Changes in Digital Speech Measures in Asymptomatic Carriers of Pathogenic Variants Associated With Frontotemporal Degeneration.

BACKGROUND AND OBJECTIVES: Clinical trials developing therapeutics for frontotemporal degeneration (FTD) focus on pathogenic variant carriers at preclinical stages. Objective, quantitative clinical assessment tools are needed to track stability and delayed disease onset. Natural speech can serve as an accessible, cost-effective assessment tool. We aimed to identify early changes in the natural speech of FTD pathogenic variant carriers before they become symptomatic. METHODS: In this cohort study, speech samples of picture descriptions were collected longitudinally from healthy participants in observational studies at the University of Pennsylvania and Columbia University between 2007 and 2020. Participants were asymptomatic but at risk for familial FTD. Status as "carrier" or "noncarrier" was based on screening for known pathogenic variants in the participant's family. Thirty previously validated digital speech measures derived from automatic speech processing pipelines were selected a priori based on previous studies in patients with FTD and compared between asymptomatic carriers and noncarriers cross-sectionally and longitudinally. RESULTS: A total of 105 participants, all asymptomatic, included 41 carriers: 12 men [30%], mean age 43 ± 13 years; education, 16 ± 2 years; MMSE 29 ± 1; and 64 noncarriers: 27 men [42%]; mean age, 48 ± 14 years; education, 15 ± 3 years; MMSE 29 ± 1. We identified 4 speech measures that differed between carriers and noncarriers at baseline: mean speech segment duration (mean difference -0.28 seconds, 95% CI -0.55 to -0.02, p = 0.04); word frequency (mean difference 0.07, 95% CI 0.008-0.14, p = 0.03); word ambiguity (mean difference 0.02, 95% CI 0.0008-0.05, p = 0.04); and interjection count per 100 words (mean difference 0.33, 95% CI 0.07-0.59, p = 0.01). Three speech measures deteriorated over time in carriers only: particle count per 100 words per month (ß = -0.02, 95% CI -0.03 to -0.004, p = 0.009); total narrative production time in seconds per month (ß = -0.24, 95% CI -0.37 to -0.12, p < 0.001); and total number of words per month (ß = -0.48, 95% CI -0.78 to -0.19, p = 0.002) including in 3 carriers who later converted to symptomatic disease. DISCUSSION: Using automatic processing pipelines, we identified early changes in the natural speech of FTD pathogenic variant carriers in the presymptomatic stage. These findings highlight the potential utility of natural speech as a digital clinical outcome assessment tool in FTD, where objective and quantifiable measures for abnormal behavior and language are lacking.

Digital markers of motor speech impairments in spontaneous speech of patients with ALS-FTD spectrum disorders.

OBJECTIVE: To evaluate automated digital speech measures, derived from spontaneous speech (picture descriptions), in assessing bulbar motor impairments in patients with ALS-FTD spectrum disorders (ALS-FTSD). METHODS: Automated vowel algorithms were employed to extract two vowel acoustic measures: vowel space area (VSA), and mean second formant slope (F2 slope). Vowel measures were compared between ALS with and without clinical bulbar symptoms (ALS + bulbar (n = 49, ALSFRS-r bulbar subscore: x¯ = 9.8 (SD = 1.7)) vs. ALS-nonbulbar (n = 23), behavioral variant frontotemporal dementia (bvFTD, n = 25) without a motor syndrome, and healthy controls (HC, n = 32). Correlations with bulbar motor clinical scales, perceived listener effort, and MRI cortical thickness of the orobuccal primary motor cortex (oral PMC) were examined. We compared vowel measures to speaking rate, a conventional metric for assessing bulbar dysfunction. RESULTS: ALS + bulbar had significantly reduced VSA and F2 slope than ALS-nonbulbar (|d|=0.94 and |d|=1.04, respectively), bvFTD (|d|=0.89 and |d|=1.47), and HC (|d|=0.73 and |d|=0.99). These reductions correlated with worse bulbar clinical scores (VSA: R = 0.33, p = 0.043; F2 slope: R = 0.38, p = 0.011), greater listener effort (VSA: R=-0.43, p = 0.041; F2 slope: p > 0.05), and cortical thinning in oral PMC (F2 slope: ß = 0.0026, p = 0.017). Vowel measures demonstrated greater sensitivity and specificity for bulbar impairment than speaking rate, while showing independence from cognitive and respiratory impairments. CONCLUSION: Automatic vowel measures are easily derived from a brief spontaneous speech sample, are sensitive to mild-moderate stage of bulbar disease in ALS-FTSD, and may present better sensitivity to bulbar impairment compared to traditional assessments such as speaking rate.

Characterizing and detecting delirium with clinical and computational measures of speech and language disturbance.

