Identifying novel linguistic biomarkers of mild cognitive impairment in Mandarin-speaking older adults: A quantitative syntactic approach

Abstract

Mild Cognitive Impairment (MCI) is an early symptom of Alzheimer’s disease, commonly observed in older adults. The use of low-cost language biomarkers is becoming an emerging trend. This study aims to investigate whether recently proposed measures in the field of Quantitative Syntax and their combinations have the potential to serve as biomarkers for distinguishing between the MCI and cognitively normal (CN) groups. A portion of the Chinese corpus MCGD (CN = 25, MCI = 16) was used. The elderly participants performed a sequential picture description task to produce connected speech. The transcription and annotation were semi-automatically conducted and manually checked. Eleven dependency-based syntactic features were calculated. We assessed the discriminability of both univariate features and multivariate feature combinations using support vector machine. Results show that all features can be grouped into four clusters. Most measures within the largest cluster demonstrate high intercorrelations and are statistically significant in distinguishing between the MCI and CN groups. Among these, mean dependency distance (MDD) exhibits the strongest discriminative ability (AUC = 0.791 [0.610, 0.944]). Two hierarchical features have relatively weaker performance, while dependency direction indicators show almost no group differentiability. Several feature combinations identified slightly improved performance, but the difference was not statistically significant. Our findings suggest that the classic syntactic biomarker MDD remains the best-performing measure for distinguishing between MCI and CN for Mandarin-speaking older adults, while most dependency-based syntactic measures can serve as alternative markers. In the future, combining MDD with features in other domains holds promising potential for early diagnosis.