Seeking the neural traces of statistical learning during implicit processing of visual words

doi:10.21203/rs.3.rs-2417846/v1

Download PDF

Article

Seeking the neural traces of statistical learning during implicit processing of visual words

https://doi.org/10.21203/rs.3.rs-2417846/v1

This work is licensed under a CC BY 4.0 License

You are reading this latest preprint version

Statistical learning (SL) plays a key role in literacy acquisition. Studies have increasingly revealed the influence of SL on visual word processing, including the effects of word frequency at the lexical level and mappings between orthography, phonology, and semantics at the sub-lexical level. However, there has been scant direct evidence supporting neural representations of statistical regularities in visual word processing. Using time-resolved representational similarity analysis (RSA), the present study examined neural representations of different types of statistical regularities in visual word processing. From the perspective of predictive coding, an equal probability sequence with low built-in prediction precision and three oddball sequences with high built-in prediction precision were designed with consistent and three types of inconsistent (orthographically inconsistent, orthography-to-phonology inconsistent, and orthography-to-semantics inconsistent) Chinese characters as visual stimuli. In the three oddball sequences, consistent characters were set as the standard stimuli (probability of occurrence p = 0.75) and three types of inconsistent characters were set as deviant stimuli (p = 0.25), respectively. In the equal probability sequence, the same consistent and inconsistent characters were presented randomly with identical occurrence probability (p = 0.25). Significant neural representation activities of character frequency were observed in the equal probability sequence. By contrast, neural representations of sub-lexical statistics only emerged in oddball sequences where short-term predictions were shaped. These findings reveal that the statistical information obtained through long-term SL continues to play a role in current word processing mechanisms and these mechanisms can be modulated by short-term predictions.

Biological sciences/Neuroscience/Cognitive neuroscience/Language

Biological sciences/Neuroscience/Cognitive neuroscience/Reading

EEG

Statistical learning

Visual word recognition

Prediction error

Representational similarity analysis

Reading is an important social skill that enables us to extract the wisdom of others from a range of symbols. These symbols are visual words with various orthographic features. Proficient reading requires the assimilation of statistical regularities present in the writing system (Siegelman et al., 2020). This statistical structure includes both information at the sub-lexical level, such as orthography-to-phonology (O-P) consistency, orthography-to-semantics (O-S) consistency, and orthographic regularities (He & Tong, 2017; Tong et al., 2020a, 2020b, 2020c; Zhao et al., 2018), and information at the lexical level such as word frequency (i.e., the rate of occurrence of an orthographic form) that also captures some of the statistical structure in the mappings from O-P and O-S at the whole-word (lexical) level (Siegelman et al., 2020).

The acquisition of such regularity information is considered to depend on Statistical Learning (SL), i.e., learning from the distributional properties of sensory input (Frost et al., 2019). Extensive studies have demonstrated that SL is involved in a wide range of basic and higher-order cognitive functions including spoken and written language (Saffran et al., 1996; Aslin et al., 1998; Christiansen, 2019; Frost et al., 2019). SL can occur implicitly, such as when stimuli are presented passively without any explicit task attached (e.g., Fiser & Aslin, 2001, 2002; Saffran et al., 1999; Toro, Sinnett, & Soto-Faraco, 2005), or when participants are engaged in an unrelated cover task (Saffran et al., 1997; Turk-Browne, Scholl, Chun, & Johnson, 2009; Turk-Browne et al., 2005). SL is thus described as occurring ‘‘incidentally’’ (Saffran et al., 1997), ‘‘automatically’’ (Fiser & Aslin, 2002), or ‘‘without intent or awareness’’ (Turk-Browne et al., 2005). These features of SL are highly similar to those of implicit learning, a term coined by Reber (1967) and defined as ‘‘adaptation to the regularities of the world that evolves without intention to learn, and without clear awareness of what we know” (Perruchet & Pacton, 2006, pp. 233). SL and implicit learning are therefore lumped together as ‘‘implicit statistical learning” (ISL) (Arnon, 2019; Christiansen, 2018; Conway & Christiansen, 2005; Perruchet & Pacton, 2006; Turk-Browne et al., 2005).

Behavioral studies have already suggested that SL plays an important role in the process of literacy acquisition and later on in guiding efficient reading. Some of these studies have proposed that patterns of reading behavior can be best explained when considering the rich array of statistical regularities that are embedded in the written input (see, e.g., Frost, 2012; Sawi & Rueckl, 2019; Seidenberg, 2011 for reviews). Other studies have estimated individuals’ SL abilities using separate tasks (such as an embedded triplet task or artificial grammar learning paradigms) and have examined the relationship between reading behavior and performance on these tasks. These studies have tied SL performance to variation in different aspects of reading skills both in participants’ first language (Arciuli & Simpson, 2012; Spencer et al., 2014; Torkildsen et al., 2019) and second language (Frost et al., 2013). However, the external validity of these behavioral studies is limited because they typically involve the repetition of a small number of artificial patterns at uniform probabilities over just a few minutes. The complexity of the regularities in a given domain, whether a spoken language or a printed text, is often significantly different from these simplified learning problems.

A major review recently highlighted the importance of evidence from tasks that tap regularities characteristic of real-world environments across different domains (Bogaerts et al., 2021). The authors suggested that future work should focus on characterizing different statistical environments across a multitude of cognitive domains. In fact, there has been a study that has made attempts in this direction (Siegelman et al., 2020). This study used an alternative approach that focused on identifying individual differences in children's reliance on statistical regularities as reflected directly in their word naming behavior. The researchers found that the measures of reliance on O-P and O-S had much stronger predictive power than the much weaker correlations observed in correlational studies of “typical” SL tasks and reading outcomes. The authors argued that these results suggest that these more complex regularities are the ones that play a role in reading acquisition, more so than the simplified regularities typically studied in SL paradigms. This makes us realize that the Chinese writing system, which has rich quasi-regularity and distributional properties, may provide excellent material for examining the processing of complex statistical regularities in reading. This marks written Chinese as being apart from alphabetic languages, which have highly consistent O-P mapping and arbitrary O-S mapping (asymmetrical) (Zhao et al., 2014). By contrast, Chinese characters convey a relatively symmetric statistical structure between the O-P mapping and O-S mapping. This allows us to better manipulate various statistical attributes of words independently. Specifically, characters that are inconsistent in any one dimension may be consistent in the other dimensions (i.e., 银/yin/, to silver, is inconsistent for common pronunciation, /hen/, but consistent with the meaning category of 钅 as a metal-related concept, and has a common radical position). Therefore, we can manipulate orthographic, phonological, and semantic consistency simultaneously within a single Chinese character.

Neurophysiological prediction error (PE) responses are thought to signify implicit statistical learning (ISL) ability: an individual’s accuracy in their implicit learning of the statistics of sensory events (Lecaignard et al., 2015; Southwell and Chait, 2018; Dzafic et al., 2020). The sensory PE response is commonly measured using electroencephalography (EEG) and a passive oddball paradigm, in which surprising sounds are embedded in a sequence of predictable sounds. Within this paradigm, the predictable sounds (standards) are subtracted from the surprising sounds (deviants) to obtain the mismatch negativity (MMN), which is interpreted as the neural manifestation of PE (Friston, 2005; Garrido et al., 2008, 2009; Den Ouden et al., 2012; Stefanics & Czigler, 2012). The strength of the PE response is determined by the volatility of the external environment. In a stable environment (i.e., one in which sensory information varies less), individuals form precise predictions about forthcoming sensory stimuli, and their brains consequently produce large PE responses to events that violate such predictions (Wacongne et al., 2012; Dzafic et al., 2020). By contrast, the PE response will be reduced in the face of an ever-changing, volatile environment. Several studies have confirmed the relationship between PE and spoken language ISL (Paraskevopoulos et al., 2012; Koelsch et al., 2016). However, although the PE response in the visual modality has been measured using the revised oddball paradigm, this method has not been properly utilized in SL studies related to reading (i.e., written language). This raises two important questions. The first is whether statistical regularities in real-world written language discerned through long-term SL are reflected in the neural processing of visual words by skilled readers. If the answer is yes, the second question is whether the neural activities in response to these statistical regularities are continuously modulated by prediction precision and PE in the current sensory environment.

