CoChat
Sign Up
Cited by 96Voice and Speech DisordersDementia and Cognitive Impairment ResearchEmotion and Mood Recognition

A Review of Automated Speech and Language Features for Assessment of Cognitive and Thought Disorders

Open access
Voleti, Rohit, Liss, Julie M., Berisha, Visar·2020·IEEE Journal of Selected Topics in Signal Processing
DOI 10.1109/jstsp.2019.2952087PMC PMC8074691
Open this paper in CoChat — free account

Abstract

It is widely accepted that information derived from analyzing speech (the acoustic signal) and language production (words and sentences) serves as a useful window into the health of an individual's cognitive ability. In fact, most neuropsychological testing batteries have a component related to speech and language where clinicians elicit speech from patients for subjective evaluation across a broad set of dimensions. With advances in speech signal processing and natural language processing, there has been recent interest in developing tools to detect more subtle changes in cognitive-linguistic function. This work relies on extracting a set of features from recorded and transcribed speech for objective assessments of speech and language, early diagnosis of neurological disease, and tracking of disease after diagnosis. With an emphasis on cognitive and thought disorders, in this paper we provide a review of existing speech and language features used in this domain, discuss their clinical application, and highlight their advantages and disadvantages. Broadly speaking, the review is split into two categories: language features based on natural language processing and speech features based on speech signal processing. Within each category, we consider features that aim to measure complementary dimensions of cognitive-linguistics, including language diversity, syntactic complexity, semantic coherence, and timing. We conclude the review with a proposal of new research directions to further advance the field.

Verified 2026-10-05 · 1 reference not checked yet

AI snapshot

Written from the paper’s full text.

TL;DR

This review surveys automated speech and language features used to assess cognitive and thought disorders, dividing them into natural language processing measures and speech signal processing measures. It argues that these features can complement subjective clinical assessment, while highlighting barriers such as small datasets, ASR errors, limited interpretability, and lack of standardization.

Key findings

  • The review organizes automated markers into NLP features—lexical diversity/density, syntactic complexity, semantic coherence—and speech signal features—prosodic timing, articulatory/spectral measures, vocal quality, and ASR-derived features.
  • Cited studies report strong discrimination between clinical and control groups: over 90% accuracy for semantic dementia classification, over 80% for MCI using LIWC categories, AUCs of 0.732, 0.703, and 0.861 for MCI when using language, language plus speech, and language plus speech plus test scores, respectively; speech-graph features classified schizophrenia, mania, and controls with over 90% accuracy.
  • Acoustic measures often add complementary information to textual measures; for example, AD versus control classification reached about 87%, MCI versus control about 80%, and MCI versus AD about 80% using temporal prosodic features, while ASR-based MCI detection reached 78% versus 82% for manual methods.
  • ASR errors reduce the usefulness of syntactic and semantic textual features, though some features are robust; one study found dementia versus functional memory disorder classification dropped from over 90% with manual transcripts to 79% with ASR.

Methods

  • Design: narrative review of existing literature on automated speech and language features for cognitive and thought disorders, with no new primary data collection.
  • Scope: covers neurodegenerative disease and mental illness examples such as MCI, dementia, Alzheimer’s disease, chronic traumatic encephalopathy, schizophrenia, schizoaffective disorder, bipolar disorder, and primary progressive aphasia.
  • Analysis: synthesizes methods and clinical applications across NLP and speech signal processing, discusses advantages and disadvantages, and proposes future research directions rather than performing a meta-analysis.

Limitations

  • Stated by the authors: clinical speech data are scarce, hard to share, require expert labeling, and raise privacy concerns; speech elicitation tasks are not standardized; published results may be biased by small datasets and limited tasks; high-dimensional features such as MFCCs are difficult for clinicians to interpret; sensitivity to noisy conditions, ASR errors, and speaker variability is not well characterized.
  • Evident from the text: the review is narrative rather than systematic or meta-analytic, and many cited studies use small samples or single datasets, limiting direct comparison across findings.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1109/jstsp.2019.2952087

Cites

116 of 117 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

0 of 120 cited sources checked · 0 supported · 0 flagged

  • Premisecitation

    TTR, BI, verb-rate, and adjective-rate indicated strong group differences between AD patients and healthy controls.

