflecting the varying degrees of reliance on each route for different types of reading. Neuroimaging studies using positron emission tomography (PET) ( Law et al., 1991; Sakurai et al., 2000, 2001b) and functional magnetic resonance imaging (fMRI) ( Horie et al., 2012a; Ino et al., 2009; Kawabata Duncan et al., 2014; Nakamura et al., 2005; Thuy et al., 2004; Twomey et al., 2013) have revealed distinct activation patterns during Kana and Kanji reading tasks that generally align with the dual-route model. The bilateral inferior posterior temporal regions are consistently activated in Kanji relative to Kana during size judgement ( Thuy et al., 2004), word recognition ( Ino et al., 2009), silent reading ( Horie et al., 2012a), oral reading ( Nakamura et al., 2005), and lexical judgement tasks ( Twomey et al., 2013; Kawabata Duncan et al., 2014), while the left lateral occipital and inferior parietal cortices are predominantly activated in Kana compared with Kanji ( Sakurai et al., 2000; Thuy et al., 2004; Nakamura et al., 2005).
Similarly, Korean—like Japanese—employs dual scripts: Hanja (logograms) and Hangul (phonograms). fMRI studies ( Cho et al., 2014; Kim et al., 2017) have demonstrated that Hangul processing activates parietal regions associated with phonological decoding, whereas Hanja processing engages the extrastriate cortex, FG, and Broca’s area. These findings echo the script-based dissociation observed in Japanese and underscore a broader principle linking script type to distinct neural processing route.
Although significant progress has been made in understanding spatial neural processing during reading, the temporal dynamics of Kana and Kanji reading remain underexplored. Psychological experiments with healthy individuals have revealed that Kana words are read aloud more rapidly, whereas Kanji words are processed for meaning more swiftly, indicating potential differences between the temporal dynamics of Kana and Kanji reading ( Feldman and Turvey, 1980; Hatta and Kawakami, 2003). However, the mechanisms by which these differences are manifested in real-time neural activity remain unclear.
Electrophysiological studies in Western languages have provided valuable insights into the time course of reading-related neural processes. Scalp electroencephalography and magnetoencephalography studies show a posterior-to-anterior cascade of activation, beginning with early visual feature analysis (approximately 100 ms) in the occipital cortex, followed by letter-string analysis (approximately 150 ms) in the left occipitotemporal cortex ( Cornelissen et al., 2003; Salmelin, 2007). Subsequent activation in the left superior temporal and inferior frontal cortices occurs between 200 and 600 ms and in the left posterior parietal cortex between 200 and 800 ms, corresponding to lexical-semantic and phonological processing. The N400 component reflects lexical-semantic processing and is sensitive to word meaning and context ( Kutas and Hillyard, 1980; Pylkkänen and Marantz, 2003). These findings support the DRC model, which suggests distinct pathways for phonological decoding and whole-word recognition. However, non-invasive electrophysiological studies have not consistently demonstrated differences in Kana and Kanji reading that align with the dual-route model for the Japanese language. One event-related potential study showed early bilateral N150 and P200 responses for both Kana and Kanji with no distinctions between them. Nonetheless, the late positive component associated with semantic processing was bilateral for Kanji and left-lateralised for Kana ( Ishiwatari et al., 2002). Another study reported similar left-dominant N170 responses for both Kana and Kanji scripts in Japanese speakers without clear distinctions between them, whereas non-Japanese speakers exhibited bilateral responses ( Maurer et al., 2008). Additionally, a study manipulating the spatial frequency of visual stimuli reported Kanji-specific changes in P100 and N170 latency, suggesting distinct processing mechanisms for Kana and Kanji during early visual processing stages ( Horie et al., 2012b). Magnetoencephalography recordings in response to single-character presentations of Kana and Kanji revealed bilateral dipolar responses over the inferior temporal regions, followed by the perisylvian regions, with no differences between Kana and Kanji reading in terms of location or latency ( Koyama et al., 1998). Given these inconsistent results, electrophysiological techniques with superior spatiotemporal resolution are required to better understand reading dynamics.
Over the last decade, electrocorticography (ECoG), conducted as a preoperative assessment for patients with epilepsy ( Mercier et al., 2022; Parvizi and Kastner, 2018; Rosenow and Lüders, 2001), has significantly advanced our understanding of human language processing ( Juphard et al., 2011; Lachaux et al., 2012; Mainy et al., 2008; Nakai et al., 2017; Woolnough et al., 2021, 2022), owing to its high spatiotemporal resolution and enhanced signal-to-noise ratios ( Dalal et al., 2009). Components of high-gamma band activity (HGBA) have specifically demonstrated correlation with the firing rates of spatially distinct neural ensembles located beneath the electrodes, thereby reflecting event-related localised cortical activity in real-time ( Manning et al., 2009). A previous case study from our laboratory, involving electrodes placed bilaterally on the basal surface of the temporo-occipital cortex, demonstrated selective HGBA for Kanji reading at electrodes on the left FG when a participant performed a visual lexical decision task presented with both words and pseudowords in Kanji ( Tanji et al., 2005). However, because of the limited number of cases involving subdural electrode implantation and because electrodes typically cover only a relatively small portion of the cortical surface, no studies to date have comprehensively evaluated HGBA across multiple brain regions during Kana and Kanji reading.