The present study aims to address these questions and considers the real-world statistical regularities of the Chinese writing system. We designed an experiment using the passive oddball paradigm containing equal probability sequences. For the materials, we carefully selected three consistent Chinese characters and nine inconsistent Chinese characters (see the Materials section). These inconsistent characters were divided into three categories, each with low consistency only in a particular sub-lexical dimension (orthographic, phonological, or semantic). Equal probability sequences were used to provoke the neural activities related to visual words that depended on the participants' long-term experience. There was no clear predictive evidence in this sequence, and all the characters were presented randomly. Oddball sequences were used to provoke the neural activities related to the same characters when the participants were given clear short-term predictions. In these sequences, consistent characters were repeated with a high probability (p = 0.75) and one of the three types of inconsistent characters occasionally appeared with a low probability (p = 0.25), giving a total of three oddball sequences. To track the neural representations of various statistical regularities, we performed a time-resolved representational similarity analysis (RSA) for the electrophysiological activities of the corresponding characters separately in the oddball (integrating all three oddball sequences) and equal probability conditions. This method allows the assessment of any concordance between the neural and cognitive representational spaces, and thus may have the potential to disentangle the specific contribution of each process (Cichy et al., 2015; Giari et al., 2020). Based on the widely reported robust effects of SL on reading in behavioral studies, we hypothesize that significant neural representations of statistical information can be detected even during implicit visual word processing. Starting from the principle that the volatility of the environment will modulate the intensity of PE related neural activities, we propose a second hypothesis that sub-lexical statistical information will be detected stronger neural representation activities in the oddball condition than in the equal probability condition.

Ethics statement

All participants gave oral and written, informed consent in accordance with procedures that were approved by the ethics committee at the School of Psychology, Shaanxi Normal University (Approval No. HR 2021-05-002). The protocols adhered to the Declaration of Helsinki.

Participants

Forty-eight healthy young adults were recruited, with three being excluded for excessive EEG artifacts. The final sample of 45 right-handed (via self-report) young adults (mean age = 18.04, SD = 0.80; 35 females) had an average of 12.3 years of education, and all had normal or corrected-to-normal vision and hearing. The final sample size surpassed that of similar work using EEG to investigate implicit character recognition during the oddball paradigm (Shtyrov et al., 2013; Wang et al., 2013; Wei et al., 2018; Hu et al., 2020) and is comparable to other EEG studies using RSA analysis to explore the representation dynamics of language processing (Hubbard et al., 2021; Hauptman et al., 2022). Participants were recruited from the undergraduate and postgraduate student population at Shaanxi Normal University and were paid 60 RMB for their participation. All participants reported no speech or hearing problems and had no prior history of neurological or psychiatric abnormalities.

An additional group of 33 paid healthy college students (19 female, mean age = 21.48 years, SD = 2.33 years) were recruited to rate the orthographic consistency, phonological consistency, and semantic consistency (transparency) of each Chinese character we selected. Take the scoring of semantic consistency, one question was asked to measure semantic transparency: “To what extent do you think the radical "X" can represent the meaning of the Chinese character "Y"?”. For example, for the character “洋”, the question was “To what extent do you think the radical "氵" can represent the meaning of the Chinese character "洋" ?”. In a similar way, each dimension of consistency was measured on a seven-point scale, with 1 = totally inconsistent and 7 = totally consistent.

Materials

There were four different sets of Chinese phonograms selected. Each set contained four characters (each row in Fig. 1A). In Chinese, a character is generally made up of a semantic radical and a phonetic radical, known as “phonograms”. For example, the character “牲” consists of a semantic radical “牜” and a phonetic radical “生”. Each phonogram set in the present study has a fixed phonetic radical and four different semantic radicals, it enables us to quantify orthographic consistency features based on phonetic radical (Fig. 1A). In the second row in Fig. 1A, for example, the phonetic radical in all characters is “生”, while the semantic radicals are “牜”, “⺮”, “月” and “忄”, respectively.

Semantic radicals are generally on the left side of characters, and phonetic radicals are on the right, which is the main orthographic rule in the majority (63%) of phonograms (Myers, 2019). Semantic radicals usually provide semantic information of characters, while phonetic radicals provide phonological clues. For example, in the Chinese character “牲”, the semantic radical “牜” is on the left side, while the phonetic radical “生” is on the right. The meaning of character “牲” is “domestic animals”. This means that it can easily be speculated from the semantic radical “牜”, which refers to “cattle”. Phonologically, the pronunciation of “牲” is “sheng”, which is also highly consistent with the sound of the phonetic radical “生” (sheng). Characters like “牲” are consistent characters. However, there are some characters (i.e., inconsistent characters) that do not follow the orthographic (positional), phonological and/or semantic rules (e.g., “笙”, “性”, “胜”). Accordingly, these 12 Chinese characters were then divided into three consistent and nine inconsistent characters. The nine inconsistent characters were further divided into three categories, including inconsistent orthographic (IOr) characters, inconsistent phonology (IPh) characters and inconsistent semantics (ISe) characters. Each category has three characters that come from three different sets. In other words, all four categories (i.e., 1 consistent and 3 inconsistent categories) have the same number of characters and the same phonetic radicals (see Fig. 1A). The inconsistent categories differ from the consistent category with regards to orthographic, phonology and semantics, respectively.

Specifically, compared to the consistent category, IOr characters differ with regards to the structure of characters. That is, phonetic radicals appear on the right side of a consistent category. On the contrary, phonetic radicals in the IOr characters are in the less common position in a character. For example, phonetic radial “生” posits at the right side of a “牲”, but it is at the bottom of the IOr character “笙”. Among all characters that are “艮”, “生” and “羊”, they act as phonetic radicals. The probabilities that “艮”, “生” and “羊” appear on the right side of characters are 72.22%, 50.00% and 72.73%, respectively. On the other hand, the percentages of the character positions for “痕”, “笙” and “痒” are 5.56%, 25%, and 18.18%, respectively (Supplementary Table 1). Furthermore, the position of phonetic radicals in IPh and ISe categories are on the right side, the same as consistent characters.

Similarly, IPh characters differ from consistent characters (CC) with regards to phonological consistency. The pronunciations of CC are high in phonological consistency of corresponding phonetic radicals, while the phonological consistency is low in IPh characters. The phonological consistency is defined as the proportion of a specific pronunciation among all characters that adopt the same phonetic radicals (see Borleffs et al., 2017; Siegelman et al., 2020). High consistency refers to the pronunciation of a regular character that is the main pronunciation of all Chinese characters utilizing the specific phonetic radical. In contrast, the pronunciation of IPh characters is a rare sound that corresponds to phonetic radicals. For instance, the pronunciation of “生” and the regular character “牲” are /sheng/, which is the same as most characters that contain “生” (e.g., “笙”, “胜”). However, pronunciation of the IPh character “性” is /xing/, not “/sheng/”. The phonological consistency is 0.39–0.92 for all three CC and is 0-0.28 for the three IPh characters (see Supplementary Table 1 for details). In addition, ISe and IOr characters have the same pronunciation as regular characters (Supplementary Table 1).

Finally, the ISe characters differ from the consistent ones with regards to the transparency of the semantic radical. The CC is high in the transparency of semantic radicals, while the transparency in ISe characters is low. The transparency is defined as the connection between the meaning of the semantic radical, and the meaning of the corresponding character (Shu et al., 2003). That is, semantic radicals of CC can reflect the meaning of corresponding characters. However, the meanings of the ISe characters cannot be speculated from the semantic radicals. For example, the semantic radical “牜” (cattle) is related to the meaning of “牲” (livestock). In contrast, the meaning of ISe character “胜” (victory) is much different from that of the corresponding semantic radical “月” (moon). The same as CC, characters in the IPh and IOr categories are high in the transparency of the semantic radical.

