Abstract
Impairment in predictive processes gained a lot of attention in recent years as an explanation for autistic symptoms. However, empirical evidence does not always underpin this framework. Thus, it is unclear what aspects of predictive processing are affected in autism spectrum disorder. In this study, we tested autistic adults on a task in which participants acquire probability-based regularities (that is, a statistical learning task). Twenty neurotypical and 22 autistic adults learned a probabilistic, temporally distributed regularity for about 40 min. Using frequentist and Bayesian methods, we found that autistic adults performed comparably to neurotypical adults, and the dynamics of learning did not differ between groups either. Thus, our study provides evidence for intact statistical learning in autistic adults. Furthermore, we discuss potential ways this result can extend the scope of the predictive processing framework, noting that atypical processing might not always mean a deficit in performance.
Similar content being viewed by others
Introduction
In the past years, several frameworks emerged to explain the neurocognitive mechanisms behind autism spectrum disorder (ASD). A line of research suggests that autistic behavior might emerge due to an atypical ability to predict future events based on experience and current sensory input—that is, predictive processing. The predictive processing framework originated from perception research: according to it, the brain generates hypotheses about the environment during perception based on previous experiences (priors) and updates the hypotheses using the prediction errors, that is, the differences between the predictions and the actual sensory inputs1. This framework has since been extended to a general framework for understanding brain functioning, including learning and memory2,3. It might benefit the understanding of ASD, and thus, help develop better supporting systems and interventions.
Various approaches to the predictive processing framework offer explanations for autistic traits by highlighting atypicalities in different components of the process. One of them assumes that autistic individuals tend to attribute a high and inflexible precision to prediction errors3. According to this view, autistic people would systematically adjust their internal representation of the world after each (minor) prediction error, instead of considering that some of these errors might simply signal unavoidable noise. Importantly, such errors indicate to the learner that the regularity is not fully learned yet3,4. Another viewpoint proposes that ASD individuals rely more on incoming sensory data (i.e., bottom-up information) compared to their prior experiences (i.e., top-down processes), which may result in less adaptive behaviour4,5,6,7,8. Lastly, in ASD, atypical predictive processing may arise from an inaccurate estimation of the extent to which environmental regularities change (as opposed to the estimation of the noise in the regularity itself, as mentioned above, see9 for different types of uncertainties), that is, the estimation of volatility10. Autistic people tend to overestimate volatility, even at the expense of learning environmental probabilities8. Altered or impaired predictive processes could explain sensory hypersensitivity3,11, deficits in sociocognitive skills12, and rigid habit-like behaviour in ASD (e.g.3,4,10). Despite its potential as a comprehensive framework for ASD, mixed empirical results on predictive processing suggest a more complex picture (for reviews, see6,13).
Predictive processing plays a crucial role in various functions, such as perception, different mechanisms of memory, and even habituation2,14,15,16. It can occur with17 or without18,19 reward. The former may be impaired or intact in ASD (e.g.20), depending on many factors, such as the reliability of the regularity to predict, whether the cues are social or nonsocial21, or the strength of the association22. However, the presence of reward or feedback can affect these results23, as reward sensitivity might be altered in ASD24. Thus, it is important to address examples of predictive processing that do not involve reward or trial-by-trial feedback.
Such is statistical learning: a form of predictive processing that entails learning probability-based regularities of the environment25,26,27. Despite its relevance, even the most comprehensive reviews often overlook or neglect studies about statistical learning in ASD6,13, although it contributes to language acquisition28, social skills29, and habit learning30—behaviors that are often altered in ASD31. Most statistical learning studies on ASD have used tasks where the regularity is predictable with a probability of one (that is, deterministic tasks). The results of these are mixed; some of them have found impaired32,33,34, and others have reported intact statistical learning in ASD35,36,37,38,39. Importantly, however, when regularities can be predicted with a probability less than one (often referred to as probabilistic regularities), no studies, to our knowledge, have found impaired statistical learning in ASD. Indeed, on probabilistic tasks, autistic individuals have similar40,41, or potentially even superior42,43 statistical learning performance compared to neurotypical peers. Thus, it appears that on probabilistic tasks, under certain circumstances, ASD participants perform similarly to (or even better than) neurotypical ones. What these circumstances are, however, is not fully understood.
Roser et al.43 used the differences in local and global processing to explain their results of superior statistical learning in autistic adults (compared to neurotypical adults). In their visuospatial task, Roser and colleagues presented participants with consecutive 3 × 3 grids containing abstract shapes. Unbeknownst to the participants, certain shapes consistently appeared in specific spatial relationships (e.g., two specific shapes always positioned diagonally to each other). The participants' (implicit) ability to differentiate these "base pairs" from other pairs was later assessed as a measure of their visual statistical learning. In this task, participants might benefit from local-level processing, which, importantly, would function superiorly in ASD (44,45; but see46 for contradicting evidence). Consequently, it is not clear whether superior performance in ASD measured by Roser et al.43 derived from better performance in statistical learning or just reflected differences in processing style. Thus, in our study, we aimed to test autistic individuals on a task where the performance assumably benefits less from local processing strategies, since regularities are temporally, and not spatially distributed18.
Note, however, that Roser et al.43 found superior learning only in autistic adults, but not in children (although they did not compare age groups directly). This is of importance, since statistical learning might change during the lifespan19,47,48. Although no study to date has compared the performance of autistic adults and children directly in a statistical learning task, the results of Roser et al.43 suggest that a superior statistical learning performance may only be present in autistic adults, but not in children, compared to their neurotypical age groups. Consequently, in our study, we aimed to compare the statistical learning performance of ASD versus neurotypical adults. We tested statistical learning using a probabilistic, temporally distributed task, where the pattern items do not follow each other directly but in a non-adjacent manner (the Alternating Serial Reaction Time (ASRT) task by Howard and Howard18). Based on Roser et al.43, we expected a superior performance of autistic compared to neurotypical adults.