    “In particular, TTR, BI, verb-rate, and adjective-rate all indicated strong group differences between the participants with AD and healthy controls”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [10]

      Not checked in the free run.

  • Premisecitation

    Snowdon et al. found that low idea density in early life was a particularly strong predictor of reduced cognitive ability or AD in later life.

    “In particular, the study found that low idea density in early life was a particularly strong predictor of reduced cognitive ability or the presence of AD in later life.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [66] source

      Not checked in the free run.

  • Premisecitation

    In Snowdon et al., about 80% of participants whose writings lacked linguistic complexity developed AD or had mental and cognitive disabilities in older age.

    “Roughly 80% of the participants that were determined to lack linguistic complexity in their writings developed AD or had mental and cognitive disabilities in their older age.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [66] source

      Not checked in the free run.

  • Premisecitation

    Asgari et al. achieved over 80% classification accuracy for MCI versus healthy controls using LIWC categories.

    “They were able to achieve over 80% classification accuracy by selecting LIWC categories that best represented the difference in the MCI and healthy control datasets.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [7] source

      Not checked in the free run.

  • Premisecitation

    Roark et al. found that content density was a strong indicator of group differences between healthy controls and MCI patients.

    “In particular, the content density was a strong indicator of group differences between healthy controls and patients with MCI.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [14]

      Not checked in the free run.

  • May combine two claims. 32
  • The quote only partly states this claim. 1
  • Classified as probably not a substantial claim. 30

  • May combine two claims.

    Premisecitation

    Diagnosis for many common psychiatric conditions has become more consistent over time, but schizoaffective disorder and some others are often evaluated inconsistently because test batteries are subjective.

    “While the diagnosis for many common psychiatric conditions has become more consistent over time as they are better understood, others (e.g. schizoaffective disorder) are often evaluated inconsistently by different clinical assessors due to the subjective nature of the test batteries applied [63], [64].”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites [63]

      Not checked in the free run.

    • cites [64]

      Not checked in the free run.

  • The quote only partly states this claim.

    Premisecitation

    Bucks et al. discriminated healthy older controls (n=16) from AD patients (n=8) using TTR, BI, and HS, with cross-validation accuracy of 87.5%.

    “In [10], the authors successfully discriminated between a small sample of healthy older control participants (n = 16) and patients with AD (n = 8) using TTR (Equation (1)), BI (Equation (2)), and HS (Equation (3)) as measures of lexical diversity or vocabulary richness.”
    • Also stated in: § First Order Lexeme-Level Analysis · hedged
      “the groups could be classified with a cross-validation accuracy of 87.5%.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [10]

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    Ratings for narrative, recall, conversational, or other spoken language tasks are often subjective and of variable reliability, which makes diagnosis more challenging.

    “However, ratings for narrative, recall, conversational, or other spoken language tasks are often subjective in nature and of variable reliability, making the underlying diagnosis more challenging [62].”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [62] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    CAINS, BNSS, and SSPA are screenings used to evaluate schizophrenia and bipolar disorder.

    “Other forms of disorders, i.e. schizophrenia and bipolar disorder, can be evaluated with screenings such as the Clinical Assessment Interview for Negative Symptoms (CAINS) [60], Brief Negative Symptom Scale (BNSS) [61], the Social Skills Performance Assessment (SSPA) [53], and several others that measure the effects of thought and mood disorders.”
    Quote in the paperNot a substantial claim0 of 3 sources checked
    • cites [60] source

      Not checked in the free run.

    • cites [61]

      Not checked in the free run.

    • cites [53] source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    Garrard et al. achieved over 90% classification accuracy for distinguishing semantic dementia from healthy controls and for classifying left- versus right-predominant SD.

    “Classification accuracy of over 90% was reached for categorizing the participants for two tasks: (1) participants with SD against the healthy control participants, and (2) classifying the left- and right-predominant variants of SD.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [4] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    The mental status examination is a commonly used multi-faceted screening tool for signs of neurological and psychiatric disorders at a given time.

    “The mental status examination (MSE) is a commonly utilized and multi-faceted tool for screening an individual at a given point in time for signs of neurological and psychiatric disorders [56].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [56] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    The MMSE, ACE, and MoCA are related screenings for conditions such as MCI, dementia, and Alzheimer's disease.