Procedure

The experimental procedure consisted of three oddball blocks and an equal probability block. One of the three categories of inconsistent characters (each category contains three specific characters), in turn, served as the deviant stimuli (dev; probability of occurrence p = 0.25; the number of presentations is divided equally among the three specific characters) across different oddball blocks (each block contains 420 trials), while the consistent characters (containing three specific characters) served as standard stimuli (std; p = 0.75) (Fig. 1D). In the equal probability block (contains 480 trials), the probability of inconsistent (three categories; named equiprobable stimuli) and consistent characters were the same (p = 0.25) (Fig. 1C). Within each block, the trial order was fully randomized, and the order of oddball blocks was also randomized while the equal probability block was implemented at the beginning. For each individual trial, the stimulus was presented for 200 ms, and then a gray image was inserted, lasting for 500–600 ms, at a random time between trials (Fig. 1B). Moreover, the color of the characters may change from white to red at random during some trials (target; p = 0.1; may appear on the standard stimuli of the oddball blocks as well as all stimuli of the equal probability block). The task throughout the experiment was to ignore attributes of the character and to press a button with the right thumb as quickly and accurately as possible when red characters (target stimuli) were presented (similar tasks have been widely used in previous studies examining implicit character processing, e.g., Zhao et al., 2019; Xue et al., 2019). The deviant stimuli in the oddball blocks would not appear twice in a row, and the target only appeared after one standard trial. Participants sat comfortably in an armchair at a distance of 60 cm from the screen, and were given a break for each block that they completed. Using the E-prime software, the images of words were presented within the central visual field (visual angle: horizontally = 2.5°; vertically = 3.8°).

Behavioral analysis

A participant’s response was counted as a hit if the button was pressed for less than 700 ms after the character color changed. Otherwise, the response was counted as a false alarm. Hit and false alarms (FAs) rates during the color change detection task were analyzed in order to evaluate the degree of commitment to unrelated tasks of the participants.

EEG recording and preprocessing

Electroencephalography (EEG) signals were recorded through the use of a 64-channel amplifier (ANT Neuro EEGO, Inc.) that was mounted on an electrode cap according to the international 10–10 system. The online reference electrode during the data collection was CPz. The EEG data was digitized at a sampling rate of 1000 Hz, and impedances were kept below 10 kΩ during the experiment.

Offline preprocessing, artifact removal, and data quality assessment was carried out via the Harvard Automated Processing Pipeline for EEG (HAPPE) in MATLAB (Gabard-Durnam et al., 2018). A spatially distributed subset of channels providing whole-head coverage was processed (excluding the EOG, M1 and M2 channels). HAPPE’s artifact removal steps included bad channel rejection, removal of 50 Hz electrical noise through CleanLine’s multi-taper approach (Mullen, 2012), and participant artifact rejection (e.g., eye blinks, movement) through wavelet-enhanced ICA with automated component rejection via EEGLAB and the Multiple Artifact Rejection Algorithm (Winkler et al., 2015). Post-artifact rejection, any channels removed during the bad channel rejection were repopulated through spherical interpolation to reduce spatial bias in re-referencing. After filtered with a 0.1–40 Hz digital Butterworth bandpass filter with a 12 dB/oct roll-off, the EEG data were then re-referenced to the average reference and mean signal detrended. Epochs were created from − 300 ms pre-stimulus to 700 ms post-stimulus for each trial and baseline corrected using the first 100 ms. Any epochs with retained artifact were rejected using amplitude criteria (± 100 µV), as in prior research (Zhao et al., 2019).

Representational Similarity Analysis

To track the representations of individual characters across time, we used representational similarity analysis (RSA; Kriegeskorte et al., 2008). First, we created neural representational dissimilarity matrices (RDMs) for each time point in the EEG epochs (10 ms resolution), reflecting the pairwise dissimilarity of the characters’ brain representations. Second, we modeled the organization of the neural RDMs using Spearman rank correlation coefficients (Giari et al., 2020; Li et al., 2022), which allowed us to track when representations are explained by the characters’ lexical (frequency) or sub-lexical (containing three dimensions of orthographic, phonological and semantic) statistical information.

Neural RDMs

At each time point from 100 ms before stimulus onset to 600 ms after stimulus onset, we correlated the EEG activity between trial pairs (for the nine different inconsistent characters), separately for the oddball condition (put the data of three oddball sequences together) and equal probability condition. This results in a distance value (1- Pearson correlation) that indicates the dissimilarity between character pairs according to brain activity. By repeating this procedure for each pair of characters we constructed a 9 × 9 neural RDM (Fig. 2A). Individual trials were used as input to the RDM calculation. To calculate the time-point by time-point neural RDMs, the vector for the 61 scalp electrodes was concatenated with those of the five preceding and the five succeeding time points, as implemented in CoSMoMVPA (Oosterhof and Connolly 2016). This resulted in a vector length of 671 features reflecting brain activity spanning 10 ms.

Model RDMs

We designed two series of model RDMs to explore and validate the representation of different statistical information in the EEG data (Fig. 2B). The first series is a number of RDMs constructed based on the current experimental design. These RDMs will be referred to as predictor RDMs in subsequent texts, and these predictor RDMs include orthographic RDM, phonological RDM, semantic RDM and radical-control RDM (based on the phonetic radical category of the material itself). The above predictor RDMs are 9 × 9 binary RDMs, in which 1 corresponded to a comparison between category character (e.g., consistent vs. inconsistent for the orthographic consistency features), and 0 corresponded to a comparison within category stimuli (e.g., consistent vs. consistent). In addition, we also constructed the frequency RDM according to the word (character) frequency (based on the data from the LCSMCS; Sun et al., 1997).

To supplement and validate the results obtained from the predictor RDMs, we constructed another series of RDMs according to the ratings before the formal experiment (from another group of subjects), which resulted in three models of 9 × 9 rating RDMs that corresponded to the orthographic consistency, phonological consistency, and semantic consistency dimensions of our stimuli. Specifically, we calculated the pairwise Euclidean distance between the rating score of each character in each dimension.

Representational similarity analysis

The lower off-diagonal of each matrix was extracted as vectors to calculate the Spearman rank correlations between each model and the EEG data. Since some models were correlated, excluding the other models allowed us to separate the contribution of these models from each other (Giordano et al., 2013; Cichy and Pantazis, 2017). In order to explicitly compare lexical and sub-lexical level statistical information models, lexical (frequency) RDM would be excluded when computing a partial correlation between neural RDM and each sub-lexical (orthographic, phonological and semantic) RDMs, and vice versa. We calculated the partial correlation coefficients at each time point for each subject. These partial correlation coefficients served as an indicator of the time course of different statistical information dimensions in the EEG data.

In addition, in order to detect the difference in the representation strength of different statistical information between the oddball condition and the equal probability condition, we calculated the difference of all partial correlation coefficients of each subject under two conditions (oddball minus equal probability).

Statistical inference

We performed a non-parametric statistical approach for all RSA results which did not depend on assumptions of the data distributions (Nichols and Holmes, 2002). Using the maximum cluster size method, significant temporal clusters were defined as adjacent time points that all exceed a statistical cutoff (cluster-inducing threshold). This cutoff was determined through a sign permutation test according to the distribution of t-values from 10000 permutations of the measured correlation values. The 95th percentile of the t-value distribution was used as the clustering induction threshold of each time point (equivalent to p < 0.05, one-sided). To identify significant clusters, we determined the 95th percentile of maximum cluster sizes across all permutations (equivalent to p < 0.05, one-sided). This approach provided us with significant temporal clusters in which correlation showed significant effects.

Visual mismatch negativity (vMMN) analysis

To verify the existence of prediction error responses, we examined vMMN activities using a data-driven approach. The differential waveforms of characters with different inconsistent categories were obtained by subtracting the ERPs of the corresponding deviant stimuli from the ERPs of the corresponding equiprobable stimuli. This method allows the comparison of ERPs that are evoked by the deviant of the oddball sequence to the ERPs that are evoked by physically identical stimuli from a sequence without any particular frequent (standard) stimulus (Stefanics et al., 2014).

