DIVA predictions about speech in MV ASD (Chenausky et al., 2021)

Purpose: Understanding what limits speech development in minimally verbal (MV) children with autism spectrum disorder (ASD) is important for providing highly effective targeted therapies. This preliminary investigation explores the extent to which developmental speech deficits predicted by Directions Into Velocities of Articulators (DIVA), a computational model of speech production, exemplify real phenotypes.Method: Implementing a motor speech disorder in DIVA predicted that speech would become highly variable within and between tokens, while implementing a motor speech plus an auditory processing disorder predicted that DIVA’s speech would become highly centralized (schwa-like). Acoustic analyses of DIVA’s output predicted that acoustically measured phoneme distortion would be similar between the two cases, but that in the former case, speech would show more within- and between-token variability than in the latter case. We tested these predictions quantitatively on the speech of children with MV ASD. In Study 1, we tested the qualitative predictions using perceptual analysis methods. Speech pathologists blinded to the purpose of the study tallied the signs of childhood apraxia of speech that appeared in the speech of 38 MV children with ASD. K-means clustering was used to create two clusters from the group of 38, and analysis of variance was used to determine whether the clusters differed according to perceptual features corresponding to within- and between-token variability. In Study 2, we employed acoustic analyses on the speech of the child from each cluster who produced the largest number of analyzable tokens to test the predictions of differences in within-token variability, between-token variability, and vowel space area.Results: Clusters produced by k-means analysis differed by perceptual features that corresponded to within-token variability. Nonsignificant differences between clusters were found for features corresponding to between-token variability. Subsequent acoustic analyses of the selected cases revealed that the speech of the child from the high-variability cluster showed significantly more quantitative within- and between-token variability than the speech of the child from the low-variability cluster. The vowel space of the child from the low-variability cluster was more centralized than that of typical children and that of the child from the high-variability cluster.Conclusions: Results provide preliminary evidence that subphenotypes of children with MV ASD may exist, characterized by (a) comorbid motor speech disorder and (b) comorbid motor speech plus auditory processing disorder. The results motivate testable predictions about how these comorbidities affect speech.Supplemental Material S1. Case Study participants’ productions of a subset of syllables containing corner vowels ([a, æ, i, u]) and [ʌ] during three baseline assessments. The number of times each response was produced (out of a total of 15 opportunities to produce each target) is in parentheses. Correct imitations of the target syllable are shown in bold. Chenausky, K. V., Brignell, A., Morgan, A. T., Norton, A. C., Tager-Flusberg, H. B., Schlaug, G., & Guenther, F. H. (2021). A modeling-guided case study of disordered speech in minimally verbal children with autism spectrum disorder. American Journal of Speech-Language Pathology. Advance online publication. https://doi.org/10.1044/2021_AJSLP-20-00121Publisher Note: This article is part of the Special Issue: Select Papers From the 2020 Conference on Motor Speech.

目的：深入理解自闭症谱系障碍（ASD）中言语发展迟缓（MV）儿童言语发展的限制因素，对于提供高效精准的针对性治疗方案具有重要意义。本研究初步探讨了由发音器官速度（DIVA）模型预测的发育性言语缺陷在多大程度上体现了真实的表型特征。方法：在DIVA模型中实施运动性言语障碍，预测言语在token内部和之间的变异性将显著增加；同时实施运动性言语障碍与听觉处理障碍，预测DIVA的言语将高度集中（类似元音）。对DIVA输出的声学分析预测，两种情况下，声学测量的音素扭曲度相似，但在前者情况下，言语的token内部和之间的变异性将高于后者。我们对MV ASD儿童的语言进行了定量测试。在研究1中，我们使用感知分析的方法测试了定性预测。对研究目的不知情的言语治疗师统计了38名MV ASD儿童言语中出现的儿童言语失用症的症状。使用K-means聚类将38人分成两组，并使用方差分析来确定两组在token内部和之间的变异性对应的感知特征上是否存在差异。在研究2中，我们对每个聚类中产生最多可分析token的儿童的语言进行了声学分析，以测试token内部变异性、token之间变异性以及元音空间面积差异的预测。结果：K-means分析产生的聚类在token内部变异性的感知特征上存在差异。在token之间变异性的特征上，两组之间没有发现显著差异。随后对所选案例的声学分析显示，高变异性聚类中的儿童言语在token内部和之间的变异性显著高于低变异性聚类中的儿童言语。低变异性聚类中的儿童元音空间比典型儿童和变异性高聚类中的儿童更为集中。结论：结果表明，MV ASD儿童的亚表型可能存在，特征为（a）合并的运动性言语障碍和（b）合并的运动性言语障碍与听觉处理障碍。这些合并症如何影响言语的结果引发了可测试的预测。补充材料S1：案例研究参与者在一项基于基线评估的子音节测试中的表现，其中包含角元音（[a, æ, i, u]）和[ʌ]。括号内为每种反应产生的次数（总共15次机会产生每种目标音节），正确的目标音节模仿以粗体显示。Chenausky, K. V., Brignell, A., Morgan, A. T., Norton, A. C., Tager-Flusberg, H. B., Schlaug, G., & Guenther, F. H. (2021). 自闭症谱系障碍中言语发展迟缓儿童言语障碍的建模指导案例研究。美国言语-语言病理学杂志。在线预发布。https://doi.org/10.1044/2021_AJSLP-20-00121。出版商注：本文是特刊《2020年运动性言语会议精选论文》的一部分。