Methods
Participants
In total, 45 participants were recruited for the study. Three neurotypical participants were excluded from the analysis due to errors in the data collection. Thus, the data of 42 participants were entered into the analyses, 20 of them were neurotypicals, and 22 of them had a diagnosis of ASD. Neurotypical participants were screened for diagnoses of any psychiatric or neurological disorders, and none of them scored higher on the autism spectrum quotient (AQ) questionnaire than 27, which means that they do not tend to show autistic behavioral patterns49. ASD diagnoses were provided by trained clinicians; both childhood scores of autism diagnostic interview-revised (ADI-R) and autism diagnostic observation schedule, IV-module (ADOS-IV)50,51 confirmed the diagnosis. We screened ASD participants for comorbid disorders: 12 of them had at least one of the following: attention deficit hyperactivity disorder (5), obsessive–compulsive disorder (3), generalized anxiety disorder (2), bipolar disorder (1), depression (1), and schizophrenia (1). Having an intellectual disability, language impairment, or active psychosis were exclusion criteria. Neurotypical participants were recruited by advertisement, while participants with ASD were recruited from the outpatient unit of the Department of Psychiatry and Psychotherapy, Semmelweis University. No participant received financial compensation for their participation.
The two groups did not differ in age, gender distribution, and years of education, see Table 1. All participants provided written informed consent. The study was conducted in accordance with the Declaration of Helsinki of 1975, as revised in 2008 and it was approved by the Regional and Institutional Committee of Science and Research Ethics, Semmelweis University, Budapest, Hungary (SERKEB No.: 145/2019). The experiment took place at the Laboratory of Brain, Memory and Language Lab, Eötvös Loránd University, Budapest.
Task and procedure
To measure statistical learning, we applied the ASRT task18, a commonly used and highly reliable task (e.g.52). In this task, participants saw four empty circles on a white background, horizontally arranged on the screen. A target stimulus (a dog’s head) appeared in one of the four locations. Participants were asked to press the button corresponding to the location of the appearing stimuli (Y, C, B, and M keys of a QWERTZ keyboard corresponded to the first, second, third, and fourth circle, from left to right respectively), using their right and left index and middle fingers. Participants were told that the goal of the task is to be as fast and as accurate as possible. Unknown to them, however, the serial order of the stimulus locations followed a specific structure: every second stimulus appeared randomly in one of the four possible locations, but every first element appeared systematically in the same order. Thus, these alternating elements formed an eight-element probabilistic sequence (e.g., 1r2r4r3r, where the numbers indicate the location of the elements belonging to the pattern, and r indicates a random position out of the four, see Fig. 1). Due to this structure, some combinations of three consecutive trials (triplets) were more likely to be formed. In the above example, 1 x 2, 2 x 4, 4 x 3, and 3 x 1 are high-probability triplets (where “x” indicates the middle element of the triplet, regardless of whether it is random or belongs to the pattern)—they can be both formed by two pattern and one random elements (PrP), or two random elements enclosing a pattern one (rPr). Out of the total of 64 possible triplets, 16 were high-probability triplets. Any other triplet (such as 1 x 3 or 2 x 1) cannot be formed by two pattern, and one random elements—thus, they occurred with low probability. Importantly, if participants perform with decreased RT and higher accuracy on the last element of a high-probability triplet (e.g., 2 in the above-mentioned 1 x 2 triplet) compared to the last element of a low-probability triplet (e.g., 3 in the above-mentioned 1 x 3 triplet), it means that the participant learned to predict the former one based on the preceding two elements, thus, acquired the underlying probability structure of the task. There were 48 low-probability triplets in this task. This task structure resulted in the following statistical structure: 50% of the trials were the last trial of a high-probability triplet formed by two pattern elements and one random (pattern-random-pattern), 12.5% of all trials were the last elements of a random-ending high-probability triplet (random-pattern-random). Therefore, high-probability triplets occurred with 62.5%, while low-probability triplets occurred with 37.5% overall probability. On the unique triplet level, high-probability triplets occurred with a 4% probability (62.5%/16), while low-probability ones occurred with a 0.8% probability (37.5%/48). As the last element of a high-probability triplet was more predictable than a low-probability triplet, we defined statistical learning as the difference in reaction times (RT) and accuracy performance between these triplet types. For further details of the ASRT task structure, see Fig. 1.
The task was divided into 40 blocks in total. Each block contained 85 trials: five random elements at the beginning (these were excluded from the analysis later), and an eight-elements alternating sequence ten times, as described above. The task was self-paced: the target stimulus remained on the screen until the first correct response, and the response-stimulus interval (RSI) was 120 ms, during which participants saw the four empty circles. Between blocks, participants received feedback on their RT and accuracy and could rest awhile. To reduce noise due to intra-individual variability in the analysis, we merged five blocks into one unit of analysis called an epoch.
To familiarize the participants with the ASRT task and to make sure they understood the instructions, participants first performed two blocks without the pattern (that is, all trials were random). After that, participants were asked to perform 8 epochs, with a ~ 15-min-long break after the 4th epoch. Despite the ASRT task being shown to be truly implicit (that is, no conscious knowledge is formed regarding the regularities hidden in the task, see53), once the ASRT was over, we administered a short questionnaire to make sure that none of the participants gained explicit knowledge of the structure of the task. It consisted of two questions increasingly specific to the nature of the structure: “Have you noticed anything special regarding the task?”, and “Have you noticed some regularity in the sequence of stimuli?”. According to this questionnaire, none of our participants gained conscious knowledge of the regularity.