The method of equal probability control was suggested in order to deal with repetition effects due to refractoriness that was assumed to be present in the deviant, minus standard activity that was obtained in classical oddball paradigms (Schröger and Wolff, 1996; Jacobsen and Schröger, 2001). Activity considered as “genuine” vMMN (i.e., vMMN without stimulus-specific refractoriness effects superimposed) emerges when the oddball deviant evokes a larger negativity than the control stimuli (Stefanics et al., 2014). Next, a cluster-based permutation test was utilized to search “genuine” differential activity between the ERPs of deviant and equiprobable stimuli (Maris and Oostenveld, 2007). We conducted this analysis through the use of the Fieldtrip toolbox (Oostenveld et al., 2011) in MATLAB. We developed grand-averages of differential waveforms across two regions of interest (ROI) that correspond to the left (P7, PO7, O1) and right (P8, PO8, O2) posterior occipital-temporal electrodes (the electrodes were selected based on previous studies, e.g., Hu et al., 2020; Kovarski et al., 2021). For each time point (within 0–600 ms) at left or right electrodes, the clusters were formed through two or more neighboring time points whenever the t values (obtained by two-tailed t-test) exceeded the cluster threshold (0.025). The number of permutations was set to 10000, and the corrected significance level was set to 0.05. That is, when the clustering level error probability of a cluster was less than 0.05, then it was considered that there were significant effects in the corresponding period (i.e., effective vMMN activities were identified). We will report the temporal range of the significant negative clusters and their mass for each inconsistent category (the sum of t values in a cluster).

Behavioral results

The mean hit rates and mean false alarm rates of the button presses as well as the mean press latencies of correct responses for each experiment are summarized in Table 1. The high hit rates (99%) and low false alarm rates (< 1%) indicated that the participants were able to accurately focus their attention on the color change detection task.

Table 1

Behavioral results
Block Type	Mean Hit Rates (Range)	Mean False Alarm Rates	Mean Press Latencies (SD)
Equal probability block	99% (96–100%)	< 1%	369.00 ms (30.58 ms)
IOr oddball block	99% (93–100%)	< 1%	393.26 ms (40.07 ms)
IPh oddball block	99% (93–100%)	< 1%	398.22 ms (37.02 ms)
ISe oddball block	99% (93–100%)	< 1%	396.64 ms (34.82 ms)

RSA based on predictor RDMs

According on the current experimental design, we attempt to detect the classification representations of consistent or inconsistent features in neural activities and obtain corresponding RSA results. Prior to the main analysis, we excluded the interference of sub-lexical category features on neural representation. This is due to the fact that we did not detect significant neural representations of specific phonetic radical in any of the conditions (partial correlation coefficient between neural RDM and radical-control RDM). For characters in the oddball condition, the RSA results revealed the time course of the representations of orthographic consistency (see Fig. 3B, significant time points: 106–403 ms and 417–523 ms), phonological consistency (177–470 ms), and semantic consistency (181–378 ms, all cluster-corrected sign permutation test, cluster definition threshold p < 0.05, cluster-corrected significance level p < 0.05). The evidence for representations of frequency (lexical statistical information) was absent in the EEG signals of oddball condition. In addition, only frequency (150–215 ms) and orthographic consistency (159–217 ms) representations can be detected in the neural activities corresponding to the characters in equal probability condition (Fig. 3A).

Further information was provided by statistical analysis of the differences in neural representation between the two conditions. The results show that the orthographic (238–336 ms), phonological (220–300 ms) and semantic (193–308 ms) consistency information in oddball condition can be more strongly predicted by neural patterns than those in equal-probability condition (Fig. 3C). It is worth noting that although significant representations of frequency were detected only in equal-probability condition, the intensity of these representations was not statistically different from that obtained in oddball condition. This means that the representation of frequency in oddball condition could be interpreted by stronger representation of sub-lexical statistical information (by partial correlation calculation).

RSA based on rating RDMs

To examine the generalizability of the results, we performed RSA analysis based on rating RDMs in the same manner. We obtained the representation dynamics of orthographic consistency (161–263 ms and 268–381 ms), phonological consistency (304–394 ms and 485–551 ms), and semantic consistency (164–267 ms and 304–394 ms) in the oddball condition (Fig. 3E). These time regions are basically consistent with the time periods of predictor RDM-based RSA results. We also detected neural representations of frequency in oddball condition from 150 to 238 ms. In the equal probability condition, only frequency is significant (152–216 ms) (Fig. 3D).

The statistical analysis of the representation differences revealed stronger representation of semantic consistency in oddball condition (Fig. 3F). However, although oddball condition also obtained significant representations of orthographic and phonological consistency that were different from equal probability sequence, the differences between these conditions did not reach statistical significance. We believe that this is due to the limited sensitivity of rating RDM detection. The results of frequency on the other hand confirm this view. Because the same frequency RDM was used in the RSA analysis based on predictor RDMs and rating RDMs, the difference between the two is the RDMs controlled in the partial correlation calculation. In the predictor RDM-based analysis, frequency representation is not significant after controlling for orthographic, phonological, and semantic predictor RDM. However, in the rating RDM-based analysis, the representation of frequency was significantly after controlling for orthographic, phonological, and semantic rating RDM. This means that predictor RDMs of various statistical information have stronger detection ability than rating RDM.

vMMNs results

According to a cluster-based permutation test, the vMMN effect was identified for inconsistent orthographic characters across ROIs in the left and right hemisphere, respectively (Fig. 4A, 4B). The cluster-based permutation test revealed that there was a significantly stronger negativity for the differential waveform in left (cluster1: sum [t] = -777.98; p = 0.0007; cluster2: sum [t] = -643.61; p = 0.0013) and right electrodes (cluster1: sum [t] = -776.88; p = 0.0003; cluster2: sum [t] = -673.28; p = 0.0007) for orthographic-related vMMN. Additionally, a phonological-related vMMN effect was also found in both electrodes that corresponded to a left cluster (cluster1: sum [t] = -678.36; p = 0.0007; cluster2: sum [t] = -198.66; p = 0.0225) and a right cluster (cluster1: sum [t] = -539.98; p = 0.0020; cluster2: sum [t] = -238.30; p = 0.0169). Moreover, a vMMN effect was also discovered for inconsistent semantic characters in the left cluster (cluster1: sum [t] = -523.46; p = 0.0033; cluster2: sum [t] = -376.73; p = 0.0062) and a right cluster (cluster1: sum [t] = -494.44; p = 0.0049; cluster2: sum [t] = -361.61; p = 0.0095; cluster3: sum [t] = -281.65; p = 0.0159). The time range for significant clusters for various vMMNs across different ROI are shown in Table 2. To summarize, these vMMN responses are distributed over two consecutive time windows: 150 to 300 milliseconds and 310 to 500 milliseconds. These active time periods of vMMNs are basically consistent with the temporal dynamics of neural representations of sub-lexical statistical information.

Table 2

The time window of existing vMMN responses in different ROIs tested by a nonparametric cluster-based permutation test
vMMN responses	ROIs	Time windows	Cluster mass
inconsistent orthographic	left occipital-temporal	Cluster 1: 313–543 ms Cluster 2: 149–286 ms	-777.98* -643.61*
inconsistent orthographic	right occipital-temporal	Cluster 1: 310–501 ms Cluster 2: 136–284 ms	-776.88* -673.28*
inconsistent phonological	left occipital-temporal	Cluster 1: 150–274 ms Cluster 2: 360–439 ms	-678.36*** -198.66*
inconsistent phonological	right occipital-temporal	Cluster 1: 154–274 ms Cluster 2: 376–474 ms	-539.98** -238.30*
inconsistent semantic	left occipital-temporal	Cluster 1: 159–302 ms Cluster 2: 319–444 ms	-523.46 -376.73
inconsistent semantic	right occipital-temporal	Cluster 1: 155–283 ms Cluster 2: 0–111 ms Cluster 3: 359–471 ms	-494.44 -361.61 -281.65*
p < 0.05, p < 0.01, **p < 0.001

The current work explored two important questions that contribute to our understanding of SL mechanisms: first, whether the statistical properties of real-world print environments as discerned through long-term SL are reflected in the neural processing of words by skilled readers, and second, whether the neural activities in response to these statistical properties are modulated by short-term implicit predictions. Equal probability sequences and oddball sequences were used in this study to detect the neural processing of Chinese characters based on long-term experience and short-term prediction, respectively. We applied RSA to elucidate the neurodynamic pattern of the multidimensional statistical information processing of Chinese characters. First, we found that word (i.e., character) frequency (i.e., lexical level statistics) could be rapidly recovered from EEG response patterns in the equal probability condition. This result answers the first question and confirms the existence of implicit neural representations of statistical information, derived from the long-term textual environment, in word reading. Second, three types of statistical information at the sub-lexical level (i.e., orthographic, phonological, and semantic consistency) could be extracted in the oddball condition. In addition, the representation strength of these sub-lexical statistics was significantly stronger in the oddball than in the equal probability condition. These differences in the strength of neural representations are consistent with the significant vMMN activities we obtained. These results answer the second question, revealing the existence of prediction error signals driven by prediction related to short-term sub-lexical statistical information, and confirming the modulation role of prediction precision in the neural representation of this statistical information.

