How to do things with (thousands of) words: Computational approaches to discourse analysis in Alzheimer's disease.
Level 5 - mechanism / opinion, no new human data
narrative review with no primary human data or systematic quantitative synthesis
PubMed 32622173 · doi:10.1016/j.cortex.2020.05.001
What was done
The authors reviewed literature on the application of Natural Language Processing (NLP), machine learning, and Automatic Speech Recognition (ASR) to analyze connected speech and writing (discourse) across the spectrum of Alzheimer's disease (AD) and preclinical Mild Cognitive Impairment (MCI). They examined key linguistic components, computational extraction tools, predictive modeling approaches, and barriers to clinical translation.
What was found
The abstract reports no quantitative metrics or pooled effect sizes. The authors describe qualitative findings: automated discourse analysis can identify measurable linguistic changes associated with AD pathology and risk, but practical deployment is limited by a lack of large and diverse datasets, ethical constraints in data sharing, suboptimal diagnostic specificity, and hurdles in clinical acceptability.
Why it matters
Automated speech and language processing offers a scalable, non-invasive method for early detection and risk stratification in AD and MCI. Identifying early-stage linguistic biomarkers could improve screening and enrich clinical trial cohorts before irreversible neuropathological damage occurs.
Limits
This is a narrative review presenting high-level conceptual summaries rather than a systematic review or meta-analysis with quantitative performance benchmarks (e.g., sensitivity, specificity, AUC). Specific cohort sizes, disease stages, diagnostic criteria, and language-specific generalizability from the reviewed primary studies are not detailed in the abstract.
Cited by
- supports Linguistic and speech cues of neurodegenerative conditions like Alzheimer's disease can be detected up to 10 years before typical clinical symptoms appear.