In this study, we aimed to elucidate the dynamic differences between Kana and Kanji reading by quantifying HGBA across extensive brain regions using ECoG recordings with high spatiotemporal resolution. By registering individual electrodes in standard space using common atlases, we enabled group-level analysis and facilitated comparisons with previous neuroimaging studies. This research addresses two key questions: (1) whether the dual-route model can be extended based on ECoG data, and (2) whether there are temporal differences in neural activity between Kana and Kanji reading. We hope that our findings will enhance the dual-route model by incorporating temporal dynamics, thereby providing insights into the distinct processes involved in Kana and Kanji reading.
2 Materials and methods
2.1 Participants
This study was approved by the Ethics Committee of Tohoku University Graduate School of Medicine (approval number: 20201083) and conducted in accordance with the 1964 Declaration of Helsinki and its later amendments. Written informed consent was obtained from all patients to publish their information and images.
This study included patients aged 16 years and older with intractable epilepsy who underwent subdural electrode implantation at Tohoku University Hospital between December 2018 and August 2023. Language dominance was determined using fMRI, the super-selective Wada test (
Kakinuma et al., 2022,
2024), or electrocortical stimulation. Only patients confirmed to have left-hemisphere language dominance based on these assessments were included in this study. The participants were evaluated preoperatively using the Wechsler Adult Intelligence Scale (Third Edition). Patients with a verbal intelligence quotient below 66, extensive cerebral infarction, or inability to perform the required tasks were excluded from the analysis. The final cohort comprised 14 patients who met the inclusion criteria. Of these, seven had electrodes implanted in the language-dominant hemisphere, three had bilateral electrode implantation, and four had electrodes implanted in the non-dominant hemisphere. Patient demographic characteristics are summarised in
Table 1
2.2 Electrode localisation
Three-dimensional T1-weighted MRIs were acquired for all participants before and after implantation, and computed tomography (CT) images were obtained post-implantation. To accurately localise the electrodes and correct for brain shifts caused by craniotomy, a standardised process was followed in 13 of the 14 cases using post-implantation MRI and CT, as described previously (
Blenkmann et al., 2017). The procedure involved co-registering the post-implantation MRI and CT images, deriving a brain mask from the MRI using FreeSurfer (
2.3 Stimuli and experimental design
All participants completed a visual lexical judgement task involving Kana and Kanji words or pseudowords. Following a previously reported method ( Tanji et al., 2005), four types of stimuli were used: (1) Kana words comprising two or three characters (mean length: 2.9 morae), (2) Kana pseudowords created by rearranging characters from real Kana words, (3) real Kanji words comprising two characters (mean length: 3.9 morae, as a single Kanji character often corresponds to multiple morae), and (4) Kanji pseudowords created by rearranging characters from real Kanji words. This method ensured that visual complexity was consistent when comparing words with their corresponding pseudowords. Each category included 40 stimuli, using only high-familiarity Kana and Kanji words, defined as those with a familiarity rating of 5.5 or higher on a scale of 0 to 7 in the NTT’s Japanese Word Familiarity Database (Nippon Telegraph and Telephone Corp., Tokyo, Japan). According to the Balanced Corpus of Contemporary Written Japanese (National Institute for Japanese Language and Linguistics, Tokyo, Japan), all words used were high-frequency, averaging 73.0 and 67.1 occurrences per million words for Kana and Kanji, respectively. Due to intrinsic structural differences between the scripts, visual complexity could not be fully controlled between Kana and Kanji. As an index of this complexity, the average stroke count was 7.2 for Kana words and 15.7 for Kanji words.
For Kana stimuli, pseudowords retain a fixed pronunciation despite lacking meaning. For example, the Kana word きれい ( ki-re-i, ‘beautiful’) has a fixed pronunciation, and its pseudoword いきれ ( i-ki-re) is read unambiguously despite being meaningless. In contrast, Kanji pseudowords lack both meaning and a single definitive reading, as individual Kanji words have multiple possible pronunciations. While valid Kanji compounds (e.g., 男性 [ dan-sei, ‘male’] and 社長 [ sha-chō, ‘president’]) exhibit fixed readings and meanings, pseudowords like 性社 lack meaning and do not have a single, fixed pronunciation, although readers may attempt to infer one.
The stimuli, totalling 160 items, were presented twice in a pseudo-randomised order (
Fig. 1
Stimuli presentation and recording were performed using a simplified version of the Detroit Procedure, as previously reported ( Kambara et al., 2018). Participants were seated during the experiment, and the stimuli were presented on a 13-inch laptop using Microsoft PowerPoint (Microsoft, Redmond, WA, USA). The stimuli were presented at a viewing distance of approximately 60 cm, with each stimulus displayed for 500 ms, followed by a 3500-ms interstimulus interval ( Fig. 1). The stimulus size for Kanji words/pseudowords was 1.4 cm × 2.8 cm (visual angle: 1.4° × 2.8°). For Kana words/pseudowords, the size was either 1.4 cm × 2.8 cm (1.4° × 2.8°) or 1.4 cm × 4.2 cm (1.4° × 4.2°), depending on the character count. For the visual naming task, the stimuli ranged from 4 to 6 cm (4°–6°) in height and 4–8 cm (4°–8°) in width.