The present study showed that stable neural representations of word (i.e., character) frequency could be detected in the equal probability condition in which there were no clear short-term predictions, and these representations were active for periods of about 150–220 ms after the character was presented. The frequency information implied by the characters could only be interpreted by the statistical distribution characteristics of the long-term reading environment. This provides direct evidence for the effect of SL on visual word recognition and reveals that the statistical information obtained through long-term SL continues to play a role in current word processing mechanisms. Prior studies have indicated that the word frequency effect (WFE, Monsell, Doyle, & Haggard, 1989) may be part of the indirect evidence for the relationship between SL and reading. Behaviorally, the WFE is a highly replicable and reliable effect that relates to the observation that high-frequency words are processed more efficiently than low-frequency words (e.g., Székely et al., 2003; Snodgrass & Yuditsky, 1996; Jescheniak & Levelt, 1994). In the temporal dynamics of neural processing, most studies have found ERP differences between high-frequency (HF) words and low-frequency (LF) words at around 200 ms after word onset (Proverbio et al., 2008; Vergara-Martínez et al., 2020; Woolnough et al., 2021). The WFE has also been reported in several recent ERP studies focusing on Chinese characters (Wang & Maurer, 2017; Yu et al., 2022). Wang and Maurer’s study found ERP differences at the time interval of 172–253 ms. Another study used an implicit color decision task to report divergence between the ERPs evoked in response to HF characters and the ERPs evoked in response to LF characters over a time period of 210 to 222 ms (Yu et al., 2022). The ERP time window corresponding to the WFE is consistent with our RSA results. By breaking through the limitation of univariate analysis (the traditional form of ERP analysis), we provide more powerful evidence for the implicit and rapid processing of word frequency information. However, almost no significant neural representation activities in response to sub-lexical level statistical information were detected in the equal probability condition. This suggests that skilled adult readers may be sensitive to larger-grain size statistical information (i.e., at the lexical rather than sub-lexical level) in their daily word processing. In fact, this finding is consistent with prior studies of alphabetic languages. Some studies have reported that adults are particularly impacted by body rime rather than grapheme-phoneme level regularities (e.g., Cortese & Simpson, 2000; Jared, 2002). It has been reported that, as humans develop, they become increasingly reliant on O-P regularities at larger grain sizes (i.e., from the sub-lexical to the lexical level) (Treiman & Kessler, 2006).

Neural representations of various sub-lexical level statistics (i.e., orthographic, phonological, and semantic consistency) become detectable when subjects are exposed to visual inputs that contain clear short-term predictions (i.e., in oddball sequences). In addition, the intensity of these neural representations is significantly stronger than those evoked in the equal probability condition. These findings reveal a short-term plasticity mechanism for neural representations of sub-lexical level statistical information. We suggest that this plasticity mechanism, as observed in the present experiment, can be appropriately explained by the predictive coding framework (Friston, 2005; Friston, 2010). In simple terms, predictive coding is an implicit process that creates an internal model of sensory inputs with the aim of minimizing surprise (i.e. a quantitative formulation of prediction error, which is the negative log probability of a sensory event) (Friston & Kiebel, 2009). In order to minimize the cost (free-energy, Friston, 2010) of reducing prediction errors, the system assigns different weights to real-time prediction errors according to the variability of the environment (i.e., prediction precision), which causes the same external input to evoke different levels of response depending on the environment in which it is located. This precision setting mechanism can be conceptually understood as a form of meta-learning: learning what is learnable (Gottlieb, 2012) or estimating the predictability of new contingencies. Returning to the current experiment, a lack of effective prediction (i.e., low precision) was observed within the equal probability sequences. All characters were decoded using the established neural processing patterns, corresponding to neural activities in skilled adult readers that are sensitive to lexical rather than sub-lexical statistical information. However, the situation changed in the oddball sequences, where the presence of unambiguous prediction (i.e., high precision, by repeated exposure to consistent characters) caused any inconsistent information to be evaluated as worthy of learning, thereby activating additional neural activity that was different from the existing processing patterns. This was ultimately reflected in stronger and more explicit neural representations of the various types of sub-lexical statistical information. In fact, a recent study using the oddball paradigm reported similar findings (Dzafic et al., 2020). That study investigated the neural mechanisms that underpin statistical learning and volatility attuning, and showed that, in stable conditions, statistical learning (as behaviorally assessed) was improved compared to the volatile conditions, prediction errors increased, and there was a greater modulation of neuronal gain, forward connections, and backward connections.

We obtained significant “genuine” vMMN responses within the active time window of neural representations of various forms of sub-lexical statistical information. The vMMN responses obtained in the oddball paradigm were interpreted as a neural manifestation of the prediction error signal (Friston, 2005; Baldeweg, 2006; Wacongne et al., 2011; Stefanics et al., 2014; Stefanics et al., 2018). The prediction error in the current experiment could be clearly attributed to the violation of the consistency category. There are three reasons for this attribution. 1) The vMMNs were calculated via subtraction of the ERPs evoked in response to the equiprobable stimuli from the ERPs evoked in response to the deviant stimuli. The equiprobable and the deviant stimuli were comprised of the same Chinese characters, so that the vMMN cannot be described as involving physical, orthographic, phonological, or semantic differences between them. 2) As the probability of presenting the equiprobable and the deviant stimuli were exactly equal (i.e., 1/4), then the vMMN cannot be explained as a difference in refractoriness between the ERPs evoked by them. 3) The vMMN cannot be explained as a violation of a phonological category or a semantic category since there was no reducible phonetic or semantic category within the deviant or equiprobable characters in the oddball sequence. In summary, we believe that our study defines a class of prediction error responses driven by consistency category violation. These categories of consistency at the sub-lexical level were only available from statistical mappings of a long-term reading environment, so these findings further illustrate the dynamic interaction between short-term plasticity driven by prediction error and long-term experience. We believe that these findings advance the understanding of the mechanisms of ISL and provide an interesting perspective on ISL from the predictive coding framework.