Statistical analysis
Statistical analyses were carried out using JASP 0.16.1.054, and data preparation and visualization were conducted using Python 3.8, using pandas, NumPy, os, matplotlib, and seaborn packages 55,56,57. First, we determined about each trial in a sliding window manner whether, based on the two elements preceding it, they were the last element of a high- or a low-probability triplet (for the sake of simplicity, henceforth referred to as high-probability and low-probability triplets). That is, considering the example in Fig. 1, if the stimuli followed the “13214232” order, first, trial “2” was categorized as a high-probability triplet (1 3 2) element. Then, trial “1” was categorized as a high-probability triplet (3–2–1) element again, and so on. After this categorization, we excluded the last elements of trill (e.g., 2 1 2), and repetition (e.g., 2 2 2) triplets since participants show a pre-existing tendency to react faster to these elements, thus, they can bias the RTs58. We also screened for outlier trials using a boxplot, meaning that we excluded all trials where the RT fell outside the range of 1.5 inter-quartile distance (IQD) from the first quartile and 1.5 IQD from the third quartile. With this method, we excluded 5.83% of all trials in the entire sample (5.46% in the neurotypical, and 6.17% in the ASD group). Using the remaining data, we calculated the mean accuracy and median RT in each epoch, separately for high- and low-probability triplets. On these data, we performed a mixed-design ANOVA described in the Results section. When applicable, pairwise comparisons were performed using Holm correction.
Additionally to the frequentist statistics, we performed Bayesian analyses using default JASP priors, to be able to detect null results. Based on the BF01 values (which indicate the ratio of the likelihood of the null hypothesis to the likelihood to the alternative hypothesis), we calculated Bayes Factorexclusion (BFexcl) values. We compared the models to the null model (which included the subject variable and random slopes) in each case, and we calculated BFexcl values across matched models. BFexcl values indicate the likeliness of a model that does not include the given effect as opposed to the one that does. The BFexcl values above one rather support the exclusion of the given factor from the model, while values below one support the inclusion59. Values close to one mean that there is not enough evidence to support either inclusion or exclusion. We suggest a similar interpretation of these values as that of BF01 scores: a score above three means substantial evidence in favour of the null hypothesis, while a score between 0.33 and 1 indicates anecdotal evidence, while a score below 0.33 substantial evidence in favour of the alternative hypothesis60,61. For the sake of transparency, however, we reported BF01 values and errors (%) in Supplementary Materials S1 Table.
The data are available at https://osf.io/mebcx/.
Significance statement
According to the predictive processing framework, autistic symptoms are the result of the weak ability to predict future events based on prior knowledge and sensory input. Despite its popularity, the validity of this framework and its limitations are still unclear. Here, we aim to test the predictive processing framework in autism by using a temporal statistical learning task. We found intact predictive processing in autism—neither the amount of learning nor the dynamics of it were altered. Our result challenges the predictive processing framework of autism. However, we suggest an update of the framework to better explain existing data and deepen our understanding of autism.
Results
To test whether statistical learning differs between ASD and neurotypical groups, we conducted two mixed-design analyses of variances (ANOVAs), separately for accuracy and RT as dependent variables. In each, epoch (1–8) and triplet type (high/low-probability) served as within-subject factors and group (ASD/neurotypical) as a between-subject factor.
Is statistical learning different between ASD and neurotypical adults? RT
We found a significant Triplet main effect in the ANOVA on RT [F(1,40) = 116.287, p < 0.001, η2p = 0.744, BFexcl < 0.001]: participants were faster on the high- compared to the low-probability triplets, indicating that statistical learning was present throughout the task. According to the significant Epoch × Triplet interaction [F(7,280) = 7.162, p < 0.001, η2p = 0.152, BFexcl < 0.001], this difference showed a gradual progress; reaching a significant level in the second epoch and remaining significant in every later epoch (pHolm ≤ 0.021). Importantly, however, based on nonsignificant Triplet × Group and Epoch × Triplet × Group interactions, the groups differed neither in the overall amount of learning [F(1,40) = 1.603, p = 0.213, η2p = 0.039, BFexcl = 2.828] nor in the dynamics of learning [F(7,280) = 0.720, p = 0.655, η2p = 0.018, BFexcl = 25.586], respectively.
Is statistical learning different between ASD and neurotypical adults? Accuracy
The ANOVA on accuracy showed a similar pattern: the significant Triplet main effect indicated that statistical learning happened [F(1,40) = 33.805, p < 0.001, η2p = 0.458, BFexcl < 0.01], i.e., participants were more accurate on the high- compared to the low-probability triplets. Moreover, there was a significant Epoch x Triplet interaction, showing a difference between epochs in the amount of learning, which, on the other hand, was not supported by the Bayesian statistics [F(7,280) = 2.443, p = 0.019, η2p = 0.058, BFexcl = 1.333]—the difference between high- and low-probability triplets reached and maintained a significant level from the 4th epoch on (from that epoch on, pHolm ≤ 0.003). Yet, both the overall learning [indicated by the Triplet × Group interaction: F(1,40) = 0.130, p = 0.721, η2p = 0.003, BFexcl = 3.606] and the dynamics of learning [indicated by the Epoch × Triplet × Group interaction: F(7,280) = 0.898, p = 0.508, η2p = 0.022, BFexcl = 15.263] were similar in ASD and neurotypical groups, see Fig. 2. Accuracy results are shown in Supplementary Materials (SM) Results Fig. S2. Results about the general, statistical learning-independent accuracy and RT are shown on Supplementary Fig. S3.