    “Related screenings include the mini-mental state examination (MMSE) [57], Addenbrooke’s Cognitive Examination (ACE) [58], and the Montreal Cognitive Assessment (MoCA) [59] for evaluating conditions like mild cognitive impairment (MCI), dementia, and Alzheimer’s disease (AD).”
    Quote in the paperNot a substantial claim0 of 3 sources checked
    • cites [57] source

      Not checked in the free run.

    • cites [58] source

      Not checked in the free run.

    • cites [59] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation

    The DSM-5 supplies standard diagnostic criteria for psychiatric and neurocognitive disorders and is revised as knowledge advances.

    “The DSM-5 [52], published by the American Psychiatric Association (APA), provides the standard diagnostic criteria for psychiatric and neurocognitive disorders, and it is updated as knowledge in the field evolves.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [52]

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    Berisha et al. found that Reagan's number of unique words steadily declined over his presidency, while Bush showed no such changes.

    “The study shows that the number of unique words used by Reagan over the period of his presidency steadily declined over time, while no such changes were seen for Bush.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [11]

      Not checked in the free run.

  • May combine two claims.

    Interpretationnone given

    LSA, word2vec, and GloVe are simple to train and effective, but they do not handle out-of-vocabulary words or distinguish words with multiple unrelated meanings.

    “For example, LSA, word2vec and GloVe are simple to train and effective, but a major disadvantage is that they do not handle out-of-vocabulary (OOV) words or consider words with multiple unrelated meanings.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Neural-network word embedding methods such as word2vec and GloVe have shown better performance than LSA for semantic modeling when enough training data is available.

    “In recent years, several new word embedding methods based on neural networks have gained popularity, such as word2vec [26] or GloVe [27], which have shown improved performance over LSA for semantic modeling when sufficient training data is available [84].”
    Quote in the paperNot a substantial claim0 of 3 sources checked
    • cites [26] source

      Not checked in the free run.

    • cites [27] source

      Not checked in the free run.

    • cites [84] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S4 (based). May combine two claims.

    Interpretationnone given

    With NLP, complexity and coherence of language can be measured; with speech signal processing, acoustic proxies related to cognitive processing can be measured.

    “With NLP, we can measure the complexity and coherence of language and with speech signal processing, we can measure acoustic proxies related to cognitive processing.”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: Speech and language analysis provides a useful window into cognitive health.
    • Rests on an assumption: Speech and language analysis can reveal underlying neural function regardless of the specific disease or condition.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Spoken language production is a complex multi-stage process involving memory, cognition, and sensorimotor function.

    “The production of spoken language in humans is a complex, multi-stage process that involves high levels of memory, cognition, and sensorimotor function.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [55]

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    In NFL players, longitudinal declines in TTR and CD were observed, while coaches and executives showed expected increases.

    “However, this study demonstrated clear longitudinal declines in both variables for the NFL players while showing the expected increase in both variables for coaches and executives in similar contexts.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [15]

      Not checked in the free run.

  • May combine two claims.

    Interpretationcitation

    Reagan's declines predated his AD diagnosis in 1994 by 6 years, suggesting lexical features may be useful in predicting AD onset pre-clinically.

    “These declines predated his diagnosis of AD in 1994 by 6 years, suggesting that these computed lexical features may be useful in predicting the onset of AD pre-clinically.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [11]

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation

    Common symptoms of formal thought disorder include poverty of speech, derailment of speech, and semantically incoherent speech.

    “Some common symptoms include poverty of speech (alogia), derailment of speech, and semantically incoherent speech (word salad) [90], [91].”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [90] source

      Not checked in the free run.

    • cites [91] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Interpretationnone given

    Adopting endpoints more sensitive to change means studies can be powered with fewer participants.

    “In addition, adopting endpoints that are more sensitive to change means that these studies can be powered with fewer participants.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Deep neural network methods such as RNN/LSTM and transformer architectures use contextual information to generate embeddings for out-of-vocabulary words.