2.4 Data acquisition and signal processing
ECoG signals were recorded at a sampling rate of 1000 Hz using a 192-channel Nihon Kohden Neurofax 1100A Digital System (Tokyo, Japan). Signal processing was performed using MNE-Python software ( Gramfort et al., 2013). Electrodes and epochs with excessive noise and spikes were excluded based on visual inspection, and the data were re-referenced to the average montage. Time-frequency analyses were performed using the standard filter-Hilbert method over a 1500-ms window (500 ms before to 1000 ms after stimulus onset) with a 10-ms step size. The ECoG signals were band-pass filtered at 80–120 Hz, and high-gamma power (HGP) was calculated using the Hilbert transform. HGP was reported as standardised Z-scores relative to baseline, defined as the period from 500 to 30 ms before stimulus onset.
2.5 Statistical analysis and visualisation
Statistical analyses were performed using R (
2.5.1 Electrode-wise analysis
To investigate the temporal dynamics of Kana and Kanji reading, we focused on the post-stimulus 50–750-ms time range, where automatic reading processes primarily occur ( Mainy et al., 2008; Woolnough et al., 2021, 2022), averaging HGP within 100-ms bins (i.e., 50–150 ms, 150–250 ms, …, 650–750 ms) and comparing them using Wilcoxon’s rank-sum tests. Kana pseudowords correspond to those in Western languages, while Kanji words resemble irregular words ( Sakurai, 2004). As the comparison between Kana words and Kanji pseudowords was not informative, we performed the following five comparisons: (1) Kanji and Kana words, (2) Kana words and Kana pseudowords, (3) Kanji words and Kanji pseudowords, (4) Kanji words and Kana pseudowords, and (5) Kanji and Kana pseudowords. The significance level was set at α = 0.05, and the false discovery rate was controlled using the Benjamini–Yekutieli method for multiple comparisons. Significant electrodes were visualised on the cortical surface of the standard MNI152 template.
2.5.2 Region-wise analysis
To compare the dynamics across brain regions, each responsive electrode was assigned to a region based on the Harvard–Oxford atlas (Supplementary Fig. 2), and all epochs recorded from the responsive electrodes within each region were treated as a single population (Supplementary Fig. 3 for representative brain regions). The average HGP, calculated from all epochs across responsive electrodes within each region, was then compared in 10-ms increments from stimulus onset to 750 ms for the same five contrasts mentioned above using Welch’s t-tests with a significance level of α = 0.01 (0.05/5 comparisons). Periods of sustained HGP differences lasting at least 30 ms (three consecutive bins) were identified. For clarity, the labels from the Harvard–Oxford atlas—occipital fusiform cortex, temporal occipital fusiform cortex, and temporal fusiform cortex—were simplified to the posterior, middle, and anterior FG, respectively.
2.5.3 Sub-analysis of Kanji visual complexity in ventral occipital regions
Given the greater visual complexity of Kanji compared with that of Kana, an additional sub-analysis was conducted to evaluate the influence of visual complexity on neural activity in the ventral occipital regions. Kanji words and pseudowords were categorised into three groups based on stroke count. HGP in the left and right occipital poles, FG, and ITG were compared between the high-stroke (top one-third) and low-stroke groups (bottom one-third). It was hypothesised that in regions sensitive to visual complexity, high-stroke Kanji would elicit greater activation compared with low-stroke Kanji. The average HGP from responsive electrodes within each region was compared at 10-ms intervals from stimulus onset to 750 ms, and periods of sustained HGP differences lasting at least 30 ms were identified, as described above. The significance level for this sub-analysis was set at α = 0.05/2.
2.5.4 Sub-analysis excluding participants with lower task performance
To investigate the potential influence of task difficulty on the results, we conducted a sub-analysis excluding four participants (#P3, #P4, #P7, and #P10), each of whom had an overall accuracy below 88 % and scored below 70 % in at least one condition. Both electrode-wise and region-wise analyses were repeated for the remaining participants.
2.5.5 Sub-analysis excluding left-handed participants
To examine the potential influence of handedness-related hemispheric lateralisation, we conducted a sub-analysis excluding two left-handed participants (#P4 and #P8). Both electrode-wise and region-wise analyses were repeated for the remaining participants.
3 Results
Of the 912 electrodes, 340 were responsive, with 179 and 161 located in the left and right hemispheres, respectively.
