Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Linguistic Indicators of Early Cognitive Decline in the DementiaBank Pitt Corpus: A Statistical and Machine Learning Study

Artsvik Avetisyan, Sachin Kumar

TL;DR

This study investigates linguistic markers of early cognitive decline using the DementiaBank Pitt Corpus across three representations: raw text, POS-enhanced, and POS-only. Employing interpretable logistic regression and random forest models, it evaluates performance under transcript-level and subject-level cross-validation, and validates findings with Mann–Whitney U tests and Cliff’s delta. Results show that lexical diversity, functional word usage, sentence structure, and discourse coherence robustly distinguish dementia from control speech, with POS-based representations often offering stronger, more generalizable signals. The work demonstrates that linguistically grounded features, validated statistically, can underpin transparent language-based cognitive screening suitable for clinical settings, while acknowledging task, demographic, and modality limitations and outlining paths for future multimodal and longitudinal extensions.

Abstract

Background: Subtle changes in spontaneous language production are among the earliest indicators of cognitive decline. Identifying linguistically interpretable markers of dementia can support transparent and clinically grounded screening approaches. Methods: This study analyzes spontaneous speech transcripts from the DementiaBank Pitt Corpus using three linguistic representations: raw cleaned text, a part-of-speech (POS)-enhanced representation combining lexical and grammatical information, and a POS-only syntactic representation. Logistic regression and random forest models were evaluated under two protocols: transcript-level train-test splits and subject-level five-fold cross-validation to prevent speaker overlap. Model interpretability was examined using global feature importance, and statistical validation was conducted using Mann-Whitney U tests with Cliff's delta effect sizes. Results: Across representations, models achieved stable performance, with syntactic and grammatical features retaining strong discriminative power even in the absence of lexical content. Subject-level evaluation yielded more conservative but consistent results, particularly for POS-enhanced and POS-only representations. Statistical analysis revealed significant group differences in functional word usage, lexical diversity, sentence structure, and discourse coherence, aligning closely with machine learning feature importance findings. Conclusion: The results demonstrate that abstract linguistic features capture robust markers of early cognitive decline under clinically realistic evaluation. By combining interpretable machine learning with non-parametric statistical validation, this study supports the use of linguistically grounded features for transparent and reliable language-based cognitive screening.

Linguistic Indicators of Early Cognitive Decline in the DementiaBank Pitt Corpus: A Statistical and Machine Learning Study

TL;DR

This study investigates linguistic markers of early cognitive decline using the DementiaBank Pitt Corpus across three representations: raw text, POS-enhanced, and POS-only. Employing interpretable logistic regression and random forest models, it evaluates performance under transcript-level and subject-level cross-validation, and validates findings with Mann–Whitney U tests and Cliff’s delta. Results show that lexical diversity, functional word usage, sentence structure, and discourse coherence robustly distinguish dementia from control speech, with POS-based representations often offering stronger, more generalizable signals. The work demonstrates that linguistically grounded features, validated statistically, can underpin transparent language-based cognitive screening suitable for clinical settings, while acknowledging task, demographic, and modality limitations and outlining paths for future multimodal and longitudinal extensions.

Abstract

Background: Subtle changes in spontaneous language production are among the earliest indicators of cognitive decline. Identifying linguistically interpretable markers of dementia can support transparent and clinically grounded screening approaches. Methods: This study analyzes spontaneous speech transcripts from the DementiaBank Pitt Corpus using three linguistic representations: raw cleaned text, a part-of-speech (POS)-enhanced representation combining lexical and grammatical information, and a POS-only syntactic representation. Logistic regression and random forest models were evaluated under two protocols: transcript-level train-test splits and subject-level five-fold cross-validation to prevent speaker overlap. Model interpretability was examined using global feature importance, and statistical validation was conducted using Mann-Whitney U tests with Cliff's delta effect sizes. Results: Across representations, models achieved stable performance, with syntactic and grammatical features retaining strong discriminative power even in the absence of lexical content. Subject-level evaluation yielded more conservative but consistent results, particularly for POS-enhanced and POS-only representations. Statistical analysis revealed significant group differences in functional word usage, lexical diversity, sentence structure, and discourse coherence, aligning closely with machine learning feature importance findings. Conclusion: The results demonstrate that abstract linguistic features capture robust markers of early cognitive decline under clinically realistic evaluation. By combining interpretable machine learning with non-parametric statistical validation, this study supports the use of linguistically grounded features for transparent and reliable language-based cognitive screening.
Paper Structure (24 sections, 2 equations, 14 figures, 9 tables)

This paper contains 24 sections, 2 equations, 14 figures, 9 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Dataset Description
  4. Methodology
  5. Feature Sets
  6. Machine Learning Models
  7. Feature Importance Extraction
  8. Statistical Association Analysis
  9. Results
  10. Baseline Transcript-Level Results
  11. Classification Performance on Raw Cleaned Text
  12. Feature Importance Analysis
  13. Classification Performance with POS-Enhanced Representation
  14. Feature Importance Analysis for POS-Enhanced Representation
  15. Classification Performance with POS-Only Representation
  16. ...and 9 more sections

Figures (14)

  • Figure 1: Top logistic regression features for the raw cleaned text representation
  • Figure 2: Top random forest features for the raw cleaned text representation
  • Figure 3: Top logistic regression features for the POS-enhanced representation
  • Figure 4: Top random forest features for the POS-enhanced representation
  • Figure 5: Top logistic regression feature coefficients for the POS-only representation. Positive coefficients indicate features associated with dementia transcripts, while negative coefficients are associated with control transcripts.
  • ...and 9 more figures