    “Some methods based on deep neural networks (DNNs), such as recurrent neural network (RNN) / long-short term memory (LSTM) networks (e.g. ELMo [85]) or transformer architectures (e.g. BERT [86]) utilize contextual information to generate embeddings for OOV words.”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [85] source

      Not checked in the free run.

    • cites [86]

      Not checked in the free run.

  • May combine two claims.

    Study design and conductcitation

    Bedi et al.'s initial study had only 34 participants total, including 5 CHR+ participants, and was intended as a proof-of-principle exploration.

    “This initial study only had 34 participants total (only 5 CHR+ participants) and was intended as a proof-of-principle exploration.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [16] source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    Yancheva and Rudzicz found that participants with Alzheimer's disease and healthy controls tended to discuss the same topics, but those with Alzheimer's disease had wider topic clusters with more irrelevant directions.

    “The study found that participants with AD and healthy controls were likely to discuss the same topics, but those with AD had wider topic clusters with more irrelevant directions.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [32] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Word embeddings are based on the distributional hypothesis that a word's meaning comes from its contextual usage.

    “Word embeddings are motivated by the distributional hypothesis in linguistics, a concept proposed by English linguist John R. Firth who famously stated “You shall know a word by the company it keeps” [83], i.e. that the inherent meaning of words is derived from their contextual usage in natural language.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [83] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Many easy-to-compute structural metrics include mean length of clause, mean length of sentence, ratio of clauses to sentences, and related statistics.

    “Many easy-to-compute and common structural metrics of language include the mean length of a clause, mean length of sentence, ratio of number of clauses to number of sentences, and other related statistics [5].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [5]

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Interpretationreasoning

    Sentence-level syntactical analysis is prone to increased complexity because there are many implementation methodologies.

    “However, while offering the ability to analyze more complex sentence structures, sentence-level syntactical analysis is also prone to increased complexity due to large range of implementation methodologies.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation

    Mota et al. visualized and quantified logorrhea in mania as denser speech-graph networks and alogia in schizophrenia as greater nodes per word and average total degree per node.

    “Using these features, the researchers were able to visualize and quantify concepts such as the logorrhea (excessive wordiness and incoherence) associated with mania, evidenced by denser networks. Similarly, the alogia (poverty of speech) typical of schizophrenia was also visible in the generated speech-graph networks, as evidenced by a greater number of nodes per word and average total degree per”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites [21] source

      Not checked in the free run.

    • cites [22] source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    Mota et al. classified controls, participants with schizophrenia, and participants with mania with over 90% accuracy, significantly better than PANSS and BPRS.

    “Control participants, participants with schizophrenia, and participants with mania were classified with over 90% accuracy, significantly improving over traditional clinical measures, such as the Positive and Negative Syndrome Scale (PANSS) and Brief Psychiatric Rating Scale (BPRS) [21].”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [21] source

      Not checked in the free run.

  • Classified as probably not a substantial claim.

    Premisecitation

    Structural features of speech graphs, such as loop density and distance between words of interest, are clinically relevant objective language measures that provide insight into cognitive function.

    “The researchers in this work suggest that structural graph features, i.e. loop density, distance (number of nodes) between words of interest, etc.) serve as clinically relevant objective language measures that give insight into cognitive function.”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites [21] source

      Not checked in the free run.

    • cites [22] source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    Formal thought disorder is most commonly associated with schizophrenia but is often present in other mental illnesses such as mania and depression.

    “FTD is most commonly associated with schizophrenia but is often present in other forms of mental illness such as mania, depression, and several others [89], [90].”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [89] source

      Not checked in the free run.

    • cites [90] source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    Hernández-Domínguez et al. found that generated features were significantly correlated with MMSE severity and achieved an AUC of approximately 0.76 for classifying Alzheimer's disease, mild cognitive impairment, and healthy controls.

    “The resulting features were significantly correlated with severity as assessed by the MMSE, and classification performance was characterized by the receiver operating characteristic (ROC) area under curve (AUC) performance of AUC ≈ 0.76 for all three groups.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [33] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Spoken language production has three distinct stages: conceptualization, formulation, and articulation.

    “There are three distinct stages [55]:”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [55]

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    Manual transcriptions classified neurodegenerative dementia versus functional memory disorder with over 90% accuracy, but accuracy dropped to 79% when ASR was used.

