MConAI: Multilingual Conversational AI for Low-Cost Early Alzheimer's Detection

Abstract

Dementia is a progressive neurocognitive disorder that affects memory and cognitive function, with Alzheimer’s disease (AD) as its most prevalent form. Early language impairments offer a non-invasive biomarker for timely detection, yet most studies rely on single-language datasets and binary classification, limiting cross-lingual applicability. We present MConAI, a multilingual conversational AI framework that leverages a unified, dementia-related conversational dataset for early detection of AD and Mild Cognitive Impairment (MCI). GPT-4 preprocesses and standardizes multilingual transcripts across English, Spanish, and Mandarin, while mBERT extracts cross-lingual semantic and syntactic embeddings to identify subtle cognitive-linguistic biomarkers. Evaluated on multiple datasets, MConAI achieves an accuracy of 87% in English and an average accuracy of 83% across all languages with multiclass classification, demonstrating robust performance across languages and tasks. The key contributions of this work include (1) a scalable multilingual AI framework for early Alzheimer’s detection, (2) a unified multilingual conversational dataset, and (3) evaluation of cross-lingual transfer performance. MConAI offers an affordable, language-independent alternative to conventional clinical assessments. These findings highlight MConAI’s potential as a scalable, affordable, and language-independent solution to address the global need for early Alzheimer’s detection.

Ähnliche Arbeiten