Discussion
In this study, we aimed to test the statistical learning of autistic adults in light of the predictive processing framework. Besides the overall statistical learning, we also tested the dynamics of the learning process—which, to our best knowledge, has not been addressed in autistic adults before. We also performed exploratory analyses to find individual differences regarding the autistic symptom severity, which are reported in the SM (see Supplementary Information 1 and Supplementary Figure S3). Our findings provide frequentist and some Bayesian evidence of intact learning performance and similar learning curves in ASD and neurotypical participants.
These results seemingly contradict both the predictive processing framework of ASD that suggests impaired statistical learning in ASD2,13 and empirical findings by Roser et al.43, who found superior statistical learning in ASD. On the other hand, they are in line with previous literature that found no impairment in probabilistic statistical learning tasks in autistic children36,40,41,42. These contradictions highlight the possibility that predictive processing in autism might depend on the task used and that some aspects of it may be intact in ASD, which has both theoretical and clinical importance. In the following paragraphs, we will discuss possible explanations for these inconsistencies. First, the general information processing style the task requires might play a role. Second, atypicalities in different components of predictive processing could provide an explanation. As mentioned in the Introduction, the predictive processing framework of ASD is not a monolithic concept but rather an umbrella term that includes different mechanisms that could explain autistic traits/symptoms—these mechanisms are not necessarily mutually exclusive, yet apply a different angle to interpret the results. We did not directly access these mechanisms in our study, moreover, all these approaches face challenges by contradicting empirical results20,62,63,64. Thus, future studies are warranted on them, yet they may still help us understand our results in the context of the predictive processing framework and provide future directions. Lastly, we will discuss the potential role of age in statistical learning.
Based on the work of Roser et al.43, we even expected a superior statistical learning performance in ASD, as compared to neurotypical adults but could not replicate their results. An important difference between their task and ours was that their visual statistical learning task presented the learnable regularities on the same slide (that is, it was spatially distributed), whereas in our ASRT task, the learnable regularities were distributed in time (that is, temporally distributed). This leads to an important difference that might explain the contradictory results: the local- versus global-level processing involved in these tasks. Roser and colleagues43 argued that their findings were attributed to the significant engagement of local processing, a cognitive style in which autistic individuals often excel compared to neurotypical peers (44, but again, see46 for contradicting evidence). It is likely that our task, in comparison with the spatially distributed one used by Roser et al.43, requires more global-level integration: if participants fail to integrate the elements that successively occur, their statistical learning might be weaker. Although we acknowledge that acquiring spatially distributed regularities requires global-level integration as well, autistic individuals seem to benefit from a relative predominance of local-level processing45. Thus, the difference between our and Roser and colleagues’43 results may not at all derive from statistical learning, but from the atypicality of local/global processing.
Besides the general information processing style, atypically high and inflexible precision of prediction errors in ASD3 could account for the benefit of probabilistic tasks compared to deterministic ones. Such errors lead autistic people to update the model after each error, rather than contributing the errors to the unavoidable imprecision of the prediction itself. This has an important implication regarding our probabilistic statistical learning task: the constant update of the model might be adaptive in a task where the regularity cannot be fully learned due to its probabilistic nature. Thus, the constant update based on the prediction errors might lead to a longer learning process—the learning curve of neurotypical participants might peak sooner, as they do not update their model after a certain point, attributing the prediction errors to the imprecision of the otherwise correct model. Meanwhile, ASD participants might keep updating, thus, learning (see also the work of Gazzaniga65 about frequency-maximizing and frequency-matching strategies). This idea highlights the possibility that autistic predictive processing might depend on the given task type. Yet, this topic needs further investigation as some empirical evidence does not even support the different weighting of prediction errors in ASD (see39), and studies have suggested that some statistical learning tasks are not error-driven66,67.
It also implies that task length might affect ASD participants differently than neurotypical participants. Namely, neurotypical participants might outperform ASD participants on shorter tasks, but given enough time, ASD participants can catch up, or maybe even exceed the performance of neurotypical ones. Empirical evidence indeed supports this idea. Autistic participants tend to differ from neurotypical ones only in early learning68. Although they draw on prior knowledge less than neurotypical individuals, their priors are dominated by longer-term statistics of preceding stimuli, rather than recent ones69,70,71. Perhaps as a consequence of the above, they can catch up72 or even outperform their neurotypical peers by the end of the task (42—note, however, that this difference was only trend-level). Given enough time to learn, the constant updating of the representations might be adaptive in statistical learning. Another potential explanation is that, according to meta-analytic evidence, the overall global/local processing is similar in the autistic and neurotypical groups, but autistic people need more time for global processing than neurotypical people46—which might influence learning processes that require global processing. The slower learning dynamics might be an important methodological consideration, as most SRT/ASRT studies where ASD participants performed well, used longer (> 15 min long) learning sessions40,41,42—and our study, with about 40 min of practice provided another example for this. Taken together, the predictive processing of the autistic brain might lead to intact (or if supported by local processing, even superior) performance in case of probabilistic regularities. However, future studies shall address this question to be able to draw firm conclusions.
Atypical use of prior knowledge (vs. using primarily mere sensory input) in ASD might be another way to explain the results. Although empirical evidence often does not support the view that autistic individuals apply weak priors (e.g.62,63,64 for review see6,7), this might help to understand our results. Performance on probabilistic tasks might benefit more from bottom-up than top-down processes: one has to rely on bottom-up processes, as prior knowledge cannot predict the next event with a 100% probability. Thus, performance on the ASRT task potentially benefits more from bottom-up processes26 while using priors might even hinder it. With a real-life example, learning the grammar of a foreign language can be harder if we are proficient in another language already: the regularities we learned before in another language can automatically come to our minds instead of the correct grammar. In conclusion, while attributing lower weight to priors might harm performance on some predictive processing tasks, complex probabilistic task performance can even benefit from it.