Fig. 2
3.1 Kanji words vs. Kana words
Fig. 3
Figs. 3B and 6A show a region-wise analysis comparing Kanji and Kana words. High-gamma modulation was initiated in the bilateral visual cortices, extended anteriorly to the bilateral FG, and subsequently left the perisylvian language-related cortices. Activation for both Kanji and Kana words increased simultaneously within each brain region. Kanji word-dominant differential high-gamma modulation occurred earlier, beginning in the left middle FG at 120 ms, left inferior division of the lateral occipital cortex (LOC) at 130 ms, right middle FG at 140 ms, left posterior FG and left temporo-occipital part of the ITG at 150 ms, left posterior part of the ITG at 170 ms, left anterior SMG at 180 ms, left IFG pars triangularis at 260 ms, and left anterior FG at 330 ms. In contrast, Kana word-dominant differential modulation appeared later, beginning at the left superior division of the LOC at 300 ms, left posterior STG at 320 ms, left IFG pars opercularis at 330 ms, left anterior SMG at 340 ms, and left IFG pars triangularis at 630 ms. Notably, in the left anterior SMG and left IFG pars triangularis, Kanji-elicited HGP peaked first, followed by a delayed peak for Kana.
3.2 Kana words vs. Kana pseudowords
In the electrode-wise analysis comparing Kana words and pseudowords, no electrodes exhibited significant HGP differences between Kana words and pseudowords within the first 250 ms post-stimulus (
Fig. 4
In the region-wise analysis, differential HGBA between Kana words and pseudowords emerged later, approximately 270 ms post-stimulus, in contrast to the earlier differences observed between Kanji and Kana words beginning around 120 ms post-stimulus ( Figs. 4B and 6B). Kana pseudoword-dominant differential high-gamma modulation was detected earlier, starting in the left posterior STG at 270 ms, left anterior STG at 310 ms, right anterior STG at 350 ms, left IFG pars opercularis at 590 ms, left anterior SMG at 600 ms, and left IFG pars opercularis at 590 ms. In contrast, Kana word-dominant differential modulation appeared later, beginning at 450 ms.
3.3 Kanji words vs. Kanji pseudowords
In the electrode-wise analysis comparing Kanji words and pseudowords, no electrodes exhibited significant HGP differences between Kanji words and pseudowords within the first 150 ms post-stimulus (
Fig. 5
In the region-wise analysis comparing Kanji words and pseudowords, Kanji pseudoword-dominant differential high-gamma modulation was detected earlier, starting in the left posterior part of the ITG at 190 ms, left middle FG at 240 ms, right middle FG at 250 ms, left anterior STG at 280 ms, left posterior STG at 330 ms, left anterior STG at 370 ms, right anterior STG at 460 ms, right anterior FG at 500 ms, right angular gyrus at 570 ms, right IFG pars triangularis at 610 ms, right temporo-occipital part of the ITG at 710 ms, and left IFG pars triangularis at 720 ms (
Figs. 5B and
6
3.4 Kanji words vs. Kana pseudowords and Kanji pseudowords vs. Kana pseudowords
To avoid redundancy, a detailed discussion of these two contrasts is omitted. In short, the comparison between Kanji words and Kana pseudowords reinforced the patterns observed in earlier analyses (Kanji words vs. Kana words and Kana words vs. Kana pseudowords). Specifically, Kanji words initially dominated the bilateral ventral regions, whereas Kana pseudowords subsequently dominated the left dorsal regions (Supplementary Fig. 7). The contrasts between Kanji and Kana pseudowords revealed Kanji pseudoword dominance in the ventral regions, including the bilateral FG, left ITG, and left SMG. In contrast, Kana pseudoword dominance emerged in the left posterior STG, right inferior LOC, left superior LOC, and left IFG pars triangularis (Supplementary Fig. 8).
3.5 Sub-analysis of visual complexity in ventral occipital regions
The sub-analysis revealed no significant differences in HGP between high- and low-stroke Kanji in any of the examined regions (Supplementary Fig. 9).
3.6 Sub-analysis excluding participants with lower task performance
The sub-analysis excluding four participants (#P3, #P4, #P7, and #P10), who showed reduced accuracy in at least one condition, yielded results consistent with the main findings (Supplementary Figs. 10 and 11).
3.7 Sub-analysis excluding left-handed participants
The sub-analysis excluding two left-handed participants (#P4 and #P8) also replicated the main results (Supplementary Figs. 12 and 13).
4 Discussion
In this study, we employed the high spatiotemporal resolution of ECoG to investigate the neural dynamics underlying Kana and Kanji reading. Our analysis of the HGP across multiple brain regions revealed distinct differences in the phonological and orthographical processing of these two scripts, which have not been captured by previous non-invasive neurophysiological techniques. Notably, while the traditional dual-route model primarily focused on left-hemisphere processes, our findings highlight the significant involvement of the right ventral occipitotemporal cortex in Kanji reading. Furthermore, our results demonstrate that Kana processing predominantly engages the dorsal pathway and occurs at a slower temporal scale than Kanji processing. These findings provide a new dimension to the dual-route model, incorporating spatiotemporal dynamics that extend beyond the left hemisphere and refine our understanding of script-specific processing in Japanese. In the following discussion, we address two key points: (1) the implications of our findings for the dual-route model of Japanese language processing and (2) the temporal dynamics underlying Kana and Kanji reading.
