A cautionary tale of two corpora: analyzing cognitive impairments using linguistic data
Language samples are increasingly used to support the early detection of cognitive impairment and to develop automated models for identifying mild cognitive impairment (MCI) and dementia. Many studies have relied on Cookie Theft picture descriptions from DementiaBank’s Pitt corpus. Recently, researchers have begun combining data from the newer Delaware corpus to enlarge datasets or train and validate models across corpora. However, demographic and linguistic differences between these datasets have not been fully characterized. This study examined participant characteristics and linguistic measures from the picture descriptions in these two corpora. Multiple logistic regression assessed the homogeneity of the two corpora while controlling for age, sex, and education. Additional analyses compared language outcomes between diagnostic groups within each corpus. Delaware participants were significantly older than Pitt participants, and the Delaware control group included significantly more females. All demographic and linguistic covariates significantly contributed to classification of corpus membership. Among control participants, five of six language measures differed significantly between corpora; among participants with MCI, only grammatical complexity differed significantly. These findings demonstrate substantial heterogeneity between the Pitt and Delaware corpora, particularly among control participants. Researchers should consider demographic, linguistic, and data collection differences when combining these datasets, interpreting cross-corpus findings, or evaluating the generalizability of models trained on these two different corpora.
Authors
- Brian MacWhinney
- Emily Wang
- Joel Greenhouse
- Alyssa Lanzi
- Quingxuan He
- S. Shria Parupudi
- Davida Fromm
- Elina Lee
Institutions
- Carnegie Mellon University (US)
- University of Delaware (US)
Publication Details
- Journal
- Frontiers in Aging Neuroscience
- Published
- 2026-11-20
- DOI
- https://doi.org/10.3389/fnagi.2026.1985729
- Primary Topic
- Dementia and Cognitive Impairment Research
- Type
- article
- Field-Weighted Citation Impact
- 0.00