    “With manual transcriptions, the classification accuracy was over 90% in classifying the two groups, but it dropped to 79% when ASR was used.”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites [46]

      Not checked in the free run.

    • cites [47] source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    Automated speech analysis produces results very similar to manual computation of the same metrics, showing potential for automated speech signal processing in MCI detection.

    “They conclude that automated speech analysis produces very similar results to manually computing these metrics from the speech samples, demonstrating the potential of automated speech signal processing for detecting MCI.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [14]

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    Adding acoustic features had the greatest positive impact for differentiating healthy controls from PPA subtypes, but was less useful for distinguishing among PPA subtypes.

    “However, the addition of acoustic features had the greatest positive impact when attempting to differentiate between healthy control participants and those with one of the PPA subtypes, but proved less useful in distinguishing the subtypes.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [42] source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    Many selected features were robust to transcription quality and may even exploit poor ASR performance to identify participants with dementia.

    “Additionally, many of the features they selected showed robustness to transcription quality, possibly even taking advantage of the poor ASR performance to identify participants with dementia.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [48] source

      Not checked in the free run.

  • May combine two claims.

    Interpretationreasoning

    Many acoustic features are not used in clinical diagnosis because they are difficult to interpret directly, leaving clinicians unable to understand a model's decisions or whether it considers clinically relevant dimensions.

    “Many of these features are not currently used in clinical diagnosis despite their powerful predictive power because they are difficult to directly interpret (e.g. MFCCs); this means that clinicians can see the output of a complicated model but not understand why the model came to that decision or if it is considering clinically-relevant dimensions.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationcitation

    Sentence-level syntactical analysis provides a multi-dimensional representation of cognitive-linguistics and allows better characterization of different clinical conditions, as Mota et al. did with schizophrenia and bipolar disorder/mania.

    “This provides a multi-dimensional representation of cognitive-linguistics and allows for better characterization of different clinical conditions, as Mota et al. did with patients with schizophrenia and those with bipolar disorder/mania [21].”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [21] source

      Not checked in the free run.

  • Classified as probably not a substantial claim.

    Interpretationreasoning

    The acoustic speech stream is a near-real-time physical manifestation of the cognitive-linguistic processing that creates a message.

    “This is because the acoustic stream is the physical manifestation of the cognitive-linguistic processing that has gone into creating the message being conveyed, in near real-time.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S4 (based).

    Premisenone given

    Acoustic speech signals are highly non-stationary.

    “Time-frequency signal processing techniques are also commonly used since acoustic speech signals are highly non-stationary.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationnone given

    Jitter, shimmer, and harmonic-to-noise ratio can provide insight for diagnosing and characterizing certain clinical conditions, though they are difficult to compute and interpret consistently.

    “These features alone are often difficult to consistently compute and interpret, but can provide insight for the diagnosis and characterization of certain clinical conditions.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Premisecitation

    Intra-subject analysis showed good specificity in classifying bipolar subjects and healthy controls across all F0 contour features, but the directions of most features were inconsistent across subjects.

    “They noted that intra-subject analysis showed good specificity in classifying bipolar subjects and healthy controls across all contour features, but that directions of most were not consistent across different subjects.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [41] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Interpretationnone given

    Decreases in the rhythmic and melodic variation of speech may indicate mood changes.

    “Conversely, reductions in the rhythmic and melodic variations in speech may be indicative of changes in mood.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation

    Participants with AD show higher proportions of voiceless speech and lower proportions of voiced speech, indicating reduced fluency and shorter fluent segments.

    “First, the spontaneous speech analysis indicated that participants with AD exhibited higher proportions of voiceless speech and lower proportions of voiced speech, indicating a loss of fluency and shortening of fluent speech segments for those with AD.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [38] source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    Acoustic features separated AD from controls at about 87% accuracy, MCI from controls at about 80%, and MCI from AD at about 80%.

    “These quantifiable alterations of speech in individuals with MCI and AD allowed the researchers to successfully separate patients with AD from healthy controls (approx. 87% accuracy), patients with MCI from controls (approx. 80% accuracy), and patients with MCI from patients with AD (approx. 80% accuracy).”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [35]

      Not checked in the free run.