4.1 High-gamma activation in relation to the dual-route model
The dual-route model indicates that the reading process diverges into two pathways in the left hemisphere following visual analysis: the ventral route for orthographic-semantic processing and the dorsal route for phonological processing ( Sakurai, 2004; Iwata, 1984). Previous neuroimaging studies have demonstrated that the ventral occipital-temporal (vOT) region shows greater activation in Kanji than in Kana, whereas the dorsal temporo-parietal-occipital region shows greater activation for Kana pseudowords than for Kana words and for Kana words than for Kanji words. In this section, we discuss whether our HGP analysis supports the predictions of the dual-route model and offers insights for further extensions.
4.1.1 vOT region
Region-wise differential HGP analysis showed that Kanji words elicited greater HGP than did Kana words or pseudowords in the left FG and left posterior ITG, with the earliest differences emerging in the left middle FG. This Kanji dominance in the posterior inferior temporal regions is consistent with the findings of previous lesion studies, indicating that damage to these areas leads to Kanji-dominant alexia with agraphia ( Iwata, 1984; Sakurai et al., 2006; Kawahata et al., 1988). Although vOT activity is influenced by familiarity, our study exclusively presented high-frequency words for both Kana and Kanji, and similar results were obtained even after excluding participants with poor performance in either script condition, suggesting that neither familiarity nor task performance differences account for the observed Kanji dominance ( Twomey et al., 2013). Instead, this activation difference in the vOT may be related to the intrinsic differences between Kana and Kanji, such as the greater complexity of Kanji forms, high spatial frequency information of Kanji, or their increased reliance on morphological processing ( Horie et al., 2012b; Sakurai, 2004). Additionally, delayed Kanji dominance was observed in the left anterior FG, a region associated with word meaning ( Purcell et al., 2014; Lambon Ralph et al., 2017). Damage to the left anterior temporal region has been linked to surface alexia for irregular Kanji words, as seen in semantic dementia ( Tanabe, 2007; Fushimi et al., 2009). Electrocortical stimulation studies have demonstrated that reading impairments in the anterior FG predominantly affect Kanji ( Shimotake et al., 2015). These findings suggest that Kanji reading involves more extensive semantic processing than Kana reading, further highlighting the differential neural mechanisms underlying the processing of these scripts.
Notably, our region-wise HGP analysis also showed Kanji word dominance in the right middle FG, where Kanji words elicit greater activation than Kana words, and Kanji pseudowords elicit greater activation than Kanji words. While significant Kanji–Kana differences were observed in the left hemisphere, the right middle FG was the only right-hemispheric region showing early activation, followed by differences emerging in the left ITG and left anterior FG. Importantly, the Kanji dominance in this right middle FG persisted even after excluding left-handed participants, indicating that it is unlikely to be attributed to handedness-related lateralisation. Although left-hemisphere dominance is a well-established characteristic of alphabetic reading, the right vOT typically contributes to reading only under specific conditions, such as left vOT damage ( Cohen et al., 2004). Similarly, an ECoG study on English reading found no lexical effects in the right middle FG when contrasting irregular words and pseudowords ( Woolnough et al., 2022). In contrast, Kanji has been shown to activate the bilateral vOT in both Japanese and Chinese ( Tan et al., 2001; Sun et al., 2011), and prior fMRI studies have demonstrated increased connectivity from the right to left vOT, specifically for Kanji ( Kawabata Duncan et al., 2014). Our time-series analysis showed that Kanji processing began in the left middle FG, followed immediately by activation in the right middle FG, then lateralised to the left before spreading anteriorly, suggesting left-lateralised integration for subsequent language processing. While visual complexity may partly account for the early right-lateralised activation for Kanji compared with that for Kana in ventral regions, this factor cannot be fully controlled between scripts and thus remains a technical limitation. Nevertheless, we propose that Kanji dominance in the right middle FG is primarily driven by lexical effects, as shown by the superiority of Kanji pseudowords over Kanji words, where visual complexity is matched between the two. Our analysis of the visual complexity in the ventral occipital regions did not reveal any significant differences, suggesting that it did not contribute to the observed activation. These findings highlight that lexical factors, rather than visual complexity, drive the unique bilateral neural dynamics of Kanji reading, in contrast to Kana processing, where such lexical effects are not observed.
Further insight emerged from the individual electrode-wise analysis. While many electrodes in the bilateral FG exhibited Kanji dominance, some in the left FG showed Kana dominance. Although this distribution may partially reflect the limited number of participants, it also highlights the potential significance of the left vOT regions in Kana processing alongside Kanji. Previous electrocortical stimulation studies of the left FG have found that electrodes impairing Kana reading are situated adjacent to those impairing Kanji reading ( Usui et al., 2005, 2009). These findings suggest that certain FG regions may be specialised for the visual and orthographic processing of Kana. Our findings indicate that rather than forming distinct clusters, Kanji- and Kana-processing areas may be broadly distributed within the left FG, creating a mosaic-like pattern in which specific regions predominantly process either Kanji or Kana. Macroscopically, this mosaic pattern may explain the relative Kanji dominance observed when data were averaged across the region. Such mosaic-like activation patterns for different stimuli may be a characteristic of processing in the vOT region. For instance, face-and letter string-selective regions are intermingled in a mosaic-like multizone organisation ( Matsuo et al., 2015).
