Contextual Priors Do Not Modulate Action Anticipation in Children with Autism

Oral Presentation
Saturday, May 12, 2018: 1:45 PM
Willem Burger Zaal (de Doelen ICC Rotterdam)
L. Amoruso1,2, A. Narzisi3, M. Pinzino4, A. Finisguerra5, F. Fabbro5, A. Volzone6, S. Calderoni7, F. Muratori4 and C. Urgesi8,9, (1)Univerisity of Udine, Udine, Italy, (2)Basque Center of Cognition, Brain and Language, San Sebastian, Spain, (3)IRCCS Fondazione Stella Maris, Calambrone (Pisa), Italy, (4)IRCCS Stella Maris Foundation, Calambrone (Pisa), Italy, (5)University of Udine, Udine, Italy, (6)Scientific Institute IRCCS Eugenio Medea, San Vito al Tagliamento, Italy, (7)Department of Clinical and Experimental Medicine, University of Pisa, Pisa, Italy, (8)Dipartimento di Lingue e Letterature, Comunicazione Formazione e Società, Uniersity of Udine, Udine, Italy, (9)Scientific Institute (IRCCS) E. Medea, Bosisio Parini, Italy
Background: Current influential models on action comprehension postulate that optimal intention inference requires the combination of two sources of information:
(a) the observers’ prior expectations about the likely intention driving other’s behaviors, given past experiences; (b) the sensory evidence conveyed by online perceptual movement kinematics.

Objectives: Here, we aimed to investigate the ability of children with ASD to extract regularities from the environment and use them to build-up prior expectations about other’s behavior.

Methods: 22 High-functioning ASD children were evaluated with the ADOS-2. Non-verbal IQ scores were derived, in ASD and Typycal control group (n=22), from the Raven’s Colored Progressive Matrices. Paradigm: (1) Familiarization Phase (Fig.1): Implicit Probabilistic Learning Task. Children observed videos (1slong) depicting a child actor grasping common objects to perform different actions and we asked them to recognize actor’s intention. Importantly, we implicitly biased action-contextual cues associations in terms of their probability of co-occurrence (i.e., given the presence of an orange plate, it was highly probable that the child would grasp the apple for eating); (2) Testing Phase: Action Prediction Task. Children observed the same videos but, in this case, the secondhalf remained occluded from view (500mslong) and participants were asked to predict action unfolding. During this phase, all possible action-contextual cues associations were equally presented. We reasoned that during this phase, where movement kinematics were partially occluded, children’s responses would be biased to contextual priors acquired during the familiarization phase, thus compensating perceptual uncertainty.

Results: A) An independent t-test comparing the overall accuracy during the familiarization phase revealed no differences (t(46) = -1.34, p = 0.18) between ASD (M = 0.87, SEM = 0.03) and typically developing (TD, M = 0.93, SEM = 0.02) children in recognizing actor’s motor intention. B) The RM-ANOVA revealed a significant Group by cooccurrence probability interaction [F(3, 138)=2.89, p=0.03]. Black solid lines indicate between-group comparisons, grey dotted lines within-group comparisons. Bars denote SEM.

Conclusions: No differences were observed between groups in recognizing actor’s intention during the familiarization phase. However, when perceptual evidence was ambiguous, only TD children were able to combine it with contextual priors and successfully predict action unfolding in a probabilistic modulated fashion. Collectively, our results suggest that ASD children have an impairment in building contextual priors and do not benefit from them (i.e., integrate with perceptual evidence) when predicting other people’s actions under situations of perceptual uncertainty.