BACKGROUND: Delirium is a critically underdiagnosed syndrome of altered mental status affecting more than 50% of older adults admitted to hospital. Few studies have incorporated speech and language disturbance in delirium detection. We sought to describe speech and language disturbances in delirium, and provide a proof of concept for detecting delirium using computational speech and language features. METHODS: Participants underwent delirium assessment and completed language tasks. Speech and language disturbances were rated using standardized clinical scales. Recordings and transcripts were processed using an automated pipeline to extract acoustic and textual features. We used binomial, elastic net, machine learning models to predict delirium status. RESULTS: We included 33 older adults admitted to hospital, of whom 10 met criteria for delirium. The group with delirium scored higher on total language disturbances and incoherence, and lower on category fluency. Both groups scored lower on category fluency than the normative population. Cognitive dysfunction as a continuous measure was correlated with higher total language disturbance, incoherence, loss of goal and lower category fluency. Including computational language features in the model predicting delirium status increased accuracy to 78%. LIMITATIONS: This was a proof-of-concept study with limited sample size, without a set-aside cross-validation sample. Subsequent studies are needed before establishing a generalizable model for detecting delirium. CONCLUSION: Language impairments were elevated among patients with delirium and may also be used to identify subthreshold cognitive disturbances. Computational speech and language features are promising as accurate, noninvasive and efficient biomarkers of delirium.

Latent Factors of Language Disturbance and Relationships to Quantitative Speech Features.

BACKGROUND AND HYPOTHESIS: Quantitative acoustic and textual measures derived from speech ("speech features") may provide valuable biomarkers for psychiatric disorders, particularly schizophrenia spectrum disorders (SSD). We sought to identify cross-diagnostic latent factors for speech disturbance with relevance for SSD and computational modeling. STUDY DESIGN: Clinical ratings for speech disturbance were generated across 14 items for a cross-diagnostic sample (N = 334), including SSD (n = 90). Speech features were quantified using an automated pipeline for brief recorded samples of free speech. Factor models for the clinical ratings were generated using exploratory factor analysis, then tested with confirmatory factor analysis in the cross-diagnostic and SSD groups. The relationships between factor scores and computational speech features were examined for 202 of the participants. STUDY RESULTS: We found a 3-factor model with a good fit in the cross-diagnostic group and an acceptable fit for the SSD subsample. The model identifies an impaired expressivity factor and 2 interrelated disorganized factors for inefficient and incoherent speech. Incoherent speech was specific to psychosis groups, while inefficient speech and impaired expressivity showed intermediate effects in people with nonpsychotic disorders. Each of the 3 factors had significant and distinct relationships with speech features, which differed for the cross-diagnostic vs SSD groups. CONCLUSIONS: We report a cross-diagnostic 3-factor model for speech disturbance which is supported by good statistical measures, intuitive, applicable to SSD, and relatable to linguistic theories. It provides a valuable framework for understanding speech disturbance and appropriate targets for modeling with quantitative speech features.

Who does what to whom? graph representations of action-predication in speech relate to psychopathological dimensions of psychosis.

Graphical representations of speech generate powerful computational measures related to psychosis. Previous studies have mostly relied on structural relations between words as the basis of graph formation, i.e., connecting each word to the next in a sequence of words. Here, we introduced a method of graph formation grounded in semantic relationships by identifying elements that act upon each other (action relation) and the contents of those actions (predication relation). Speech from picture descriptions and open-ended narrative tasks were collected from a cross-diagnostic group of healthy volunteers and people with psychotic or non-psychotic disorders. Recordings were transcribed and underwent automated language processing, including semantic role labeling to identify action and predication relations. Structural and semantic graph features were computed using static and dynamic (moving-window) techniques. Compared to structural graphs, semantic graphs were more strongly correlated with dimensional psychosis symptoms. Dynamic features also outperformed static features, and samples from picture descriptions yielded larger effect sizes than narrative responses for psychosis diagnoses and symptom dimensions. Overall, semantic graphs captured unique and clinically meaningful information about psychosis and related symptom dimensions. These features, particularly when derived from semi-structured tasks using dynamic measurement, are meaningful additions to the repertoire of computational linguistic methods in psychiatry.

Natural language processing methods are sensitive to sub-clinical linguistic differences in schizophrenia spectrum disorders.