Finally, let us consider an additional question about categorization. Categorization is a prerequisite for generating vMMN responses, only when the target characteristics of the deviant and standard stimuli are classified as different types, can the deviants produce violations (i.e., surprises) of the predictions that are established by the standard. Categorization is a basic process that includes visual perception, and this process goes beyond the physical characteristics of the stimuli (Czigler, 2014). For example, there are hundreds of colors that fall under the category “blue” even though they have different combinations and values of hue, saturation, and brightness. In the field of character processing, the vMMN effects of word meaning (Hu et al., 2020) and phonological (Wang et al., 2013) categorization have been investigated. However, the consistency features of concern in the current study are rather special. Phonological consistency, for example, represents how frequently a phonetic radical represents a given sound by calculating the relative proportion of characters with the same pronunciation among those that share the radical (e.g., Lee et al., 2005). This means that phonological consistency is essentially a continuous variable ranging from 0 to 1, and the other two types of consistency are essentially the same. However, the consistency effect reported in previous studies reflects a dichotomy. These studies often anchor certain most common (i.e. high probability) body rime correspondences as consistent words and others as inconsistent words, and found that consistent words are read aloud faster and more accurately than inconsistent words (e.g., Glushko, 1979; Jared, 2002). This consistency effect has also been verified with Chinese characters (Fang et al., 1986; Lee et al., 2006; Lee et al., 2005). With regard to neural mechanisms, previous fMRI studies have reported greater activation in the left inferior frontal gyrus, the left temporoparietal (i.e., inferior parietal gyrus and supramarginal gyrus) region, and the left temporal–occipital junction when naming inconsistent characters compared to consistent ones (Lee et al., 2004). Additionally, several event-related potential (ERP) studies have validated the consistency effects of characters within the time windows of the N170, P200, and N400 components (phonological consistency: Lee et al., 2007; Hsu et al., 2009; Yum et al., 2014; semantic consistency: Hsu et al., 2021). In fact, the current experimental design perpetuates the above idea of exploring the consistency effect. Although we did not focus on specific ERP components, our results provide new evidence for the rationality of this line of research. Furthermore, from the perspective of predictive coding, the vMMN activities evoked by consistency category violation may reveal the true state of the brain's statistical learning product (i.e., neural activities in response to statistical information contained in the reading environment). In other words, the rich distributional statistical information obtained through SL can be aggregated and reclassified into consistent and inconsistent features and reflected in different neural representation patterns.

In summary, our multivariate RSA study demonstrated the contribution of long-term SL to the neural activity related to current word processing. Importantly, we found that the short-term prediction provided by the visual input environment evoked neural representations of sub-lexical statistical information. In addition, we also obtained significant vMMNs, which indexed an implicit processing mechanism that operates within the predictive coding framework. These findings reveal the link between predictive coding and SL, and confirm the short-term plasticity of neural activities corresponding to long-term SL.

Data availability

The datasets generated during and/or analyzed during the current study are available from the corresponding author on reasonable request.

Acknowledgements

The study was funded by National Natural Science Foundation of China (61807023), Humanities and Social Science Fund of Ministry of Education of the People's Republic of China (17XJC190010), Shaanxi Province Natural Science Foundation (2018JQ8015), and Fundamental Research Funds for the Central Universities (CN) (GK201702011) to Jingjing Zhao. Queries of the study should correspond to [email protected].

Author contributions

Jianyi Liu: Conceptualization; methodology; software; validation; formal analysis; investigation; data curation; writing - original draft; visualization; project administration. Tengwen Fan: Investigation; resources. Yan Chen: Investigation; data curation. Jingjing Zhao: Conceptualization; writing - review & editing; supervision; funding acquisition.

Competing interests

All authors declare no financial or non-financial competing interests.

