Predictive language processing in young autistic children

Abstract

Recent theories propose that domain-general deficits in prediction (i.e., the ability to anticipate upcoming information) underlie the behavioral characteristics associated with autism spectrum disorder (ASD). If these theories are correct, autistic children might be expected to demonstrate difficulties on linguistic tasks that rely on predictive processing. Previous research has largely focused on older autistic children and adolescents with average language and cognition. The present study used an eye-gaze task to evaluate predictive language processing among 3- to 4-year-old autistic children (n = 34) and 1.5- to 3-year-old, language-matched neurotypical (NT) children (n = 34). Children viewed images (e.g., a cake and a ball) and heard sentences with informative verbs (e.g., Eat the cake) or neutral verbs (e.g., Find the cake). Analyses of children's looking behaviors indicated that young autistic children, like their language-matched NT peers, engaged in predictive language processing. Regression results revealed a significant effect of diagnostic group, when statistically controlling for age differences. The NT group displayed larger difference scores between the informative and neutral verb conditions (in looks to target nouns) compared to the ASD group. Receptive language measures were predictive of looking behavior across time for both groups, such that children with stronger language skills were more efficient in making use of informative verbs to process upcoming information. Taken together, these results suggest that young autistic children can engage in predictive processing though further research is warranted to explore the developmental trajectory relative to NT development. LAY SUMMARY: This study found that 3- to 4-year-old autistic children and younger, language-matched neurotypical (NT) children both used verbs to predict upcoming nouns in sentences like "Eat the cake." For both autistic and NT children, those with stronger language skills were able to predict upcoming nouns more quickly.

Keywords: anticipatory language processing; autism spectrum disorder; language development; prediction; receptive language.

Figure 1. Trial Schematic

Note. The total duration of each trial was 6000 ms. The cue sentence duration was 3818 ms. The onset of the verb (e.g., eat/find) occurred 1800 ms after trial onset, and the onset of the target noun (e.g., cake) occurred 2790 ms after trial onset, such that verb onset was 990 ms before noun onset. The regression analysis window began at 200 ms after verb onset and ended 200 ms after noun onset (1000 ms total duration) from 2000 ms to 3000 ms from trial onset. Following the cue sentence, in order to help maintain children’s attention to the task, a reward sentence (e.g., That’s great!) followed each target noun.

Figure 2. Cluster Permutation Analysis Results

Note. Proportion of target looks for children with autism spectrum disorder (ASD, n = 34) and neurotypical (NT) children (n = 34) during Informative sentences (green) and Neutral sentences (grey). The onset of the target noun (e.g., cake) is at 0 ms. Line shading represents one standard error from the mean, averaged by subjects. Horizontal dashed line indicates chance performance. The full analysis window for the cluster permutation analysis (1000 ms before noun onset to 1000 ms after noun onset) is depicted. Area shading indicates significant clusters from permutation analyses (ps < 0.001).

Figure 3. Regression Analysis Results

Note. Children’s proportion of looks to the target image (e.g., cake) during Informative sentences (e.g., Eat the cake) minus their proportion of looks to the target image during Neutral sentences (e.g., Find the cake) for neurotypical (NT) children (n = 34) and children with ASD (n = 34). To aid in visualizing regression results, we collapsed across groups and divided children into a Lower PLS group (pink) and a Higher PLS group (orange) based on a median split of PLS-5 Auditory Comprehension raw scores. Target noun onset is at 0 ms. Vertical dashed lines indicate the analysis time window (200 ms after verb onset to 200 ms after noun onset). Horizontal dashed lines indicate equivalent target looks for Informative and Neutral conditions. Line shading indicates one standard error from the mean, averaged by subjects.