This study investigates the use of cosine similarity measures across syntactic, lexical, and semantic vector repre- sentations to detect repetitions in the spontaneous speech of autistic children. It focuses on direct repetitions (i.e., immediate verbatim repetitions of linguistic output produced by another individual) and self-repetitions (i.e., within-speaker recurrence). The performance of similarity-based methods is then compared with state-of-the-art black-box classification models based on BERT, trained on the same data. Using spontaneous speech data from French- and Dutch- speaking autistic children, the results show that lexical and semantic similarity provide reliable cues for identifying self-repetitions, achieving high precision and recall, with F1-scores exceeding 83%, comparable to those obtained by BERT-based models. In contrast, direct repetitions are more difficult to detect using similarity-based approaches, with BERT models clearly outperforming them and reaching F1-scores above 73%. Across all conditions, syntactic similarity consistently underperforms relative to lexical and semantic measures. These findings highlight the strengths and limitations of similarity-based approaches and suggest directions for future research, particularly in improving the detection of direct repetitions and assessing the cross-linguistic generalizability of these methods.

Autism Spectrum Disorders (ASD) are a group of complex developmental conditions whose effects and severity show high intraindividual variability. However, one of the main symptoms shared along the spectrum is social interaction impairments that can be explored through acoustic analysis of speech production. In this paper, we compare 14 Italian-speaking children with ASD and 14 typically developing peers. Accordingly, we extracted and selected the acoustic features related to prosody, quality of voice, loudness, and spectral distribution using the parameter set eGeMAPS provided by the openSMILE feature extraction toolkit. We implemented four supervised machine learning methods to evaluate the extraction performances. Our findings show that Decision Trees (DTs) and Support Vector Machines (SVMs) are the best-performing methods. The overall DT models reach a 100% recall on all the trials, meaning they correctly recognise autistic features. However, half of its models overfit, while SVMs are more consistent. One of the results of the work is the creation of a speech pipeline to extract Italian speech biomarkers typical of ASD by comparing our results with studies based on other languages. A better understanding of this topic can support clinicians in diagnosing the disorder.