Early multimodal behavioral cues in autism: a micro-analytical exploration of actions, gestures and speech during naturalistic parent-child interactions

Early signs of autism often emerge through distinct developmental pathways, particularly in communication, social interaction, and play. While naturalistic parent-child interactions during free play are ideal for observing spontaneous social behaviors, few autism studies have adopted this ecological and developmental approach. To address this gap, we used a fine-grained microanalytic method to examine motor, gestural, and vocal behaviors in young children, integrating machine learning to explore how combinations of these traits distinguish early autistic neurodivergence. We analyzed video recordings of 58 autistic and non-autistic children (aged 13–40 months) engaged in naturalistic parent-child play. A frame-by-frame micro-coding scheme was applied to capture actions, gestures, speech, and their multimodal integration. Clear differences emerged between neurotypical (NT) and autistic (ASC) children. NT children displayed more gestures, particularly deictic and conventional-interactive, greater gesture–gaze coordination, more functional object play, and more frequent multi-word utterances. In contrast, ASC children showed fewer deictic and conventional-interactive gestures and greater use of instrumental gestures, reduced gesture–gaze coordination, a higher reliance on vocalizations rather than words, and increased object manipulation compared to functional play. Feature selection using ANOVA F-tests identified a core set of key predictors most frequently and independently selected across folds of cross-validation: Alternate Gaze, Reaching, and Instrumental Gesture. Higher values of Alternate Gaze were associated with NT classification, while elevated frequencies of Reaching and Instrumental Gestures were linked to ASC classification. A logistic regression classifier trained on these features achieved over 85% accuracy in distinguishing the two groups. These findings underscore the value of an ecologically valid, and developmentally informed approach to identifying early behavioral markers of autism, supporting earlier recognition and the design of more personalized, strengths-based interventions.