  • Classified as probably not a substantial claim.

    Premisecitation

    An automated method estimating the F0 contour over time with a moving window can serve as a proxy for mood changes in bipolar disorder.

    “In this study, the authors propose an automated method for estimating the contour of F0 over time with a moving window approach as a proxy for mood changes.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [41] source

      Not checked in the free run.

  • Classified as probably not a substantial claim.

    Premisecitation

    Iter et al. identified that repetitive sentences and phrases would be scored as highly coherent even though repetition of ideas is common in formal thought disorder and should be scored negatively.

    “The third reason they identified (but did not address) is that repetitive sentences and phrases would be scored as highly coherent, even though repetition of ideas is common in FTD and should be scored negatively.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [18] source

      Not checked in the free run.

  • May combine two claims.

    Interpretationreasoning

    Pauses in speech can be linked to word-finding difficulty or to additional time needed to formulate a message.

    “In this way, pauses during speech can be associated with difficulty in lexical retrieval (word-finding difficulties) or with extra processing time needed for message formulation.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Premisecitation

    Diagnostic accuracy decreases as WER increases, but the correlation is relatively weak when using features that are more robust to ASR errors, such as word frequency and word length features.

    “Additionally, they were able to confirm that even with their model, diagnostic accuracy decreases with increasing WER, as expected, but the correlation between the two is relatively weak when selecting certain features that are more robust to ASR errors (such as word frequency and word length related features)15.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [45] source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    ASR accuracy is reduced for elderly patients and for patients with dementia.

    “Previous work has shown that ASR accuracy is reduced for both elderly patients and those with dementia [99], [100], [101].”
    Quote in the paperSubstantial claim0 of 3 sources checked
    • cites [99] source

      Not checked in the free run.

    • cites [100] source

      Not checked in the free run.

    • cites [101]

      Not checked in the free run.

  • May combine two claims.

    Interpretationnone given

    Clinical speech data is harder to collect and share than speech recognition data because it is scarcer, requires expert labeling, and raises privacy issues.

    “Unlike speech recognition research, data for clinical speech research is much more difficult to collect and share because it’s scarcer, requires experts to label, and there are privacy issues.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationreasoning

    Understanding variability can inform stratified sampling schemes that match healthy and clinical cohorts on relevant speech and language production aspects.

    “Furthermore, this understanding can inform stratified sampling schemes that allow experimenters to match healthy and clinical cohorts on relevant aspects of speech/language production.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S4 (based).

    Interpretationnone given

    The sensitivity of the described features and the models they drive is not well understood under noisy conditions.

    “The sensitivity of the features we describe herein and the follow-on models they drive are not well understood under noisy conditions.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Interpretationnone given

    The individual sources of variability have not yet been fully characterized quantitatively for the features described in this paper.

    “These individual sources of variability have yet to be fully characterized quantitatively for the features that we described in this paper.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Interpretationreasoning

    The area of interpretable AI models has received some recent attention and will become more important as AI systems are deployed in healthcare.

    “This area has received some attention recently and will continue to become more important as AI systems are deployed in healthcare.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable). May combine two claims.

    Interpretationreasoning

    Many of the described features are readily interpretable and, based on existing literature, likely have clinical utility.

    “Many of the features described herein are readily interpretable and, given the existing literature, it is reasonable to posit that they have clinical utility.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable), S4 (based).

    Interpretationreasoning

    Joint exploration of the elicitation-analytics space has the potential to improve sensitivity in detecting cognitive-linguistic change.

    “We posit that joint exploration of the elicitation-analytics space has the potential to result in improved sensitivity in detecting cognitive-linguistic change.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S4 (based).

    Interpretationnone given

    Improved performance from application-specific features likely comes at the cost of reduced feature interpretability.

    “This improved performance likely comes at the expense of reduced feature interpretability.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationreasoning

    A more complete understanding of variability in healthy populations helps interpret changes seen in clinical populations.

    “A more complete understanding of this variability in healthy populations helps to interpret changes observed in clinical populations.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable), S4 (based).

    Interpretationreasoning

    Better characterizing the sensitivity of nuisance features can inform the development of representations robust to various noise sources.

