Multivariate computational linguistic analysis for early detection of cognitive impairment

Abstract

AbstractBackgroundThe overarching goal of this study was to develop a novel computational linguistic algorithm that can be used to build a machine learning platform for recognizing changes in cognitive abilities during the progression of psychopathological disorders or predicting clinical prognosis following a brain injury or therapeutic interventions.MethodThe computational linguistic algorithm developed in this study evaluates lexical diversity, lexical complexity, and emotional tones in a quantitative manner. An existing algorithm that quantifies idea density was also included in this study. The initial proof‐of‐concept study was performed on various published books by well‐known authors, including Iris Murdoch who developed Alzheimer’s disease (AD) later in life. Ten interviews of patients with mild AD and ten interviews of healthy age‐matched individuals were evaluated. The transcript of each video was produced manually. Each transcript was analyzed using the computational linguistic algorithm for lexical diversity/complexity, emotional tones, and idea density. Quantitative linguistic values were compared between patients with mild AD and healthy age‐matched individuals.ResultA computational linguistic analysis of various books resulted in findings that were consistent with the nature and subject of each book. Among all the computational linguistic variables evaluated in this study, the book written by Irish Murdoch after the onset of AD showed a decrease in the level of lexical complexity by 17.5%. Interestingly, a 20% decrease in the level of lexical complexity was also noted when video interviews of patients with mild AD were compared to normal age‐matched individuals. The video transcripts of AD patients showed a statistically significant increase in the levels of anger, fear, and sadness. We are currently conducting a study to compare computational linguistic variables in response to writing prompts between patients with mild AD and healthy age‐matched individuals. Preliminary results indicate that patients with AD show a similar decrease in the level of lexical complexity as well as quantifiable changes in emotional tones.ConclusionThe computational linguistic algorithm developed in this study may lead to a machine learning approach or artificial intelligence platform to detect linguistic changes that may occur in the human cognitive state for both diagnosis and/or prognosis of mental health conditions.