Transformer‐based Deep Learning Architecture Improves Detection of Associations between Spontaneous Speech Language Markers and Cognition
Bibliographic record
Abstract
Abstract Background Spontaneous speech is easily obtainable and has the potential to become an accessible and low‐cost marker for cognitive function. The time‐consuming and labor‐intensive nature of speech analysis has been a major obstacle to utilizing this promising tool. This study uses a novel transformer‐based methodology to explore associations between spontaneous speech language features and global cognition. Method Speech recordings were obtained from participants with clinical diagnoses of mild cognitive impairment (MCI), dementia, and cognitively unimpaired, from the Alzheimer’s Disease Research Center (ADRC) at University of California, Irvine. Audio samples were denoised and transcribed using our transformer‐based model (Figure 1). The model conducted association analyses between global cognition, measured by Montreal Cognitive Assessment (MoCA) scores, and 5 linguistic features pre‐specified based on popularity in language analysis. Features were extracted from transcripts via Bidirectional Encoder Representations from Transformers (BERT), which assigns each word a special “token” (tokenizing), translates “tokens” into a numerical format for computer comprehension (encoding), and generates unique identifiers (embeddings) that mathematically capture meanings and relationships between words. A linear regression (LR) model was then trained on power‐transformed BERT‐extracted linguistic features and its performance was tested against both untransformed and power‐transformed data without BERT feature extraction using the same linguistic features. Results In a cohort comprising 73 healthy controls and 12 individuals with MCI or dementia, our analysis, utilizing pre‐specified linguistic features and leveraging BERT for enhancing feature extraction, revealed that vocabulary richness ( p = 0.009), average word length ( p = 0.005), and semantic coherence ( p = 0.047) were significantly associated with MoCA scores (Table 1). The non‐BERT untransformed model found no significant associations with MoCA while the non‐BERT power‐transformed model found only one significant association: average word length ( p = 0.016). Conclusion Our study's novel approach of employing BERT for linguistic feature extraction in association analysis increased the number of pre‐selected speech features significantly associated with global cognition, namely vocabulary richness, average word length, and semantic coherence. This improvement in association detection highlights the potential for better deep‐learning dementia detection methods and might lead to increased utility of spontaneous speech as an easily obtainable and scalable cognitive measure.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.000 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one teacher head, not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".