    “A better characterization of the sensitivity of these nuisance features can inform the development of new representations that are robust to various sources of noise.”
    Quote in the paperNot a substantial claim
57 of 57

  • The paper relies on this without support.

    Speech and language analysis provides a useful window into cognitive health.

    “It is widely accepted that information derived from analyzing speech (the acoustic signal) and language production (words and sentences) serves as a useful window into the health of an individual’s cognitive ability.”

  • The paper relies on this without support.

    Many aspects of cognitive and thought disorders are evident in how speech is produced and what is said.

    “Many aspects of cognitive and thought disorders are manifest in the way speech is produced and what is said.”

  • The paper relies on this without support.

    Speech and language analysis can reveal underlying neural function regardless of the specific disease or condition.

    “Irrespective of the underlying disease or condition, the analysis of speech and language can provide insight to the underlying neural function.”

  • The paper relies on this without support.

    All neuropsychological batteries for cognitive health include a significant speech and language assessment because cognitive-linguistic function is a strong biomarker of neuropsychological health.

    “All of these neuropsychological batteries for evaluating cognitive health have a significant speech and language assessment, as cognitive-linguistic function is a strong biomarker for neuropsychological health in many dimensions.”

  • The paper relies on this without support.

    Speech and language measures themselves are agnostic to the underlying disorder; it is the pattern of deficits that associates with different etiologies.

    “To reiterate, the speech and language measures are, themselves, agnostic to the underlying disorder. Rather, it is the constellation of deficit patterns that associate with different etiologies.”

  • The paper relies on this without support.

    Natural spoken language contains measurable dimensions that indicate aspects of cognitive health.

    “Natural spoken language contains several measurable dimensions that indicate various aspects of cognitive health.”

  • The paper relies on this without support.

    Sentence complexity provides insight into cognitive-linguistic health.

    “The complexity of the sentences we structure provides a great deal of insight into cognitive-linguistic health.”

  • The paper relies on this without support.

    The ordering of words in sentences and sentences in paragraphs provides important insight into cognitive function.

    “The ordering of words in sentences and sentences in paragraphs can also provide important insight into cognitive function.”

  • The paper relies on this without support.

    Automated first-order lexical analysis can generate objective language metrics that provide valuable insight into cognitive function.

    “Automated first-order lexical analysis, i.e. at the lexeme-level or word-level, can generate objective language metrics to provide valuable insight into cognitive function.”

  • The paper relies on this without support.

    Measures of lexical density may be useful for cognitive assessment.

    “Measures of lexical density, which quantify the degree of information packaging within an utterance, may also be useful for cognitive assessment.”

  • The paper relies on this without support.

    Content words carry more information than function words.

    “Content words1 (i.e. nouns, verbs, adjectives, adverbs) tend to carry more information than function words2 (e.g. prepositions, conjunctions, interjections, etc.).”

  • The paper relies on this without support.

    TTR is negatively affected by longer utterances because unique-word diversity typically plateaus as total word count increases.

    “However, TTR is negatively impacted for longer utterances, as the diversity of unique words typically plateaus as the number of total words increase.”

  • The paper relies on this without support.

    MATTR does not have a length-based bias but is more variable because parameters are estimated on smaller samples.

    “This approach does not have a length-based bias, but is considerably more variable as the parameters are estimated on smaller speech samples.”

  • The paper relies on this without support.

    The exponential and logarithm in BI and HS reduce dependence on text length while using all samples.

    “The exponential and logarithm in the BI and the HS reduce the dependence on the text length, while still using all samples to estimate the diversity measure, unlike the MATTR.”

  • The paper relies on this without support.

    POS tagging has reliable and accurate implementations.

    “POS tagging has several reliable and accurate implementations, and these features are simple and easy to compute.”

  • The paper relies on this without support.

    A person's cognitive function includes the capacity to express organized and coherent thoughts through spoken or written language.

    “Cognitive function is also characterized by one’s ability to convey organized and coherent thoughts through spoken or written language.”

  • The authors state this as an assumption.

    In latent semantic analysis, words that appear together in a group are assumed to be semantically similar.

    “As per Firth’s hypothesis, the principal assumption of LSA is that words which occur together within a group of words will be semantically similar.”

  • The paper relies on this without support.

