Applying Computational Linguistics to Fundamental Components of Schizophrenia

描述（由申请人提供）：精神分裂症是一种独特的人类疾病，对人类独特的语言能力有特定影响。事实上，精神分裂症的语言异常可以被视为基本的疾病组成部分，甚至可能是“生物特征”的一部分。“将现代语言学知识和工具引入这种疾病是一种有前途的方法。我们已经形成了独特的跨学科合作（康普顿博士，精神分裂症研究者;科文顿博士，计算语言学家;伦登博士，语音学语言学家;克利里博士，统计学家;和布兰查德博士，一位测量阴性症状的专家）来研究精神分裂症的一些最令人困惑和致残的方面，语言/言语异常与混乱和阴性症状密切相关。我们将分析精神分裂症患者和正常对照组的言语异常。我们将从语言学的“句法”、“语义”、“语用”和“语音”领域来评估语音，而不是考察单个语言学参数。我们将在这一研究领域引入前沿创新通过使用卡温顿博士小组开发的心理语言学软件评估这些指标这样我们对言语异常的评级将是高度客观和超可靠的。 我们的长期目标是开发多变量模型，以及用于临床和研究环境的新方法，这些方法基于计算语言指数，具有自动化的固有可靠性和已证明的有效性。在这项探索性/发展性研究中，我们将收集100名精神分裂症患者的详细症状评分，以及这些患者和100名对照组的录音语音样本和神经认知评分。本研究涉及可能产生重大转化影响的新工具和模型的早期/概念阶段。我们努力获得新知识，然后将其付诸行动。例如，我们的新方法可以转化为先进的临床应用（例如，高度可靠的、基于语音的症状进展或缓解监测）。此外，我们的新模型和方法可能是迈向有前途的预测模型的第一步（例如，对高危青少年风险预测有用的因素组合）。这些目标与NIMH战略计划高度一致。我们的四个目标是：（1）使用计算机语音分析检查句法、语义和语用语言参数，并评估它们与紊乱症状的关系;（2）使用计算机语音傅立叶频谱分析检查语音语言参数，并评估它们与阴性症状的关系;（3）确定最能预测患者与对照状态的心理语言参数的组合;以及（4）确定最能预测患者的混乱评分和阴性症状评分的心理语言学参数的组合。鉴于我们将收集的丰富数据，我们还将能够协变药物和物质使用的影响;检查基于中性v.情感负载内容和自发v.阅读语音的结果的变化;评估归因于认知领域的语言测量的差异;并比较首次发作和慢性患者的结果。

DESCRIPTION (provided by applicant): Schizophrenia is a uniquely human disorder with specific effects on the uniquely human capacity of language. Indeed, the gross and subtle language abnormalities of schizophrenia can be seen as fundamental illness components, perhaps even as part of a "biosignature." Bringing modern linguistics knowledge and tools to this disorder is a promising approach. We have formed a unique, inter-disciplinary collaboration (Dr. Compton, a schizophrenia researcher; Dr. Covington, a computational linguist; Dr. Lunden, a linguist specializing in phonetics; Dr. Cleary, a statistician; and Dr. Blanchard, an expert in measuring negative symptoms) to study some of the most perplexing and disabling facets of schizophrenia, the language/speech abnormalities linked closely to disorganization and negative symptoms. We will analyze speech abnormalities in patients with schizophrenia and unaffected controls. Rather than examining a single linguistic parameter, we will assess speech in "syntactic," "semantic," "pragmatic," and "phonetic" domains of linguistics. We will introduce cutting- edge innovation to this area of study by assessing these indices using psycholinguistics software developed by Dr. Covington's group so that our ratings of speech abnormalities will be highly objective and ultra-reliable. Our long-term goal is to develop multivariable models, and new methods for clinical and research settings, based on computational linguistic indices with inherent reliability from automation and proven validity. In this exploratory/developmental study, we will collect detailed symptom ratings from 100 schizophrenia patients, as well as audio-recorded speech samples and neurocognition scores from these patients and 100 controls. This study involves early/conceptual stages of new tools and models that could have a major translational impact. We strive to acquire new knowledge and then put it into action. For example, our new methods could translate into advanced clinical applications (e.g., highly reliable, voice-based monitoring of symptom progression or remission). Furthermore, our new models and methods could be a first step toward promising predictive models (e.g., combinations of factors useful in risk prediction among at-risk youth). These objectives are highly aligned with the NIMH Strategic Plan. Our 4 aims are to: (1) examine syntactic, semantic, and pragmatic linguistic parameters using computer analysis of speech, and assess their relation to disorganized symptoms; (2) examine phonetic linguistic parameters using computerized Fourier spectrum analysis of speech, and assess their relation to negative symptoms; (3) determine the combination of psycholinguistic parameters that best predicts patient versus control status; and (4) determine the combination of psycholinguistic parameters that best predicts disorganization scores and negative symptom scores among patients. Given the rich data we will collect, we will also be able to covary the effects of medication and substance use; examine variation in findings based on neutral v. emotionally laden content and spontaneous v. read speech; assess variance in linguistic measures attributable to cognitive domains; and compare results in first-episode and chronic patients.