Linguistic Markers of Genetic Liabitiy to Autism

Background: Deficits in language discourse (i.e., connected speech such as conversation and narrative) are a hallmark of autism, and a principal feature of the broad autism phenotype, or BAP -- i.e., the constellation of subtle features often observed in relatives of autistic individuals that are qualitatively similar to the core features of autism and are believed to index genetic liability to autism. While existing clinically-based measures of discourse have provided highly valid and richly descriptive information, the vast majority of such measures tend to assess behavior at a functional rather than mechanistic level. Thus, relatively little is known regarding the underlying cognitive mechanisms that give rise to observed discourse abnormalities, and importantly, whether such mechanisms are impaired in both in both autism and the BAP.  

Objectives: This project aimed to clarify the mechanisms underlying discourse impairments in autism and the BAP, which may index genetic liability and be useful for genetic studies. We performed computational linguistic analyses to examine the relative frequency of different classes of words as a measure of psychological processes underlying discourse practices. The use of certain classes of words provides a powerful implicit measure of important psychological processes. This is particularly so for classes of words (e.g., pronouns and articles), which are essential to establishing a referential model of the meaning of a discourse exchange. While impaired processing of linguistic reference has been described in autism, no systematic research has been conducted on autism or the BAP in parents using the comprehensive tools for studying word usage applied in this study.

Methods: The frequency of different word classes was examined in existing language corpora using a computerized tool called Linguistic Inquiry and Word Count, or LIWC (Pennebaker et al., 2001). LIWC was designed to analyze psychological characteristics of individuals, and social characteristics of interactions, using samples of natural language (either written language or transcripts of spoken language). LIWC was applied to transcripts of discourse produced during the semi-structured interviews with 28 high-functioning individuals with autism and 23 age and IQ-matched controls, as well as transcripts of comparable language samples from 15 autism parents who had been clinically rated as BAP (+) using established methods, and 11 parents rated as BAP (-).

Results: Specific patterns of word usage were detected in both individuals with autism and parents with the BAP. Specifically, pronouns and other devices used to create coherence in discourse, discriminated individuals with autism from controls (p values < .05) and parents with the BAP from parents without the BAP (p < .01). Discriminant function analysis was conducted using these linguistic measures as predictors of autism, and the clinically-based BAP classification of the parents. Analyses correctly classified 92% of children and 84.6% of parents.

Conclusions: Findings indicate specific patterns of word class usage that may underlie the discourse abnormalities associated with autism and the BAP. These features appear to serve as markers capable of defining with strong sensitivity individuals with autism and parents with the BAP.