Dalal Tyler C, Liang Liangbing, Silva Angelica M, Mackinley Michael, Voppel Alban, Palaniyappan Lena
Schulich School of Medicine and Dentistry, Western University, London, Ontario, Canada.
Robarts Research Institute, London, Ontario, Canada.
Acta Psychiatr Scand. 2025 Mar;151(3):332-347. doi: 10.1111/acps.13685. Epub 2024 Apr 10.
Speech markers are digitally acquired, computationally derived, quantifiable set of measures that reflect the state of neurocognitive processes relevant for social functioning. "Oddities" in language and communication have historically been seen as a core feature of schizophrenia. The application of natural language processing (NLP) to speech samples can elucidate even the most subtle deviations in language. We aim to determine if NLP based profiles that are distinctive of schizophrenia can be observed across the various clinical phases of psychosis.
Our sample consisted of 147 participants and included 39 healthy controls (HC), 72 with first-episode psychosis (FEP), 18 in a clinical high-risk state (CHR), 18 with schizophrenia (SZ). A structured task elicited 3 minutes of speech, which was then transformed into quantitative measures on 12 linguistic variables (lexical, syntactic, and semantic). Cluster analysis that leveraged healthy variations was then applied to determine language-based subgroups.
We observed a three-cluster solution. The largest cluster included most HC and the majority of patients, indicating a 'typical linguistic profile (TLP)'. One of the atypical clusters had notably high semantic similarity in word choices with less perceptual words, lower cohesion and analytical structure; this cluster was almost entirely composed of patients in early stages of psychosis (EPP - early phase profile). The second atypical cluster had more patients with established schizophrenia (SPP - stable phase profile), with more perceptual but less cognitive/emotional word classes, simpler syntactic structure, and a lack of sufficient reference to prior information (reduced givenness).
The patterns of speech deviations in early and established stages of schizophrenia are distinguishable from each other and detectable when lexical, semantic and syntactic aspects are assessed in the pursuit of 'formal thought disorder'.
言语标记是通过数字方式获取、经计算得出的可量化测量指标集，反映与社会功能相关的神经认知过程状态。语言和交流中的“怪异之处”历来被视为精神分裂症的核心特征。将自然语言处理（NLP）应用于言语样本可以阐明语言中即使是最细微的偏差。我们旨在确定在精神病的各个临床阶段是否能观察到基于NLP的、有精神分裂症特征的概况。
我们的样本包括147名参与者，其中有39名健康对照者（HC）、72名首发精神病患者（FEP）、18名临床高危状态者（CHR）、18名精神分裂症患者（SZ）。一项结构化任务引出了3分钟的言语，然后将其转化为关于12个语言变量（词汇、句法和语义）的定量测量指标。然后应用利用健康变异的聚类分析来确定基于语言的亚组。
我们观察到一个三类解决方案。最大的一类包括大多数健康对照者和大多数患者，表明是“典型语言概况（TLP）”。其中一个非典型类在词汇选择上具有显著高的语义相似性，感知性词汇较少，衔接性和分析结构较低；这类几乎完全由精神病早期患者组成（EPP - 早期概况）。第二个非典型类有更多已确诊精神分裂症的患者（SPP - 稳定期概况），有更多感知性但认知/情感类词汇较少，句法结构更简单，且对先前信息缺乏充分提及（既定信息减少）。
在精神分裂症的早期和确诊阶段，言语偏差模式彼此有别，并且在评估“形式思维障碍”时，从词汇、语义和句法方面进行评估时是可检测到的。
