Graph-Based Word Alignment for Clinical Language Evaluation.

Area of Science:

• Computational linguistics
• Neuroscience
• Medical informatics

Background:

• Neurological disorders, such as dementia and mild cognitive impairment, often impair narrative abilities.
• There is a growing need for objective, non-invasive screening tools for these conditions.
• Automated analysis of spoken language, particularly narrative retellings, shows promise for early detection.

Purpose of the Study:

• To develop an automated method for extracting narrative recall scores from spoken language.
• To improve the accuracy and efficiency of word alignment for analyzing narrative retellings.
• To assess the diagnostic potential of automated narrative recall scores for neurological disorders.

Main Methods:

• Utilized natural language processing (NLP) for analyzing spoken narrative retellings.
• Developed a novel word alignment method based on random walks on a graph, outperforming expectation maximization techniques.
• Extracted narrative recall scores from high-quality word alignments.

Main Results:

• The novel random walk-based word alignment achieved superior accuracy in a fraction of the time compared to traditional methods.
• Automated narrative recall scores yielded diagnostic classification accuracy comparable to manual scoring.
• The proposed method significantly outperformed summary-level text similarity metrics for diagnostic classification.

Conclusions:

• Automated analysis of narrative retellings using improved word alignment is a viable and accurate method for diagnosing neurological disorders.
• The developed NLP techniques offer a flexible, generalizable, and efficient approach to spoken language analysis for diagnostic purposes.
• This method holds potential as a component of objective, unobtrusive screening tools for conditions like dementia.