A growing body of literature aims to capture another type of uncertainty in the prediction process. According to Palmer et al.10 and Lawson et al.8, autistic people in fact struggle with the estimation of volatility, rather than the estimation of the noise inherently present even when the regularity remains the same. Overestimating volatility leads to an aberrant learning process, which adds to the interpretation of our current results: although the ASRT task operates with some uncertainty (as in it is probabilistic), it is not volatile at all, which might explain the intact performance. This issue could be deeper understood by adding volatility to the ASRT task, for example by switching between different sequences to learn (see for example73,74). Such a study would provide insight into how different types of uncertainties affect learning in ASD. Moreover, using computational models such as hierarchical Gaussian filter would enable us to track the learning of volatility individually, c.f. Lawson et al.8. Given that volatility appears to offer an excellent explanation for our results, it would be particularly worthwhile for future studies to explore this concept.
However, statistical learning studies only ever have found an impairment in autistic children, not in adults. Moreover, all the previous studies that used our task showed no statistical learning impairment in autistic children41,42, which is in line with our findings on adults. All the studies to date, however, compare autistic individuals to neurotypical peers—to our knowledge, no study to date compared the statistical learning performance of autistic children with autistic adults—even though it might be of relevance, as statistical learning tends to change over the lifespan: neurotypical children can outperform adults on probabilistic tasks19,47. Most empirical evidence, including this current paper, suggests similar statistical learning throughout the lifespan in autistic and neurotypical individuals. On the other hand, the nature of the task (e.g., probabilistic/deterministic) might affect this as well, as results found on the SRT task in neurotypical children show a different developmental curve than on the ASRT task47,48,75, moreover, several functions show an altered developmental curve in ASD (see7 for review)—thus, we need further empirical evidence that directly tests this question.
Taken together, our paper aimed to investigate statistical learning in autistic adults from the predictive processing point of view. Predicting probabilistic, temporally distributed regularities seems to be intact, but not superior in ASD. It raises the possibility that predictive processing in ASD, even if it is atypical, can result in intact performance. Importantly, atypicality might affect the performance differently in seemingly similar tasks—here, we discussed how certain factors may contribute to predictive processing in ASD. We would like to inspire future studies not to consider predictive processing as a monolithic concept—for example, the same mechanisms might impair the performance in a deterministic task but not in a complex, probabilistic one. Furthermore, it might be useful for clinicians too; we suggest using strength-based methods in therapy and education of ASD patients, e.g., using probabilistic methods or giving enough time. These suggestions might help understand more about autistic predictive processing, and to autistic individuals to reach their best competencies.
Data availability
The raw datasets and the analyzed data for the current study are available at the following link: https://osf.io/mebcx/. The code to preprocess the raw data is available on GitHub: https://github.com/OrsPesthy/ASDstatlearning.
References
Gregory, R. L. Perceptions as hypotheses. Philos. Trans. R. Soc. Lond. B Biol. Sci. 290(1038), 181–197. https://doi.org/10.1098/RSTB.1980.0090 (1980).
Sinha, P. et al. Autism as a disorder of prediction. Proc. Natl. Acad. Sci. 111(42), 15220–15225. https://doi.org/10.1073/PNAS.1416797111 (2014).
van de Cruys, S. et al. Precise minds in uncertain worlds: Predictive coding in autism. Psychol. Rev. 121(4), 649–675. https://doi.org/10.1037/a0037665 (2014).
Pellicano, E. & Burr, D. When the world becomes “too real”: A Bayesian explanation of autistic perception. Trends Cogn. Sci. 16(10), 504–510. https://doi.org/10.1016/J.TICS.2012.08.009 (2012).
Brock, J. Alternative Bayesian accounts of autistic perception: Comment on Pellicano and Burr. Trends Cogn. Sci. 16(12), 573–574. https://doi.org/10.1016/j.tics.2012.10.005 (2012).
Angeletos Chrysaitis, N. & Seriès, P. 10 years of Bayesian theories of autism: A comprehensive review. Neurosci. Biobehav. Rev. 145, 105022. https://doi.org/10.1016/j.neubiorev.2022.105022 (2022).
Hadad, B.-S. & Yashar, A. Sensory perception in autism: What can we learn. Annu. Rev. Vis. Sci. 8, 239–264. https://doi.org/10.1146/ANNUREV-VISION-093020-035217 (2022).
Lawson, R. P., Mathys, C. & Rees, G. Adults with autism overestimate the volatility of the sensory environment. Nat. Neurosci. 20(9), 1293–1299. https://doi.org/10.1038/nn.4615 (2017).
Mathys, C. D. et al. Uncertainty in perception and the hierarchical Gaussian filter. Front. Hum. Neurosci. 8, 825. https://doi.org/10.3389/fnhum.2014.00825 (2014).
Palmer, C. J., Lawson, R. P. & Hohwy, J. Bayesian approaches to autism: Towards volatility, action, and behavior. Psychol. Bull. 143(5), 521–542. https://doi.org/10.1037/BUL0000097 (2017).
Van De Cruys, S., Perrykkad, K. & Hohwy, J. Explaining hyper-sensitivity and hypo-responsivity in autism with a common predictive coding-based mechanism. Cogn. Neurosci. 10(3), 164–166. https://doi.org/10.1080/17588928.2019.1594746 (2019).
Palmer, C. J., Seth, A. K. & Hohwy, J. The felt presence of other minds: Predictive processing, counterfactual predictions, and mentalising in autism. Conscious Cogn. 36, 376–389. https://doi.org/10.1016/J.CONCOG.2015.04.007 (2015).
Cannon, J., O’Brien, A. M., Bungert, L. & Sinha, P. Prediction in autism spectrum disorder: A systematic review of empirical evidence. Autism Res. 14(4), 604–630. https://doi.org/10.1002/AUR.2482 (2021).
