The effect of volatility in linguistic input on prediction behavior in autistic toddlers.

The children included in the main comparisons were selected after the experiment because they had learned the first word pairings, and the selected autistic and non-autistic groups differed in age, cognitive ability and receptive language. Because of that selection, a true group difference could still be present if children were matched on those characteristics. The same trial order was given to all children because of a pseudorandomization error, and image side was partially confounded with adjective. The task used only two familiar adjective-noun pairs that were not semantically related, so it may not reflect everyday language prediction. The sample was monolingual English-speaking and primarily white and non-Hispanic or Latino, so results may not generalise widely. The non-autistic group had higher maternal education, though the authors say this did not appear to explain the results. The .50 cut-off for including children was arbitrary, although a stricter .55 check gave the same pattern. Analyzing switches separately did not allow interpretation of change across the whole experiment.

The authors reported no conflicts of interest. The supplied metadata lists NIH extramural research support, but the text does not state a funding source or whether the funder designed or wrote the study.

Do not read the autistic children's reduced looking after the first switch as proof that autistic toddlers over-rely on the newest input. The main group comparison was null, and the within-group pattern changed by the second switch.