Implicit Processing of Category Information in Autism

Background: Statistical learning in low-level contexts is intact in autism. However, statistical learning in more complex contexts has not yet been investigated in autism. The tendency to generalize is reduced in autism and could lead to over-specificity in category formation. Statistical learning also operates at the semantic categorical level in the typical population. If statistical learning is atypical in autism in higher-level contexts, such as at the categorical level, then we should expect atypical predictive ability in individuals with autism.

Objectives: (1) To investigate whether individuals with autism implicitly process statistical regularities occurring in complex images to the same extent as typically developing individuals. (2) To assess whether learning occurs when these regularities are at the level of the semantic category.

Methods: 40 male adults with autism and 40 male controls (matched for age and IQ) took part. Participants were exposed to image streams consisting of natural scenes. The order of image presentation was dictated by a deterministic underlying structure based on the semantic categories of the scene images (e.g., images of mountain ranges would always be followed by images of bathrooms). Participants were randomly assigned to one of two conditions where the underlying structure occurred at the level of the individual image or occurred at the level of the image categories. In the first condition a single image was used to represent each semantic category. However, in the second condition a novel image was presented for each incidence of a particular semantic category. Thus statistical learning could occur in the first condition without attending to the semantic category, but not in the second condition. A cover task was used to ensure participants did not explicitly attend to image-order. A forced two-choice paradigm was then used to assess participant’s ability to correctly identify images occurring in the correct order over foil sequences. 

Results: Individuals with autism showed equal performance to controls when the underlying structure was based on individual images. This suggests that individuals with autism do not show any impairment in implicit learning even when the stimuli used are complex, real-world images. However, performance in the autism group was below controls when the structure occurred at the semantic level of the category.

Conclusions: Individuals with ASC do not automatically process complex images in the same way as controls. Specifically, the semantic context of images is not processed as much in people with autism compared to controls. This may reflect their greater attention to detail. These findings have implications for how environmental regularities are processed and for learning, in individuals with autism.