Clark, A. Whatever next? Predictive brains, situated agents, and the future of cognitive science. Behav. Brain Sci. 36(3), 181–204. https://doi.org/10.1017/S0140525X12000477 (2013).
Walsh, K. S., McGovern, D. P., Clark, A. & O’Connell, R. G. Evaluating the neurophysiological evidence for predictive processing as a model of perception. Ann. N. Y. Acad. Sci. 1464(1), 242–268. https://doi.org/10.1111/NYAS.14321 (2020).
Utzerath, C., Schmits, I. C., Buitelaar, J. & de Lange, F. P. Adolescents with autism show typical fMRI repetition suppression, but atypical surprise response. Cortex 109, 25–34. https://doi.org/10.1016/J.CORTEX.2018.08.019 (2018).
Behrens, T. E. J., Woolrich, M. W., Walton, M. E. & Rushworth, M. F. S. Learning the value of information in an uncertain world. Nat. Neurosci. 10(9), 1214–1221. https://doi.org/10.1038/nn1954 (2007).
Howard, J. H. & Howard, D. V. Age differences in implicit learning of higher order dependencies in serial patterns. Psychol. Aging 12(4), 634–656. https://doi.org/10.1037/0882-7974.12.4.634 (1997).
Nemeth, D., Janacsek, K. & Fiser, J. Age-dependent and coordinated shift in performance between implicit and explicit skill learning. Front. Comput. Neurosci. 7, 147. https://doi.org/10.3389/fncom.2013.00147 (2013).
Manning, C., Kilner, J., Neil, L., Karaminis, T. & Pellicano, E. Children on the autism spectrum update their behaviour in response to a volatile environment. Dev. Sci. 20(5), e12435. https://doi.org/10.1111/DESC.12435 (2017).
Robic, S. et al. Decision-making in a changing world: A study in autism spectrum disorders. J. Autism Dev. Disord. 45(6), 1603–1613. https://doi.org/10.1007/s10803-014-2311-7 (2015).
Solomon, M. et al. Feedback-driven trial-by-trial learning in autism spectrum disorders. Am. J. Psychiatry 172(2), 173–181. https://doi.org/10.1176/appi.ajp.2014.14010036 (2015).
Goris, J. et al. The relation between preference for predictability and autistic traits. Autism Res. 13(7), 1144–1154. https://doi.org/10.1002/AUR.2244 (2020).
Keifer, C. M., Day, T. C., Hauschild, K. M. & Lerner, M. D. Social and nonsocial reward anticipation in typical development and autism spectrum disorders: Current status and future directions. Curr. Psychiatry Rep. 23(6), 1–6. https://doi.org/10.1007/s11920-021-01247-7 (2021).
Conway, C. M. How does the brain learn environmental structure? Ten core principles for understanding the neurocognitive mechanisms of statistical learning. Neurosci. Biobehav. Rev. 112, 279–299. https://doi.org/10.1016/J.NEUBIOREV.2020.01.032 (2020).
Janacsek, K. & Nemeth, D. Predicting the future: From implicit learning to consolidation. Int. J. Psychophysiol. 83(2), 213–221. https://doi.org/10.1016/J.IJPSYCHO.2011.11.012 (2012).
Schapiro, A. & Turk-Browne, N. ‘Statistical learning. In Brain mapping: An encyclopedic reference (ed. Toga, A. W.) 501–506 (Academic Press, 2015).
Christiansen, M. H., Conway, C. M. & Onnis, L. Similar neural correlates for language and sequential learning: Evidence from event-related brain potentials. Lang. Cogn. Process. 27(2), 231–256. https://doi.org/10.1080/01690965.2011.606666 (2012).
Lieberman, M. D. Intuition: A social cognitive neuroscience approach. Psychol. Bull. 126(1), 109–136. https://doi.org/10.1037/0033-2909.126.1.109 (2000).
Horváth, K., Nemeth, D. & Janacsek, K. Inhibitory control hinders habit change. Sci. Rep. 12(1), 1–11. https://doi.org/10.1038/s41598-022-11971-6 (2022).
American Psychiatric Association. Diagnostic and statistical manual of mental disorders. Am. Psychiatr. Assoc. https://doi.org/10.1176/appi.books.9780890425596 (2013).
Gordon, B. & Stark, S. Procedural learning of a visual sequence in individuals with autism. Focus Autism Dev. Disabl. 22(1), 14–22. https://doi.org/10.1177/10883576070220010201 (2007).
Mostofsky, S. H., Goldberg, M. C., Landa, R. J. & Denckla, M. B. Evidence for a deficit in procedural learning in children and adolescents with autism: Implications for cerebellar contribution. J. Int. Neuropsychol. Soc. 6(7), 752–759. https://doi.org/10.1017/S1355617700677020 (2000).
Travers, B. G. et al. Spatial and identity cues differentially affect implicit contextual cueing in adolescents and adults with autism spectrum disorder. J. Autism Dev. Disord. 43(10), 2393–2404. https://doi.org/10.1007/S10803-013-1787-X (2013).
Izadi-Najafabadi, S., Mirzakhani-Araghi, N., Miri-Lavasani, N., Nejati, V. & Pashazadeh-Azari, Z. Implicit and explicit motor learning: Application to children with autism spectrum disorder (ASD). Res. Dev. Disabil. 47, 284–296. https://doi.org/10.1016/j.ridd.2015.09.020 (2015).
Sharer, E. et al. Neural correlates of visuomotor learning in autism. J. Child. Neurol. 30(14), 1877–1886. https://doi.org/10.1177/0883073815600869 (2015).
Travers, B. G., Klinger, M. R., Mussey, J. L. & Klinger, L. G. Motor-linked implicit learning in persons with autism spectrum disorders. Autism Res. 3(2), 68–77. https://doi.org/10.1002/aur.123 (2010).