Arciuli, J., & Simpson, I. C. (2012). Statistical learning is related to reading ability in children and adults. Cognitive Science, 36(2), 286–304. https://doi.org/10.1111/j.1551-6709.2011.01200.x.
Arnon, I. (2019). Do current statistical learning tasks capture stable individual differences in children? An investigation of task reliability across modality. Behavior Research Methods. https://doi.org/10.3758/s13428-019-01205-5
Aslin, R. N., Saffran, J. R., & Newport, E. L. (1998). Computation of conditional probability statistics by 8-month-old infants. Psychological Science, 9, 321–324
Baldeweg, T. (2006). Repetition effects to sounds: evidence for predictive coding in the auditory system. Trends in cognitive sciences, 10, 93–94.
Bogaerts, L., Siegelman, N., Christiansen, M. H., & Frost, R. (2022). Is there such a thing as a ‘good statistical learner’? Trends in cognitive sciences, 26(1), 25-37.
Borleffs, E., Maassen, B. A. M., Lyytinen, H., & Zwarts, F. (2017). Measuring orthographic transparency and morphological-syllabic complexity in alphabetic orthographies: A narrative review. Reading and Writing, 30(8), 1617–1638. https://doi.org/10.1007/s11145-017-9741-5.
Christiansen, M. H. (2019). Implicit statistical learning: A tale of two literatures. Topics in cognitive science, 11(3), 468-481.
Christiansen, M. H. (2019). Implicit-statistical learning: A tale of two literatures. Topics in Cognitive Science, 11, 468–481.
Cichy, R. M., Khosla, A., Pantazis, D., & Oliva, A. (2017). Dynamics of scene representations in the human brain revealed by magnetoencephalography and deep neural networks. NeuroImage, 153, 346-358.
Cichy, R. M., Ramirez, F. M., & Pantazis, D. (2015). Can visual information encoded in cortical columns be decoded from magnetoencephalography data in humans? Neuroimage, 121, 193-204.
Conway, C. M., & Christiansen, M. H. (2005). Modality constrained statistical learning of tactile, visual, and auditory sequences. Journal of Experimental Psychology: Learning, Memory & Cognition, 31, 24–39.
Cortese, M. J., & Simpson, G. (2000). Regularity effects in word naming: What are they? Memory & Cognition, 28(8), 1269–1276. https://doi.org/10.3758/bf03211827.
Czigler, I. (2014). Visual mismatch negativity and categorization. Brain Topography, 27(4), 590-598.
Den Ouden, H. E., Kok, P., & De Lange, F. P. (2012). How prediction errors shape perception, attention, and motivation. Frontiers in psychology, 3, 548.
Dzafic, I., Randeniya, R., Harris, C. D., Bammel, M., & Garrido, M. I. (2020). Statistical learning and inference is impaired in the nonclinical continuum of psychosis. Journal of Neuroscience, 40(35), 6759-6769.
Fang, S.-P., Horng, R.-Y., and Tzeng, O. J. L. (1986). “Consistency effects in the Chinese characters and pseudo-character naming tasks,” in Linguistics, Psychology, and the Chinese Language, eds H. S. R. Kao and R. Hoosain (Hong Kong: Centre of Asian Studies, University of Hong Kong), 11–21.
Fiser, J., & Aslin, R. N. (2001). Unsupervised statistical learning of higher-order spatial structures from visual scenes. Psychological Science, 12, 499–504.
Friston, K. (2005). A theory of cortical responses. Philosophical transactions of the Royal Society B: Biological sciences, 360(1456), 815-836.
Friston, K. J. (2010). The free-energy principle: A unified brain theory? Nature Reviews Neuroscience, 11, 127–138. doi:10.1038/nrn2787
Friston, K., & Kiebel, S. (2009). Predictive coding under the free-energy principle. Philosophical transactions of the Royal Society B: Biological sciences, 364(1521), 1211-1221.
Frost, R. (2012). Towards a universal model of reading. Behavioral and Brain Sciences, 35(5), 263–279. https://doi.org/10.1017/S0140525X11001841.
Frost, R., Armstrong, B. C., & Christiansen, M. H. (2019). Statistical learning research: A critical review and possible new directions. Psychological Bulletin, 145(12), 1128.
Frost, R., Siegelman, N., Narkiss, A., & Afek, L. (2013). What predicts successful literacy acquisition in a second language? Psychological Science, 24(7), 1243–1252. https://doi.org/10.1177/0956797612472207.
Gabard-Durnam, L. J., Mendez Leal, A. S., Wilkinson, C. L., & Levin, A. R. (2018). The Harvard Automated Processing Pipeline for Electroencephalography (HAPPE): standardized processing software for developmental and high-artifact data. Frontiers in neuroscience, 12, 97.
Garrido, M. I., Friston, K. J., Kiebel, S. J., Stephan, K. E., Baldeweg, T., and Kilner, J. M. (2008). The functional anatomy of the MMN: A DCM study of the roving paradigm. Neuroimage, 42, 936–944. doi: 10.1016/j.neuroimage.2008.05.018
Garrido, M., Kilner, J. M., Kiebel, S. J., Stephan, K. E., Baldeweg, T., and Friston, K. J. (2009). Repetition suppression and plasticity in the human brain. Neuroimage, 48, 269–279. doi: 10.1016/j.neuroimage.2009.06.034
Giari, G., Leonardelli, E., Tao, Y., Machado, M., & Fairhall, S. L. (2020). Spatiotemporal properties of the neural representation of conceptual content for words and pictures–an MEG study. Neuroimage, 219, 116913.
Giordano, B. L., McAdams, S., Zatorre, R. J., Kriegeskorte, N., & Belin, P. (2013). Abstract encoding of auditory objects in cortical activity patterns. Cerebral cortex, 23(9), 2025-2037.
Glushko, R. J. (1979). The organization and activation of orthographic knowledge in reading aloud. Journal of Experimental Psychology: Human Perception and Performance, 5(4), 674–691. doi:10.1037/0096-1523.5.4.674.
Gottlieb, J. (2012). Attention, learning, and the value of information. Neuron, 76, 281–295. doi:10.1016/j.neuron.2012.09.034
Hauptman, M., Blanco-Elorrieta, E., & Pylkkänen, L. (2022). Inflection across Categories: Tracking Abstract Morphological Processing in Language Production with MEG. Cerebral Cortex, 32(8), 1721-1736.
He, X., & Tong, X. (2017). Statistical learning as a key to cracking Chinese orthographic codes. Scientific Studies of Reading, 21(1), 60-75.
Hsu, C. H., Tsai, J. L., Lee, C. Y., & Tzeng, O. J. L. (2009). Orthographic combinability and phonological consistency effects in reading Chinese phonograms: an event-related potential study. Brain and Language, 108(1), 56-66.
Hsu, C. H., Wu, Y. N., & Lee, C. Y. (2021). Effects of Phonological Consistency and Semantic Radical Combinability on N170 and P200 in the Reading of Chinese Phonograms. Frontiers in psychology, 12.
Hu, A., Gu, F., Wong, L. L., Tong, X., & Zhang, X. (2020). Visual mismatch negativity elicited by semantic violations in visual words. Brain Research, 1746, 147010.
Hubbard, R. J., & Federmeier, K. D. (2021). Representational pattern similarity of electrical brain activity reveals rapid and specific prediction during language comprehension. Cerebral Cortex, 31(9), 4300-4313.
Jacobsen, T., and Schröger, E. (2001). Is there pre-attentive memory-based comparison of pitch? Psychophysiology, 38, 723–727. doi: 10.1111/1469-8986.3840723
Jared, D. (2002). Spelling-sound consistency and regularity effects in word naming. Journal of Memory and Language, 46(4), 723-750.
Jescheniak, J. D., & Levelt, W. J. M. (1994). Word frequency effects in speech production: Retrieval of syntactic information and of phonological form. Journal of Experimental Psychology: Learning, Memory, and Cognition, 20, 824–843.
Koelsch, S., Busch, T., Jentschke, S., & Rohrmeier, M. (2016). Under the hood of statistical learning: A statistical MMN reflects the magnitude of transitional probabilities in auditory sequences. Scientific reports, 6(1), 1-11.
Kovarski, K., Charpentier, J., Roux, S., Batty, M., Houy-Durand, E., & Gomot, M. (2021). Emotional visual mismatch negativity: a joint investigation of social and non-social dimensions in adults with autism. Translational Psychiatry, 11(1), 1-12.
Kriegeskorte, N., Mur, M., & Bandettini, P. A. (2008). Representational similarity analysis-connecting the branches of systems neuroscience. Frontiers in systems neuroscience, 4.
Lecaignard, F., Bertrand, O., Gimenez, G., Mattout, J., & Caclin, A. (2015). Implicit learning of predictable sound sequences modulates human brain responses at different levels of the auditory hierarchy. Frontiers in Human Neuroscience, 9, 505.
Lee, C. Y., Tsai, J. L., Chan, W. H., Hsu, C. H., Hung, D. L., & Tzeng, O. J. (2007). Temporal dynamics of the consistency effect in reading Chinese: an event-related potentials study. Neuroreport, 18(2), 147-151.
Lee, C. Y., Tsai, J. L., Huang, H. W., Hung, D. L., & Tzeng, O. J. L. (2006). The temporal signatures of semantic and phonological activations for Chinese sublexical processing: An event-related potential study. Brain Research, 1121, 150–159.
Lee, C. Y., Tsai, J. L., Kuo, W. J., Yeh, T. C., Wu, Y. T., Ho, L. T., ... & Hsieh, J. C. (2004). Neuronal correlates of consistency and frequency effects on Chinese character naming: an event-related fMRI study. Neuroimage, 23(4), 1235-1245.
Lee, C. Y., Tsai, J. L., Su, E. C. I., Tzeng, O. J. L., & Hung, D. L. (2005). Consistency, regularity, and frequency effects in naming Chinese characters. Language and Linguistics, 6, 75–107.
Li, Y., Zhang, M., Liu, S., & Luo, W. (2022). EEG decoding of multidimensional information from emotional faces. NeuroImage, 258, 119374.
Maris, E., & Oostenveld, R. (2007). Nonparametric statistical testing of EEG-and MEG-data. Journal of neuroscience methods, 164(1), 177-190.
Monsell, S., Doyle, M. C., & Haggard, P. N. (1989). Effects of frequency on visual word recognition tasks: Where are they? Journal of Experimental Psychology: General, 118(1), 43.
Mullen, T. NITRC: CleanLine: Tool/Resource Info. (2012).
Myers, J. (2019). The Grammar of Chinese Characters: Productive Knowledge of Formal Patterns in an Orthographic System. London: Routledge. doi: 10.4324/9781315265971.