4.1.2 Left dorsal temporal-parietal-occipital region
Region-wise differential analysis revealed significant differences, with greater activation for Kana pseudowords than for Kana words and for Kana words than for Kanji words in the left posterior STG and left anterior SMG but no significant difference in the left AG. The left STG and SMG are critical for dorsal route processing during reading, particularly in grapheme-to-phoneme conversion ( Sakurai et al., 1998, 2010), and lesions in these regions are known to cause phonological alexia ( Rapcsak et al., 2009). Previous PET studies have reported that the left perisylvian temporoparietal area and superior temporal sulcus are activated by both Kana words and nonwords ( Sakurai et al., 2000), and previous fMRI studies have found that the left SMG shows greater activation in Kana ( Sakurai et al., 2000; Thuy et al., 2004; Nakamura et al., 2005). These findings are consistent with the differential HGP results observed in this study, which underscore the importance of the dorsal route, particularly for phonological processing in Kana.
Interestingly, activation in the left anterior SMG initially showed a brief period of Kanji dominance, followed by a decrease in Kanji activation and a subsequent increase in Kana dominance, despite Kanji words containing more morae than Kana words. Given that the left SMG is associated with the phonological store ( Deschamps et al., 2014), our findings suggest that phonological representations of Kanji words are accessed early through whole-word reading. This is likely because the word form is already processed in the vOT region, minimising the need for grapheme-phoneme conversion through the dorsal route. This interpretation is supported by fMRI studies in Chinese speakers, which have demonstrated that phonological information for Kanji is represented within the ventral network ( Zhao et al., 2017; Li et al., 2022a). In contrast, Kana words and pseudowords may require more extended grapheme-phoneme conversion, as indicated by the sustained Kana-dominant activation in this region.
4.2 Temporal dynamics of Kana and Kanji reading
In this section, we discuss the temporal dynamics of Kana and Kanji reading in the context of the dual-route hypothesis. Previous non-invasive electrophysiological studies have not consistently identified clear temporal distinctions between these scripts ( Ishiwatari et al., 2002; Maurer et al., 2008; Koyama et al., 1998), likely due to limitations in spatial resolution and reliance on single-dipole modelling. Advances in distributed source analysis, which offer higher spatial precision and more comprehensive mapping of distributed neural activity, may help elucidate these differences in future non-invasive studies.
Our HGP analysis partially aligns with these findings, showing that HGBA for both Kana and Kanji begins at similar latencies in each brain region, suggesting that Kana and Kanji are processed concurrently across two distinct pathways. This interpretation is consistent with previous reports indicating that reading both scripts involves both routes simultaneously ( Usui et al., 2009). However, differential HGP analysis revealed that Kanji dominance was evident in the ventral pathway during the early phase (120–550 ms), whereas Kana dominance emerged later in the dorsal pathway (300–750 ms). This contrasting pattern reflects the time intervals during which processing occurs in the ventral and dorsal regions, with Kanji being more reliant on the ventral route and Kana on the dorsal route, leading to different processing times.
The DRC model hypothesises that the dorsal route generally operates slower than the ventral route. The ventral route primarily generates phonological output for real words through whole-word reading, often completing this process before the dorsal route completes phonological processing. In contrast, pseudoword reading relies on the slower dorsal route, leading to prolonged processing time and greater cognitive effort within the dorsal pathway ( Taylor et al., 2013). We propose that a similar relationship exists between Kanji and Kana words. As Kanji relies more on the faster ventral route, once ventral processing is complete, phonological information is rapidly accessed, quickly concluding dorsal processing. Conversely, the prolonged activation in the dorsal pathway during Kana reading indicates that although it occurs simultaneously with Kanji processing in the ventral pathway, Kana processing requires additional processing in the dorsal route. Kana reading involves sequential grapheme-to-phoneme conversion for each character, followed by the synthesis of these phonemes into a word, a process that extends the processing time in the dorsal pathway.
Previous ECoG studies on English reading have shown contrasts between irregular words and pseudowords, revealing that lexical effects initially emerge in the FG, followed by activation in the inferior parietal sulcus and IFG ( Woolnough et al., 2022). Similarly, a study on French reading demonstrated a word-length effect in the left SMG, with longer pseudowords showing prolonged activation, indicating that processing in the dorsal route persists longer than that in the ventral route ( Juphard et al., 2011). While these observations are largely consistent across alphabetic languages, the temporal dynamics of Kana and Kanji reading suggest that the dual-route model, originally developed for alphabetic languages, can be extended to the difference between logographic and syllabic scripts. That is, Kanji engages the ventral route with faster lexical-semantic processing, whereas Kana relies on the dorsal route for sequential grapheme-to-phoneme conversion. The coexistence of Kana and Kanji provides a unique model for examining how lexical and sublexical processing unfolds with temporal separation within a single language. These findings highlight that, despite script differences, the neural mechanisms for reading across languages share common principles, expanding the traditional dual-route model to accommodate diverse script types.
