Statistical Word Learning in Children with ASD

Background: Using statistical cues (i.e., transitional probabilities between syllables) to determine word boundaries is critical for learning language; extracting words from a continuous speech stream, in order to learn their meanings, hinges upon this ability. Children with Specific Language Impairment (SLI), a language disorder, are impaired in using transitional probabilities to determine word boundaries (Evans et al., 2009).  Individuals with Autism Spectrum Disorders (ASD) have impaired communication, including delays in acquisition, but the underlying mechanism is poorly understood.  It is unknown if weakness in tracking transitional probabilities contributes to these impairments.  Based on common language difficulties, some have proposed an overlap between language deficits in SLI and those in ASD. Exploring whether children with ASD have a similar weakness in statistical word learning will clarify the relationship between these disorders.

Objectives: We examined whether children with ASD demonstrate similar deficits in using transitional probabilities to extract words from a continuous speech stream as those reported in children with SLI. Similar weaknesses would support a connection between these disorders. In contrast, successful performance would suggest that language delays in ASD reflect different underlying processes.

Methods: To clarify the role that transitional probabilities play in word learning, 17 children with ASD and 24 children with typical development (TD) ages 7-17 were recruited. Participants listened to a 21-minute speech stream containing 12 syllables.  Six combinations of syllables formed trisyllabic “words” with high internal transitional probabilities (32-100%); transitional probabilities of syllables not forming “words” were lower (10-20%). As they listened, children engaged in a drawing task that directed their attention away from explicit processing of the speech stream.  Immediately following presentation of the speech stream, children completed a 36-trial, two-alternative forced-choice test, indicating which trisyllabic “word” sounded more like the speech sounds that they heard while drawing.

Results: Participants had IQ>80 (ASD=103.88(11.46); TD=105.38(11.46) and receptive and expressive vocabulary>80 (expressive vocabulary: ASD=105.44(15.24), TD=110.96(16.33); receptive vocabulary(ASD=109.88(13.95), TD=115.92(10.76)).  Groups were matched on age, verbal IQ, and expressive and receptive language. On the syllables task, the ASD and TD groups performed significantly above chance, and, importantly, did not differ from each other. Children in both groups were better able to detect “words” with the highest internal transitional probabilities (>75%) than words with lower transitional probabilities (<42%).

Conclusions: In contrast to children with SLI, participants with ASD were as sensitive as their IQ- and language-matched peers to the transitional probabilities that distinguished “words” from “non-words” in an artificial miniature language. This finding could indicate that (1) children with SLI and ASD do not share a common underlying mechanism for language impairment; or, (2) the strong language skills in the ASD group precluded the detection of any deficits. The participants in this study had strong language skills that do not reflect the large variability of language skills among people with ASD. The relationship between SLI and ASD may be further clarified in subsequent studies exploring the ability of children with ASD and weaker language skills to use statistical cues to learn words.