Treves, I. N. et al. Autistic adults show intact learning on a visuospatial serial reaction time task. J. Autism Dev. Disord. 2023, 1–9. https://doi.org/10.1007/S10803-023-05894-Y (2023).
Ward, E. K., Buitelaar, J. K. & Hunnius, S. Implicit learning in 3-year-olds with high and low likelihood of autism shows no evidence of precision weighting differences. Dev. Sci. 25(2), e13158. https://doi.org/10.1111/DESC.13158 (2022).
Brown, J., Aczel, B., Jiménez, L., Kaufman, S. B. & Grant, K. P. Intact implicit learning in autism spectrum conditions. Quart. J. Exp. Psychol. 63(9), 1789–1812. https://doi.org/10.1080/17470210903536910 (2010).
Nemeth, D. et al. Learning in autism: Implicitly superb. PLoS ONE 5(7), 1–7. https://doi.org/10.1371/journal.pone.0011731 (2010).
Barnes, K. A. et al. Intact implicit learning of spatial context and temporal sequences in childhood autism spectrum disorder. Neuropsychology 22(5), 563–570. https://doi.org/10.1037/0894-4105.22.5.563 (2008).
Roser, M. E., Aslin, R. N., McKenzie, R., Zahra, D. & Fiser, J. Enhanced visual statistical learning in adults with autism. Neuropsychology 29(2), 163–172. https://doi.org/10.1037/neu0000137 (2015).
Frith, U. & Happé, F. Autism: Beyond “theory of mind”. Cognition 50(1–3), 115–132. https://doi.org/10.1016/0010-0277(94)90024-8 (1994).
Kourkoulou, A., Leekam, S. R. & Findlay, J. M. Implicit learning of local context in autism spectrum disorder. J. Autism Dev. Disord. 42(2), 244–256. https://doi.org/10.1007/s10803-011-1237-6 (2012).
Van Der Hallen, R., Evers, K., Brewaeys, K., Van Den Noortgate, W. & Wagemans, J. Global processing takes time: A meta-analysis on local-global visual processing in ASD. Psychol. Bull. 141(3), 549–573. https://doi.org/10.1037/BUL0000004 (2015).
Janacsek, K., Fiser, J. & Nemeth, D. The best time to acquire new skills: Age-related differences in implicit sequence learning across the human lifespan. Dev. Sci. 15(4), 496–505. https://doi.org/10.1111/j.1467-7687.2012.01150.x (2012).
Juhasz, D., Nemeth, D. & Janacsek, K. Is there more room to improve? The lifespan trajectory of procedural learning and its relationship to the between- and within-group differences in average response times. PLoS ONE 14(7), e0215116. https://doi.org/10.1371/JOURNAL.PONE.0215116 (2019).
Baron-Cohen, S., Wheelwright, S., Skinner, R., Martin, J. & Clubley, E. The autism-spectrum quotient (AQ): Evidence from asperger syndrome/high-functioning autism, males and females, scientists and mathematicians. J. Autism Dev. Disorders 31, 5–17 (2001).
Lord, C. et al. Austism diagnostic observation schedule: A standardized observation of communicative and social behavior. J. Autism Dev. Disord. 19(2), 185–212. https://doi.org/10.1007/BF02211841 (1989).
Lord, C., Rutter, M. & Le Couteur, A. Autism diagnostic interview-revised: A revised version of a diagnostic interview for caregivers of individuals with possible pervasive developmental disorders. J. Autism Dev. Disord. 24(5), 659–685. https://doi.org/10.1007/BF02172145 (1994).
Farkas, B. C., Krajcsi, A., Janacsek, K. & Nemeth, D. The complexity of measuring reliability in learning tasks: An illustration using the alternating serial reaction time task. PsyArXiv https://doi.org/10.31234/OSF.IO/5NW4Y (2022).
Vékony, T., Ambrus, G. G., Janacsek, K. & Nemeth, D. Cautious or causal? Key implicit sequence learning paradigms should not be overlooked when assessing the role of DLPFC (Commentary on Prutean et al.). Cortex 148, 222–226. https://doi.org/10.1016/J.CORTEX.2021.10.001 (2021).
JASP Team, ‘JASP (Version 0.16.1)’ 2022, (accessed 22 August 2022); https://jasp-stats.org/
Harris, C. R. et al. Array programming with NumPy. Nature 585(7825), 357–362. https://doi.org/10.1038/s41586-020-2649-2 (2020).
Hunter, J. D. Matplotlib: A 2D graphics environment. Comput. Sci. Eng. 9(3), 90–95. https://doi.org/10.1109/MCSE.2007.55 (2007).
Mckinney, W. Data structures for statistical computing in Python, (2010).
Howard, D. V. et al. Implicit sequence learning: Effects of level of structure, adult age, and extended practice. Psychol. Aging 19(1), 79–92. https://doi.org/10.1037/0882-7974.19.1.79 (2004).
Peter Rosenfeld, J. & Olson, J. M. Bayesian data analysis: A fresh approach to power issues and null hypothesis interpretation. Appl. Psychophysiol. Biofeedback 46(2), 135–140. https://doi.org/10.1007/S10484-020-09502-Y (2021).
Wagenmakers, E. J., Wetzels, R., Borsboom, D. & van der Maas, H. L. J. Why psychologists must change the way they analyze their data: The case of Psi: Comment on Bem (2011). J. Pers. Soc. Psychol. 100(3), 426–432. https://doi.org/10.1037/A0022790 (2011).
Wagenmakers, E. J. et al. Bayesian inference for psychology. Part I: Theoretical advantages and practical ramifications. Psychon. Bull. Rev. 25(1), 35–57. https://doi.org/10.3758/s13423-017-1343-3 (2018).
