Machine learning methods for analyzing morphological and lexical characteristics of speech of boys with autism spectrum disorders and Down syndrome

Purpose. In this paper, we propose an approach to identifying significant differences in the speech of typically developing boys (TD), boys with Autism Spectrum Disorder (ASD) and Down syndrome (DS) based on a comparison of morphological and lexical characteristics of their speech. The linguistic characteristics were extracted automatically using the morphological analyzer pymorphy2. Sixty nine boys were interviewed. In total, 45 linguistic features were extracted from each dialogue.Results. The Mann - Whitney U test was used for assessing the differences in linguistic features of speech, and differences were identified for 31 linguistic features of speech of boys with TD and with ASD, 31 linguistic features of speech of boys with TD and with DS, and 15 linguistic features of speech of boys with ASD and with DS. These features were used to build classification models using machine learning methods: gradient boosting, random forest, and AdaBoost algorithm. The identified features showed good separability, and the accuracy of the classification of the dialogues of boys with typical development, autism spectrum disorders and Down syndrome equal to 88 % was achieved.