Translingual Language Markers for Cognitive Assessment from Spontaneous Speech

Area of Science:

• Computational linguistics
• Neuroscience
• Artificial Intelligence

Background:

• Mild Cognitive Impairment (MCI) is a precursor to dementia, necessitating early detection for intervention.
• Current detection methods can be invasive or costly.
• Language markers in spontaneous speech offer a non-intrusive, accessible approach for MCI identification.

Purpose of the Study:

• To investigate the efficacy of bilingual language markers from spontaneous speech for MCI detection.
• To predict Mini-Mental Status Examination (MMSE) scores using these language markers.
• To develop a robust framework for MCI identification and cognitive assessment.

Main Methods:

• Extracted bilingual acoustic, semantic, and syntactic features from spontaneous speech.
• Employed a framework addressing model complexity for small sample sizes and imbalanced demographics.
• Utilized a multi-task learning approach, incorporating clinical classification prediction as a prior for MMSE score prediction.

Main Results:

• Achieved an average Balanced Accuracy of 78.2% for MCI detection.
• Obtained an average Root Mean Squared Error (RMSE) of 2.705 for MMSE score prediction.
• Demonstrated that translingual language markers enhance MCI detection from spontaneous speech.

Conclusions:

• Bilingual language markers derived from spontaneous speech are effective for identifying MCI.
• This approach shows promise for early and accessible detection of cognitive decline.
• The developed framework offers a valuable tool for clinical research and potential diagnostic support.