Unlocking Hidden Voices – Detecting Intact Language Comprehension in Non-Verbal Autistic Children Using Electroencephalography

Background: A third of the autistic population remains minimally-verbal at school age. In order to tailor early targeted interventions for these individuals, it is crucial that we grasp the scope of their receptive language and cognitive abilities. We are interested in the extent to which minimally-verbal autistic children understand spoken language. Due to difficulties with obtaining reliable scores for this population with standard testing materials, it is necessary to develop passive methods that allow us to assess language comprehension in individual children, without requiring behavioural answers.

Objectives: Our first goal was to develop a neural marker of language comprehension that is sensitive to each individual using electroencephalography (EEG). We sought to compare the individual sensitivity of different child-friendly, covert paradigms, analyses, and EEG systems. Our second goal was to apply these paradigms to test a minimally-verbal child with autism.

Methods: First, we developed three child-friendly paradigms in which we presented identical auditory words (targets) in either a semantically congruent or incongruent context. The semantic contexts consisted of either a single probe word that was related or unrelated with the target (Experiment 1), or a sentence frame that was congruent or incongruent with the target (Experiment 2 and 3). In Experiment 3, we added short visual animations that represented the sentence frame to support the semantic context and increase the children’s engagement. We measured the brain responses of 50 typically-developing children to the target words. We used both traditional univariate analyses of the N400 event-related potential, an index of semantic integration of words into their context, and Multivariate Pattern Analyses (MVPA) to examine whether we could decode the semantic condition from the brain activity. We also assessed the quality of the signal recorded by a low-cost gaming EEG system, Emotiv EPOC+, to that recorded by a research-grade system, Neuroscan Synamps², recording from the two systems simultaneously.
Second, we tested BM, a 9-year-old minimally-verbal autistic child, on Experiment 1, using the EPOC+. We independently verified BM’s preserved language comprehension using the K-BIT-2 and PPVT-4 which we adapted so that she could respond non-verbally.

Results: Our three paradigms yielded a medium to high detection rate of differential brain responses to the two semantic conditions in typically-developing children. We found the highest detection rate (88% of participants) in Experiment 2, using congruent and incongruent sentences, when we used MVPA with the Neuroscan’s data. EPOC+’s data yielded lower sensitivity (about 50% of participants). Our case-study BM showed N400-like effects in two electrodes, consistent with her known preserved semantic comprehension.

Conclusions: Using EEG to record the brain’s response to language, we present a proof-of-concept for a neural index of language processing in typically-developing children and one case with autism. We plan to test the methods in a larger group of non-verbal autistic children. Our results may bring us closer to offering objective and reliable measures of language abilities in this population.