    Language data from people with schizophrenia is scarce, which makes research in this area difficult.

    “The limited availability of language data in schizophrenia is always a difficult challenge”

  • The paper relies on this without support.

    Left-branching sentences tend to place greater cognitive demands on speakers because they must retain more information about the subject before it is named.

    “Left-branching sentences are often cognitively more taxing as they involve more complex constructions that require a speaker to remember more information about the subject before the subject is explicitly mentioned.”

  • The paper relies on this without support.

    Building sentences places demands on the cognitive-linguistic system beyond those involved in finding words.

    “Since sentence construction further taxes the cognitive-linguistic system beyond word finding, methods that capture sentence complexity provide more insight into the neurological health of the individual producing these utterances.”

  • The paper relies on this without support.

    Formal thought disorder reduces a person's ability to produce semantically coherent language.

    “Many forms of mental illness can result in a condition known as formal thought disorder (FTD), which impairs an individual’s ability to produce semantically coherent language.”

  • The paper relies on this without support.

    Dementias such as PPA, MCI, and AD involve a general slowing of thought processes in affected people.

    “Disorders such as PPA, MCI, AD, and other forms of dementia are associated with a general slowing of thoughts in affected individuals.”

  • The paper relies on this without support.

    Speech and language data are expected to show substantial variability.

    “There is a great deal of variability to be expected in speech and language data.”

  • The paper relies on this without support.

    Published algorithms usually rely on speech and language samples that were collected earlier for other purposes.

    “Algorithms published in the literature typically make use of previously-collected speech and language samples.”
24 of 24

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

  • Reported AUC falls when speech features are added

    Minor

    The paper says combining language and speech-duration metrics improved classification, but its reported AUC drops from 0.732 for language features alone to 0.703 when speech features are added. The improvement appears only after test scores are added as well.

1 of 1

Cites

Flagged first, then by citations

Research works printed but not linked: 5

  1. wechsler, 1997, Wechsler Memory ScaleThird Edition Manual
  2. pies, 2007, How “objective” are psychiatric diagnoses, Psychiatry, 4, 18
  3. brunét, 1978, Le vocabulaire de Jean Giraudoux Structure et évolution
  4. honoré, 1979, Some simple measures of richness of vocabulary, Bull Assoc Lit Linguist Comput, 7, 172
  5. colman, 2015, A Dictionary of Psychology

Cited by

Sorted by citations

Related

Authors
Voleti, R., Liss, J. M., & Berisha, V.
Affiliations
Arizona State University (US)
ORCID
Venue
IEEE Journal of Selected Topics in Signal Processing
Publisher
Institute of Electrical and Electronics Engineers (IEEE)
Published
2020-02
Volume
14
Issue
2
Pages
282-298
ISSN
  • 1932-4553 (ISSN-L)
  • 1941-0484
DOI
10.1109/jstsp.2019.2952087
PMC
PMC8074691
OpenAlex
W2984124903
Semantic Scholar
81ef57662f3a9410f8e7008005758c719a0293e7
Cited by
96
References
117
Citations by year
2026: 13 · 2025: 19 · 2024: 11 · 2023: 29 · 2022: 13 · 2021: 7 · 2020: 4
FWCI
4.12
Type
journal article
Language
en
Open access
Licence
Locations
  • IEEE Journal of Selected Topics in Signal Processing (publishedVersion)
  • PubMed (publishedVersion)
  • PubMed Central (acceptedVersion)
  • arXiv (Cornell University) (submittedVersion)
  • arXiv (Cornell University)
Funders
NIH R01 (5R01DC006859)
Topics
  • Health Sciences › Medicine › Physiology › Voice and Speech Disorders
  • Health Sciences › Medicine › Psychiatry and Mental health › Dementia and Cognitive Impairment Research
  • Social Sciences › Psychology › Experimental and Cognitive Psychology › Emotion and Mood Recognition
Keywords
speech features, linguistic features, cognitive disorders, thought disorder, neuropsychological assessment, natural language processing, speech processing, linguistic diversity, syntactic complexity, semantic coherence, speech timing, disease progression monitoring

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Claims read under definition version 3 (extract-v3) on 2026-09-28.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-09-28.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review