[Autism detection based on computational analysis of parental language]

Early detection of autism spectrum disorder (ASD) requires integrating clinical observation with developmental history and everyday functioning, yet care pathways are often strained by high demand and limited specialist resources. In this context, parents' narratives provide ecologically valid signals about social communication, language, and repetitive behaviors; however, free-form accounts are difficult to analyze systematically in routine practice. This article reviews a special study, which automatically analyzed caregivers' openended responses to 12 questions inspired by the ADI-R in 51 families (children aged 2-8 years; 25 ASD and 26 controls). The study aimed to automatically detect ASD in children through computational analysis of parents' speech. The best subject-level classification strategy was achieved by converting text into semantic representations (using OpenAI's text-ada-embedding-large-v3), training models per question, including the question in the input, and aggregating decisions via majority voting (84% accuracy; ROC-AUC 1.0). The clinical goal of this pilot study was to demonstrate the potential to support screening, prioritization, and referral, and potentially reduce waiting times and professional burden, thereby improving timely access to evaluation and early intervention. We discuss limitations (small sample, selected population, possible influence of prior intervention) and ethical risks (sociolinguistic bias, privacy, stigmatization, false positives/negatives), and we propose steps toward responsible clinical translation.