While our findings broadly support the dual-route model, important questions remain regarding its applicability to non-alphabetic writing systems, particularly those based on morphographic or logographic principles ( Li et al., 2022b). In Chinese, for example, the existence and functional relevance of a dorsal phonological route have been debated due to lack of systematic grapheme-to-phoneme correspondence and predominant reliance on a ventral, direct semantic pathway. Although some studies have applied the dual-route framework to classify alexia subtypes in Chinese ( Wang and Yang, 2014), its cross-script generalisability remains unresolved. Similarly, the functional significance of dorsal activation during Kanji reading is not yet fully understood. Our results indicate that dorsal activity during Kanji is relatively weak and temporally restricted, in contrast to the sustained dorsal engagement observed during Kana or alphabetic reading. This suggests that the dorsal and ventral pathways may not operate in parallel for Kanji as they do for phonographic scripts. Instead, dorsal involvement during Kanji reading may reflect automatic access to phonological word forms, rather than the character-to-sound conversion processes characteristic of Kana or alphabetic decoding. These findings highlight the need to adapt the dual-route model to accommodate the structural and functional diversity of writing systems such as Kanji and Chinese. In this context, incorporating evidence from non-alphabetic languages is essential for advancing our understanding of the universal and script-specific neural mechanisms of reading. Multiscriptal languages like Japanese provide a unique empirical setting for testing and refining theoretical reading models, enabling the development of a broader and more flexible framework for understanding cross-linguistic variation in reading mechanisms.
4.3 Limitations and future directions
Our study has some limitations. First, the relatively small sample size and heterogeneous electrode coverage limit both the generalisability and statistical power of our findings. The spatial distribution of implanted electrodes was determined based on clinical needs, which limited access to certain regions implicated in reading, such as the intraparietal sulcus. Additionally, the limited number of electrodes in speech motor regions—such as the central opercular cortex and IFG pars opercularis—precluded definitive conclusions about their role in reading. The relatively weak HGP responses in the precentral gyrus may reflect signal averaging or the low articulatory demand of the binary (yes/no) response task. Future studies incorporating overt reading tasks and broader electrode coverage in larger cohorts are needed to clarify the role of motor-related language areas and validate our present findings.
Second, the inclusion of patients with epilepsy introduces potential limitations in generalising our findings to the broader population, as focal epilepsy can alter typical patterns of hemispheric lateralisation and regional functional organisation ( Hamberger and Cole, 2011; Berl et al., 2014; Gil et al., 2020). To mitigate these concerns, we adopted two standard safeguards commonly used in intracranial language mapping studies: (i) all participants had confirmed left-hemisphere language dominance, determined through the Wada test, fMRI, or electrical stimulation mapping ( Tzourio-Mazoyer et al., 2017); and (ii) electrodes located within seizure onset zones or over structural lesions were excluded from time-frequency analyses to avoid artefacts and functional reorganisation ( Kojima et al., 2012, 2013; Wu et al., 2011; Nakai et al., 2017), thereby minimising artefactual distortion from interictal activity and local functional reorganisation linked to epileptiform discharges. Although ECoG evidence on reading specific networks remains limited ( Wu et al., 2011), prior high-gamma mapping studies in epilepsy cohorts—using tasks such as auditory naming, picture naming, and single-word reading—have consistently demonstrated activation patterns that align with findings from non-invasive studies in neurotypical individuals, thereby supporting the validity of our approach ( Kojima et al., 2012, 2013; Wu et al., 2011; Nakai et al., 2017; Kambara et al., 2018). These findings suggest that ECoG can provide insights beyond clinical populations. Nevertheless, further research with larger and neurotypical cohorts is necessary to confirm the generalisability of our findings. While invasive studies in patients with epilepsy—such as direct cortical stimulation ( Penfield and Jasper, 1954; Ojemann, 1991) and lesion-deficit correlations ( Scoville and Milner, 1957; Rasmussen and Milner, 1977)—have significantly advanced cognitive neuroscience, our results should be regarded as provisional until corroborated by complementary methods in non-clinical populations.
Third, we did not measure response times during the lexical decision tasks, limiting our ability to directly link decision-making processes to the temporal dynamics of Kana and Kanji reading observed in the ECoG data. Although response-time differences are unlikely to influence early visual processing in the vOT regions, previous studies on word reading ( Woolnough et al., 2022) have shown that late-stage high-gamma responses—particularly in the auditory cortex beyond 500 ms—are delayed for pseudowords compared with real words, likely reflecting longer processing times. Incorporating response-time measurements in future research would enable a more precise interpretation of late neural activity and its relation to cognitive processing. Fourth, our study focused exclusively on high-familiarity words, and we did not manipulate variables such as word length, frequency, or familiarity, which are known to influence the reading process. Future research examining these variables could provide deeper insights into the dual-route model and the differential effects on Kana and Kanji processing.
