Objectives: The main objective of this paper is to propose an approach for developing an Artificial Intelligence (AI)-powered Language Assessment (LA) tool. Such tools can be used to assess language impairments associated with dementia in older adults. The Machine Learning (ML) classifiers are the main parts of our proposed approach, therefore to develop an accurate tool with high sensitivity and specificity, we consider different binary classifiers and evaluate their performances. We also assess the reliability and validity of our approach by comparing the impact of different types of language tasks, features, and recording media on the performance of ML classifiers. Approach: Our approach includes the following steps: 1) Collecting language datasets or getting access to available language datasets; 2) Extracting linguistic and acoustic features from subjects' speeches which have been collected from subjects with dementia (N=9) and subjects without dementia (N=13); 3) Selecting most informative features and using them to train ML classifiers; and 4) Evaluating the performance of classifiers on distinguishing subjects with dementia from subjects without dementia and select the most accurate classier to be the basis of the AI tool. Results: Our results indicate that 1) we can nd more predictive linguistic markers to distinguish language impairment associated with dementia from participants' speech produced during the Picture Description (PD) language task than the Story Recall (SR) task; and 2) phone-based recording interfaces provide more high-quality language datasets than the web-based recording systems Conclusion: Our results verify that the tree-based classifiers, which have been trained using the linguistic and acoustic features extracted from interviews' transcript and audio, can be used to develop an AI-powered language assessment tool for detecting language impairment associated with dementia.

Objectives: The main objective of this paper is to propose a methodology based on machine learning classifiers for assessing language impairments associated with dementia in older adults. To do so, we compare the impact of different types of language tasks, features, and recording media on our ML-based methodology’s efficiency. Methodology: The methodology encompasses the following steps: 1) Extracting linguistic and acoustic features from subjects’ speeches which have been collected from subjects with dementia ( N =9) and subjects without dementia ( N =13); 2) Employing feature selection methods to rank informative features; 3) Training ML classifiers using extracted features to recognize subjects with dementia from subjects without dementia; 4) Evaluating the classifiers; 5) Selecting the most accurate classifiers to develop the languages assessment tools. Results: Our results indicate that 1) we can find more predictive linguistic markers to distinguish language impairment associated with dementia from participants’ speech produced during the picture description language task than the story recall task. 2) a phone-based recording interface provides a more high-quality language dataset than the web-based recording systems; 3) classifiers trained with selected features from acoustic features or linguistic features show higher performance than the classifiers trained with pure features. Conclusion: Our results show that the tree-based classifiers that have been trained using the PD dataset can be used to develop an ML-based language assessment tool that can detect language impairment associated with dementia as quickly as possible.