Response: Commentary: Perceptual learning in autism: over-specificity and possible remedies

Hila Harris, David Israeli, Nancy J. Minshew, Yoram Bonneh, David J. Heeger, Marlene Behrmann, Dov SagiFrontiers in Integrative Neuroscience 2016 · DOI · OpenAlex

Abstract

In the study of perceptual learning (PL), individuals with Autism Spectrum Disorder (ASD) have been shown to exhibit "over-specificity" (OS)—a phenomenon where training in one context actively interferes with learning in another. This finding has been contested by critics who suggest that OS is a statistical artifact of poor initial performance rather than a unique feature of the autistic brain. This paper provides a formal response to these critiques, re-analyzing individual participant data to demonstrate that OS in ASD is an anomalous learning process independent of initial task sensitivity.

Introduction: The Challenge of Generalization

Perceptual learning typically involves a high degree of "specificity." For instance, practicing a visual task at one screen location often fails to improve performance at a new location. In a landmark 2015 study, Harris et al. discovered that for individuals with ASD, this lack of generalization goes a step further: it becomes "over-specificity." In these cases, initial training at one location actually makes subsequent learning at a new location significantly harder than if no training had occurred at all.

Critics (Mercado et al., 2016) argued that this OS was not a unique characteristic of autism but rather a consequence of higher "initial thresholds." They posited that any learner starting with a lower baseline sensitivity (higher threshold) would naturally show similar patterns of interference. This paper addresses this critique by examining the relationship between initial performance and the degree of over-specificity.

Methods and Data Re-analysis

The authors re-examined data from previous experiments utilizing the Texture Discrimination Task (TDT). In this task, participants identify the orientation of a target hidden within distracting lines, with difficulty modulated by the Stimulus Onset Asynchrony (SOA)—the timing between the target and a masking stimulus.

The study compared two training protocols:

  1. Standard Training: Repetitive practice at a single location.
  2. Dummy Training: A modified protocol interleaving "dummy trials" (no target) to prevent rigid tuning.

An OS Index was calculated for each participant, representing the difference in learning curves between the first and second locations. A positive OS value indicates active interference. The researchers then performed a correlation analysis between these OS indices and Day 1 "initial thresholds."

Key Results and Insights

The re-analysis revealed a fundamental difference in how typical and autistic brains process new information after training. In typically developing (TD) adults, there is a strong negative correlation ($r = -0.66, p < 0.001$) between starting performance and generalization. Counter-intuitively, typical learners who start with "worse" performance (higher thresholds) actually tend to show better transfer to new locations.

Correlation between initial threshold and over-specificity (OS) across groups.

However, the ASD group under standard training conditions broke this rule entirely. As shown in the figure above, their OS values showed no correlation ($r = 0.03, p = 0.94$) with their initial thresholds. This lack of relationship suggests that OS in ASD is not a byproduct of poor initial sensitivity but is instead a distinct, anomalous learning phenotype.

Discussion: Interference and the "Dummy" Remedy

The findings support the hypothesis that OS in ASD represents active interference rather than a simple lack of transfer. The autistic brain appears to create a "template" for the first location that is so rigid it suppresses learning when the stimulus moves. This "over-optimization" to a specific scenario prevents the flexible application of learned skills.

Crucially, the "dummy trial" method—interleaving trials without targets—eliminated this over-specificity in both groups. By providing the brain with a "blank" context, the protocol prevents the formation of an overly rigid internal model, allowing ASD participants to generalize as effectively as their typical peers.

Conclusion

This response confirms that over-specificity is a core feature of perceptual learning in ASD, independent of initial performance levels. While standard repetitive training can "lock" learning in the autistic brain, modified protocols like dummy training offer a potential remedy. These insights have significant implications for designing educational and therapeutic interventions for individuals with ASD, suggesting that variety and "blank" contexts may be essential for fostering flexible, generalizable skills.

Summary updated

Abstract

GENERAL COMMENTARY article Front. Integr. Neurosci., 09 November 2016 Volume 10 - 2016 | https://doi.org/10.3389/fnint.2016.00036
BibTeX
@inproceedings{harris2016response,
  author = {Hila Harris and David Israeli and Nancy J. Minshew and Yoram Bonneh and David J. Heeger and Marlene Behrmann and Dov Sagi},
  title = {{Response: Commentary: Perceptual learning in autism: over-specificity and possible remedies}},
  booktitle = {Frontiers in Integrative Neuroscience},
  year = {2016},
  doi = {10.3389/fnint.2016.00036},
}
Figure