Additionally, recent studies on direct electrical stimulation have highlighted the importance of neural fibres in both Kana and Kanji reading ( Tamai et al., 2022). The anterior-inferior portion of the inferior longitudinal fasciculus has been implicated in Kanji reading, while the posterior-superior portions of the arcuate fasciculus, posterior superior longitudinal fasciculus, inferior fronto-occipital fasciculus, and inferior longitudinal fasciculus appear to support Kana reading. However, no studies to date have investigated Kana and Kanji alexia using diffusion tensor imaging. Future studies combining stereotactic electroencephalography with diffusion tensor imaging could help delineate the broader brain networks, including white matter pathways, involved in reading different scripts, thereby providing a more comprehensive understanding of the neural mechanisms underlying Japanese reading.
5 Conclusion
This study used the high spatiotemporal resolution of ECoG to elucidate the neural dynamics of Kana and Kanji reading. Our findings extend the dual-route model by revealing distinct activation patterns for each script. Kanji reading primarily engages the bilateral ventral route, with early activation observed in the occipital-temporal regions, whereas Kana reading involves prolonged activation in the left-lateralised dorsal route, particularly during pseudoword processing. These results underscore the temporal and spatial differences in processing these scripts, advancing our understanding of the neural mechanisms underlying Japanese script reading. Future studies that incorporate expanded electrode coverage and electrostimulation mapping will further refine these insights into reading dynamics.
Data statement
Readers can contact the corresponding author for a data request. Access will be granted to qualified individuals in compliance with the ethical guidelines governing the reuse of sensitive data. Requesters must obtain approval from the ethics committee of Tohoku University Graduate School of Medicine to access the data.
Funding
This work was partially supported by the
CRediT authorship contribution statement
Kazuto Katsuse: Writing – original draft, Visualization, Methodology, Investigation, Funding acquisition, Formal analysis, Data curation, Conceptualization. Kazuo Kakinuma: Writing – original draft, Methodology, Investigation, Formal analysis, Data curation, Conceptualization. Shin-Ichiro Osawa: Writing – review & editing, Resources, Methodology, Investigation, Data curation. Shoko Ota: Writing – review & editing, Investigation, Data curation. Hana Kikuchi: Writing – review & editing, Investigation, Data curation. Ai Kawamura: Writing – review & editing, Investigation. Kazushi Ukishiro: Writing – review & editing, Methodology, Data curation. Kazuyo Tanji: Writing – review & editing, Methodology, Conceptualization. Nobuko Kawakami: Writing – review & editing, Investigation. Chifumi Iseki: Writing – review & editing, Investigation. Shigenori Kanno: Writing – review & editing, Investigation. Yuichiro Shirota: Writing – review & editing, Investigation. Masashi Hamada: Writing – review & editing, Investigation. Tatsushi Toda: Writing – review & editing, Investigation. Hidenori Endo: Writing – review & editing, Methodology, Investigation. Nobukazu Nakasato: Writing – review & editing, Methodology, Investigation. Kyoko Suzuki: Writing – review & editing, Supervision, Methodology, Investigation, Funding acquisition, Data curation, Conceptualization.
Declaration of competing interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgements
We sincerely appreciate all the patients who participated in this study and their families for their invaluable cooperation and support.
Supplementary materials
Supplementary material associated with this article can be found, in the online version, at
Appendix Supplementary materials
Table 1
| Participant no. | Age (years) | Educational history (years) | VIQ | Handedness | Language Dominance | Implanted hemisphere | Channel numbers | Epileptic foci | Aetiology |
| P1 | 40 | 12 | 90 | R | L | L | 72 | L T | AVM |
| P2 | 16 | 9 | 109 | R | L | L | 102 | L T | Unknown |
| P3 | 22 | 15 | 90 | R | L | L | 45 | B O | Ulegyria |
| P4 | 18 | 11 | 67 | L | L | L | 39 | L O | FCD |
| P5 | 26 | 10 | 85 | R | L | L | 55 | L T | Ganglioglioma |
| P6 | 22 | 13 | 100 | R | L | L | 38 | L C | FCD |
| P7 | 21 | 12 | 83 | R | L | L | 37 | L C | FCD |
| P8 | 26 | 12 | 80 | L | L | B | 37 | R T | HS |
| P9 | 19 | 12 | 70 | R | L | B | 94 | R I | Post-encephalitis |
| P10 | 37 | 12 | 71 | R | L | B | 80 | R T | Ganglioglioma |
| P11 | 26 | 16 | 81 | R | L | R | 116 | R O, T | HS |
| P12 | 30 | 16 | 92 | R | L | R | 71 | R T | HS |
| P13 | 24 | 16 | 93 | R | L | R | 55 | R T | Encephalocele |
| P14 | 28 | 12 | 88 | R | L | R | 51 | R C | FCD |
Table 2
| Participant no. | Accuracy (%) | Participant no. | Accuracy (%) |
| P1 | 88.5 | P8 | 94.3 |
| P2 | 98.1 | P9 | 95.5 |
| P3 | 70.1 | P10 | 85.0 |
| P4 | 80.1 | P11 | 97.5 |
| P5 | 96.2 | P12 | 96.2 |
| P6 | 95.2 | P13 | 96.5 |
| P7 | 87.3 | P14 | 92.7 |
© 2025 The Authors