Croydon, A., Karaminis, T., Neil, L., Burr, D. & Pellicano, E. The light-from-above prior is intact in autistic children. J. Exp. Child. Psychol. 161, 113–125. https://doi.org/10.1016/J.JECP.2017.04.005 (2017).
Sapey-Triomphe, L. A., Temmerman, J., Puts, N. A. J. & Wagemans, J. Prediction learning in adults with autism and its molecular correlates. Mol. Autism 12(1), 1–17. https://doi.org/10.1186/S13229-021-00470-6 (2021).
Van de Cruys, S., Vanmarcke, S., Van de Put, I. & Wagemans, J. The use of prior knowledge for perceptual inference is preserved in ASD. Clin. Psychol. Sci. 6(3), 382–393. https://doi.org/10.1177/2167702617740955 (2018).
Gazzaniga, M. S. Cerebral specialization and interhemispheric communication: Does the corpus callosum enable the human condition?. Brain 123(7), 1293–1326. https://doi.org/10.1093/brain/123.7.1293 (2000).
Beesley, T. & Shanks, D. R. Investigating cue competition in contextual cuing of visual search. J. Exp. Psychol. Learn. Mem. Cogn. 38(3), 709–725. https://doi.org/10.1037/A0024885 (2012).
Nazlı, İ, Ferrari, A., Huber-Huber, C. & de Lange, F. P. Statistical learning is not error-driven. bioRxiv https://doi.org/10.1101/2022.02.07.479428 (2022).
Solomon, M., Smith, A. C., Frank, M. J., Ly, S. & Carter, C. S. Probabilistic reinforcement learning in adults with autism spectrum disorders. Autism Res. 4(2), 109–120. https://doi.org/10.1002/aur.177 (2011).
Lieder, I. et al. Perceptual bias reveals slow-updating in autism and fast-forgetting in dyslexia. Nat. Neurosci. 22(2), 256–264. https://doi.org/10.1038/s41593-018-0308-9 (2019).
Sapey-Triomphe, L.-A., Weilnhammer, V. A. & Wagemans, J. Associative learning under uncertainty in adults with autism: Intact learning of the cue-outcome contingency, but slower updating of priors. Autism 26(5), 1216–1228. https://doi.org/10.1177/13623613211045026 (2022).
Vishne, G. et al. Slow update of internal representations impedes synchronization in autism. Nat. Commun. 12(1), 1–15. https://doi.org/10.1038/s41467-021-25740-y (2021).
Schipul, S. E. & Just, M. A. Diminished neural adaptation during implicit learning in autism. Neuroimage 125, 332–341. https://doi.org/10.1016/J.NEUROIMAGE.2015.10.039 (2016).
Zolnai, T. et al. Measuring statistical learning by eye-tracking. Exp. Results 3, e10. https://doi.org/10.1017/EXP.2022.8 (2022).
Kobor, A., Janacsek, K., Takacs, A. & Nemeth, D. Statistical learning leads to persistent memory: Evidence for one-year consolidation. Sci. Rep. 7(1), 1–10. https://doi.org/10.1038/s41598-017-00807-3 (2017).
Lukács, Á. & Kemény, F. Development of different forms of skill learning throughout the lifespan. Cogn. Sci. 39(2), 383–404. https://doi.org/10.1111/COGS.12143 (2015).
Acknowledgements
This research was supported by the IDEXLYON Fellowship of the University of Lyon as part of the Programme Investissements d’Avenir (ANR-16-IDEX-0005) (to D.N.); National Brain Research Program (project 2017-1.2.1-NKP-2017-00002); János Bolyai Research Scholarship of the Hungarian Academy of Sciences (to K.J.). Project No. 128016 has been implemented with the support provided by the Ministry of Innovation and Technology of Hungary from the National Research, Development, and Innovation Fund, financed under the NKFI/OTKA K funding scheme (to D.N.). Project No. 124148 has been implemented with the support provided by the Ministry of Innovation and Technology of Hungary from the National Research, Development, and Innovation Fund, financed under the NKFI/OTKA PD funding scheme (to K.J.), and the ÚNKP-22-3 and the ÚNKP-19-2-I-ELTE-332 New National Excellence Program of the Ministry for Culture and Innovation from the source of the National Research, Development and Innovation Fund (O.P. and A.G, respectively). L-A.S-T. was supported by a post-doctoral fellowship of the Fondation pour la Recherche Médicale. The authors are thankful to Bálint Szuromi for his help with participant recruitment and advice on the study design.
Funding
Open access funding provided by Eötvös Loránd University.
Author information
Authors and Affiliations
Contributions
O.P.: conceptualization, methodology, validation, formal analysis, investigation, data curation, writing—original draft, writing—review and editing, visualization. K.F.: conceptualization, methodology, validation, formal analysis, investigation, data curation, writing—review and editing, visualization. L.-A.S.-T.: writing—review and editing. A.G.: data curation, writing—review and editing. E.K.: data curation, writing—review and editing. K.J.: supervision, writing—review and editing. J.M.R.: supervision, writing—review and editing. D.N.: conceptualization, methodology, investigation, data curation, writing—review and editing, supervision, project administration, funding acquisition.
Corresponding author
Ethics declarations
Competing interests
The authors declare no competing interests.
Additional information
Publisher's note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Supplementary Information
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Pesthy, O., Farkas, K., Sapey-Triomphe, LA. et al. Intact predictive processing in autistic adults: evidence from statistical learning. Sci Rep 13, 11873 (2023). https://doi.org/10.1038/s41598-023-38708-3
Received:
Accepted:
Published:
DOI: https://doi.org/10.1038/s41598-023-38708-3
- Springer Nature Limited