Nichols, T. E., & Holmes, A. P. (2002). Nonparametric permutation tests for functional neuroimaging: a primer with examples. Human brain mapping, 15(1), 1-25.
Oostenveld, R., Fries, P., Maris, E., & Schoffelen, J. M. (2011). FieldTrip:open source software for advanced analysis of MEG, EEG, and invasive electrophysiological data. Computational intelligence and neuroscience, 2011.
Oosterhof, N. N., Connolly, A. C., & Haxby, J. V. (2016). CoSMoMVPA: multi-modal multivariate pattern analysis of neuroimaging data in Matlab/GNU Octave. Frontiers in neuroinformatics, 10, 27.
Paraskevopoulos, E., Kuchenbuch, A., Herholz, S. C., & Pantev, C. (2012). Statistical learning effects in musicians and non-musicians: An MEG study. Neuropsychologia, 50(2), 341-349.
Perruchet, P., & Pacton, S. (2006). Implicit learning and statistical learning: One phenomenon, two approaches. Trends in Cognitive Sciences, 10, 233–238.
Proverbio, A.M., Zani, A., Adorni, R., 2008. The left fusiform area is affected by written frequency of words. Neuropsychologia, 6, 2292–2299. https://doi.org/10.1016/j.neuropsychologia.2008.03.024.
Reber, A. S. (1967). Implicit learning of artificial grammars. Journal of verbal learning and verbal behavior, 6(6), 855-863.
Saffran, J. R., Aslin, R. N., & Newport, E. L. (1996). Statistical learning by 8-month-old infants. Science, 274, 1926–1928. http://dx.doi.org/10.1126/ science.274.5294.1926
Saffran, J. R., Johnson, E. K., Aslin, R. N., & Newport, E. L. (1999). Statistical learning of tone sequences by human infants and adults. Cognition, 70, 27–52. http://dx.doi.org/10.1016/S0010-0277(98)00075-4
Saffran, J. R., Newport, E. L., Aslin, R. N., Tunick, R. A., & Barrueco, S. (1997). Incidental language learning: Listening (and learning) out of the corner of your ear. Psychological Science, 8, 101–105.
Sawi, O. M., & Rueckl, J. (2019). Reading and the neurocognitive bases of statistical learning. Scientific Studies of Reading, 23(1), 8–23. https://doi.org/10.1080/10888438.2018.1457681.
Schröger, E., & Wolff, C. (1996). Mismatch response of the human brain to changes in sound location. Neuroreport, 7(18), 3005-3008.
Seidenberg, M. S. (2011). Reading in different writing systems: One architecture, multiple solutions. In P. McCardle, B. Miller, J. R. Lee, & O. J. L. Tzeng (Eds.). yslexia across languages: Orthography and the brain–gene–behavior link (pp. 146–168). Baltimore, MD: Paul H Brookes Publishing.
Shtyrov, Y., Goryainova, G., Tugin, S., Ossadtchi, A., & Shestakova, A. (2013). Automatic processing of unattended lexical information in visual oddball presentation: neurophysiological evidence. Frontiers in human neuroscience, 7, 421.
Shu, H., Chen, X., Anderson, R. C., Wu, N., & Xuan, Y. (2003). Properties of school Chinese: implications for learning to read. Child Development, 74(1), 27e47.
Siegelman, N., Rueckl, J. G., Steacy, L. M., Frost, S. J., van den Bunt, M., Zevin, J. D., ... & Morris, R. D. (2020). Individual differences in learning the regularities between orthography, phonology and semantics predict early reading skills. Journal of memory and language, 114, 104145.
Snodgrass, J. G., & Yuditsky, T. (1996). Naming times for the Snodgrass and Vanderwart pictures. Behavior Research Methods, Instruments, & Computers, 28, 516–536.
Southwell, R., & Chait, M. (2018). Enhanced deviant responses in patterned relative to random sound sequences. Cortex, 109, 92-103.
Spencer, M., Kaschak, M. P., Jones, J. L., & Lonigan, C. J. (2014). Statistical learning is related to early literacy-related skills. Reading and Writing, 28(4), 467–490. https://doi.org/10.1007/s11145-014-9533-0.
Stefanics, G., and Czigler, I. (2012). Automatic prediction error response to hands with unexpected laterality: an electrophysiological study. Neuroimage, 63, 253–261. doi: 10.1016/j.neuroimage.2012.06.068
Stefanics, G., Heinzle, J., Horváth, A. A., & Stephan, K. E. (2018). Visual mismatch and predictive coding: A computational single-trial ERP study. Journal of Neuroscience, 38(16), 4020-4030.
Stefanics, G., Kremláček, J., & Czigler, I. (2014). Visual mismatch negativity: a predictive coding view. Frontiers in human neuroscience, 8, 666.
Sun, H. L., Huang, J. P., Sun, D. J., Li, D. J., & Xing, H. B. (1997). Introduction to language corpus system of modern Chinese study. In Paper collection for the fifth world Chinese teaching symposium (pp. 459-466). Beijing: Peking University Press.
Székely, A., D’amico, S., Devescovi, A., Federmeier, K., Herron, D., Iyer, G., et al. (2003). Timed picture naming: Extended norms and validation against previous studies. Behavior Research Methods, Instruments, & Computers, 35, 621–633.
Tong, S., Zhang, P., & He, X. (2020a). Statistical learning of orthographic regularities in Chinese children with and without dyslexia. Child development, 91(6), 1953-1969.
Tong, X., Wang, Y., & Tong, S. X. (2020b). The neural signature of statistical learning of orthography. Frontiers in Human Neuroscience, 26.
Tong, X., Wang, Y., & Tong, S. X. (2020c). Neurocognitive Correlates of Statistical Learning of Orthographic–Semantic Connections in Chinese Adult Learners. Neuroscience Bulletin, 36(8), 895-906.
Torkildsen, J.v. K., Arciuli, J., & Wie, O. B. (2019). Individual differences in statistical learning predict children’s reading ability in a semi-transparent orthography. Learning and Individual Differences, 69, 60–68. https://doi.org/10.1016/J.LINDIF.2018.11.003.
Toro, J. M., Sinnett, S., & Soto-Faraco, S. (2005). Speech segmentation by statistical learning depends on attention. Cognition, 97, B25–B34.
Treiman, R., & Kessler, B. (2006). Spelling as statistical learning: Using consonantal context to spell vowels. Journal of Educational Psychology, 98(3), 642–652. https://doi.org/10.1037/0022-0663.98.3.642.
Turk-Browne, N. B., Jungé, J., & Scholl, B. J. (2005). The automaticity of visual statistical learning. Journal of Experimental Psychology: General, 134, 552–564. http://dx.doi.org/10.1037/0096-3445.134.4.552
Turk-Browne, N. B., Scholl, B. J., Chun, M. M., & Johnson, M. K. (2009). Neural evidence of statistical learning: Efficient detection of visual regularities without awareness. Journal of Cognitive Neuroscience, 21, 1934–1945.
Vergara-Martínez, M., Gomez, P., & Perea, M. (2020). Should I stay or should I go? An ERP analysis of two-choice versus go/no-go response procedures in lexical decision. Journal of Experimental Psychology: Learning, Memory, and Cognition, 46(11), 2034.
Wacongne, C., Labyt, E., van Wassenhove, V., Bekinschtein, T., Naccache, L., & Dehaene, S. (2011). Evidence for a hierarchy of predictions and prediction errors in human cortex. Proceedings of the National Academy of Sciences, 108(51), 20754-20759.
Wang, F., & Maurer, U. (2017). Top-down modulation of early print-tuned neural activity in reading. Neuropsychologia, 102, 29-38.
Wang, X. D., Liu, A. P., Wu, Y. Y., & Wang, P. (2013). Rapid extraction of lexical tone phonology in Chinese characters: a visual mismatch negativity study. PLoS One, 8(2), e56778.
Wei, D., & Gillon-Dowens, M. (2018). Written-word concreteness effects in non-attend conditions: Evidence from mismatch responses and cortical oscillations. Frontiers in Psychology, 2455.
Winkler, I., Debener, S., Muller, K.-R. & Tangermann, M. On the influence of high-pass filtering on ICA-based artifact reduction in EEG-ERP. in 2015 37^th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC) 2015, 4101–4105 (IEEE, 2015).
Woolnough, O., Donos, C., Rollo, P. S., Forseth, K. J., Lakretz, Y., Crone, N. E., ... & Tandon, N. (2021). Spatiotemporal dynamics of orthographic and lexical processing in the ventral visual pathway. Nature Human Behaviour, 5(3), 389-398.
Xue, L., Maurer, U., Weng, X., & Zhao, J. (2019). Familiarity with visual forms contributes to a left-lateralized and increased N170 response for Chinese characters. Neuropsychologia, 134, 107194.
Yu, R., Chen, J., Peng, Y., & Gu, F. (2022). Visual event-related potentials reveal the early lexical processing of Chinese characters. Neuropsychologia, 165, 108132.
Yum, Y. N., Law, S. P., Su, I. F., Lau, K. Y. D., & Mo, K. N. (2014). An ERP study of effects of regularity and consistency in delayed naming and lexicality judgment in a logographic writing system. Frontiers in psychology, 5, 315.
Zhao, J., Li, T., Elliott, M. A., & Rueckl, J. G. (2018). Statistical and cooperative learning in reading: An artificial orthography learning study. Scientific Studies of Reading, 22(3), 191-208.
Zhao, J., Maurer, U., He, S., & Weng, X. (2019). Development of neural specialization for print: Evidence for predictive coding in visual word recognition. PLoS biology, 17(10), e3000474.
Zhao, J., Wang, X., Frost, S. J., Sun, W., Fang, S. Y., Mencl, W. E., ... & Rueckl, J. G. (2014). Neural division of labor in reading is constrained by culture: A training study of reading Chinese characters. Cortex, 53, 90-106.

(Not answered)

SupplementaryTable1.docx

Download PDF

Editorial decision: revise
04 Jul, 2023
Review #3 received at journal
13 Jun, 2023
Reviewer #3 agreed at journal
22 May, 2023
Review #2 received at journal
28 Feb, 2023
Reviewer #2 agreed at journal
07 Feb, 2023
Reviewer #1 agreed at journal
15 Jan, 2023
Reviewers invited by journal
11 Jan, 2023
Submission checks completed at journal
28 Dec, 2022
First submitted to journal
27 Dec, 2022
Unknown event
27 Dec, 2022
Editor assigned by journal
27 Dec, 2022

You are reading this latest preprint version

Seeking the neural traces of statistical learning during implicit processing of visual words

Status:

Version 1

Abstract

Figures

Introduction

Materials And Methods

Results

RSA based on predictor RDMs

Discussion

Conclusion

Declarations

References

Additional Declarations

Supplementary Files

Status:

Version 1