Pre-trained artificial intelligence language model represents pragmatic language variability central to autism and genetically related phenotypes.

Many individuals with autism experience challenges using language in social contexts (i.e., pragmatic language). Characterizing and understanding pragmatic variability is important to inform intervention strategies and the etiology of communication challenges in autism; however, current manual coding-based methods are often time and labor intensive, and not readily applied in ample sample sizes. This proof-of-concept methodological study employed an artificial intelligence pre-trained language model, Bidirectional Encoder Representations from Transformers, as a tool to address such challenges. We applied Bidirectional Encoder Representations from Transformers to computationally index pragmatic-related variability in autism and in genetically related phenotypes displaying pragmatic differences, namely, in parents of autistic individuals, fragile X syndrome, and FMR1 premutation. Findings suggest that without model fine-tuning, Bidirectional Encoder Representations from Transformers's Next Sentence Prediction module was able to derive estimates that differentiate autistic from non-autistic groups. Moreover, such computational estimates correlated with manually coded characterization of pragmatic abilities that contribute to conversational coherence, not only in autism but also in the other genetically related phenotypes. This study represents a step forward in evaluating the efficacy of artificial intelligence language models for capturing clinically important pragmatic differences and variability related to autism, showcasing the potential of artificial intelligence to provide automatized, efficient, and objective tools for pragmatic characterization to help advance the field.Lay abstractAutism is clinically defined by challenges with social language, including difficulties offering on-topic language in a conversation. Similar differences are also seen in genetically related conditions such as fragile X syndrome (FXS), and even among those carrying autism-related genes who do not have clinical diagnoses (e.g., the first-degree relatives of autistic individuals and carriers of the FMR1 premutation), which suggests there are genetic influences on social language related to the genes involved in autism. Characterization of social language is therefore important for informing potential intervention strategies and understanding the causes of communication challenges in autism. However, current tools for characterizing social language in both clinical and research settings are very time and labor intensive. In this study, we test an automized computational method that may address this problem. We used a type of artificial intelligence known as pre-trained language model to measure aspects of social language in autistic individuals and their parents, non-autistic comparison groups, and individuals with FXS and the FMR1 premutation. Findings suggest that these artificial intelligence approaches were able to identify differences in social language in autism, and to provide insight into the individuals' ability to keep a conversation on-topic. These findings also were associated with broader measures of participants' social communication ability. This study is one of the first to use artificial intelligence models to capture important differences in social language in autism and genetically related groups, demonstrating how artificial intelligence might be used to provide automatized, efficient, and objective tools for language characterization.