Computerized natural language processing (NLP) allows for objective and sensitive detection of speech disturbance, a hallmark of schizophrenia spectrum disorders (SSD). We explored several methods for characterizing speech changes in SSD (n = 20) compared to healthy control (HC) participants (n = 11) and approached linguistic phenotyping on three levels: individual words, parts-of-speech (POS), and sentence-level coherence. NLP features were compared with a clinical gold standard, the Scale for the Assessment of Thought, Language and Communication (TLC). We utilized Bidirectional Encoder Representations from Transformers (BERT), a state-of-the-art embedding algorithm incorporating bidirectional context. Through the POS approach, we found that SSD used more pronouns but fewer adverbs, adjectives, and determiners (e.g., "the," "a,"). Analysis of individual word usage was notable for more frequent use of first-person singular pronouns among individuals with SSD and first-person plural pronouns among HC. There was a striking increase in incomplete words among SSD. Sentence-level analysis using BERT reflected increased tangentiality among SSD with greater sentence embedding distances. The SSD sample had low speech disturbance on average and there was no difference in group means for TLC scores. However, NLP measures of language disturbance appear to be sensitive to these subclinical differences and showed greater ability to discriminate between HC and SSD than a model based on clinical ratings alone. These intriguing exploratory results from a small sample prompt further inquiry into NLP methods for characterizing language disturbance in SSD and suggest that NLP measures may yield clinically relevant and informative biomarkers.

Linguistic camouflage in girls with autism spectrum disorder.

BACKGROUND: Autism spectrum disorder (ASD) is diagnosed more frequently in boys than girls, even when girls are equally symptomatic. Cutting-edge behavioral imaging has detected "camouflaging" in girls with ASD, wherein social behaviors appear superficially typical, complicating diagnosis. The present study explores a new kind of camouflage based on language differences. Pauses during conversation can be filled with words like UM or UH, but research suggests that these two words are pragmatically distinct (e.g., UM is used to signal longer pauses, and may correlate with greater social communicative sophistication than UH). Large-scale research suggests that women and younger people produce higher rates of UM during conversational pauses than do men and older people, who produce relatively more UH. Although it has been argued that children and adolescents with ASD use UM less often than typical peers, prior research has not included sufficient numbers of girls to examine whether sex explains this effect. Here, we explore UM vs. UH in school-aged boys and girls with ASD, and ask whether filled pauses relate to dimensional measures of autism symptom severity. METHODS: Sixty-five verbal school-aged participants with ASD (49 boys, 16 girls, IQ estimates in the average range) participated, along with a small comparison group of typically developing children (8 boys, 9 girls). Speech samples from the Autism Diagnostic Observation Schedule were orthographically transcribed and time-aligned, with filled pauses marked. Parents completed the Social Communication Questionnaire and the Vineland Adaptive Behavior Scales. RESULTS: Girls used UH less often than boys across both diagnostic groups. UH suppression resulted in higher UM ratios for girls than boys, and overall filled pause rates were higher for typical children than for children with ASD. Higher UM ratios correlated with better socialization in boys with ASD, but this effect was driven by increased use of UH by boys with greater symptoms. CONCLUSIONS: Pragmatic language markers distinguish girls and boys with ASD, mirroring sex differences in the general population. One implication of this finding is that typical-sounding disfluency patterns (i.e., reduced relative UH production leading to higher UM ratios) may normalize the way girls with ASD sound relative to other children, serving as "linguistic camouflage" for a naïve listener and distinguishing them from boys with ASD. This first-of-its-kind study highlights the importance of continued commitment to understanding how sex and gender change the way that ASD manifests, and illustrates the potential of natural language to contribute to objective "behavioral imaging" diagnostics for ASD.

Automated recognition of malignancy mentions in biomedical literature.

BACKGROUND: The rapid proliferation of biomedical text makes it increasingly difficult for researchers to identify, synthesize, and utilize developed knowledge in their fields of interest. Automated information extraction procedures can assist in the acquisition and management of this knowledge. Previous efforts in biomedical text mining have focused primarily upon named entity recognition of well-defined molecular objects such as genes, but less work has been performed to identify disease-related objects and concepts. Furthermore, promise has been tempered by an inability to efficiently scale approaches in ways that minimize manual efforts and still perform with high accuracy. Here, we have applied a machine-learning approach previously successful for identifying molecular entities to a disease concept to determine if the underlying probabilistic model effectively generalizes to unrelated concepts with minimal manual intervention for model retraining. RESULTS: We developed a named entity recognizer (MTag), an entity tagger for recognizing clinical descriptions of malignancy presented in text. The application uses the machine-learning technique Conditional Random Fields with additional domain-specific features. MTag was tested with 1,010 training and 432 evaluation documents pertaining to cancer genomics. Overall, our experiments resulted in 0.85 precision, 0.83 recall, and 0.84 F-measure on the evaluation set. Compared with a baseline system using string matching of text with a neoplasm term list, MTag performed with a much higher recall rate (92.1% vs. 42.1% recall) and demonstrated the ability to learn new patterns. Application of MTag to all MEDLINE abstracts yielded the identification of 580,002 unique and 9,153,340 overall mentions of malignancy. Significantly, addition of an extensive lexicon of malignancy mentions as a feature set for extraction had minimal impact in performance. CONCLUSION: Together, these results suggest that the identification of disparate biomedical entity classes in free text may be achievable with high accuracy and only moderate additional effort for each new application domain.