Natural Language Processing for Multilingual Chatbots in Healthcare

Abstract

This paper investigates the development of multilingual NLP-enabled chatbots for healthcare, emphasizing the need for equitable access across linguistic barriers—especially critical amid global health crises. Grounded solely in 2020 research, we reference HHH, a medical chatbot framework that combines a knowledge graph with a Hierarchical BiLSTM Attention Model (HBAM), demonstrating superior performance compared to BERT and MaLSTM on medical question matching tasks . We also draw from the ―Conversational Agents in Health Care‖ scoping review, which highlights widespread use of chatbots in treatment, monitoring, healthcare support, and patient education, while noting a lack of robust evaluation .

We propose a modular model that integrates multilingual translation, knowledge-graph reasoning, and domain-specific intent detection. The methodology leverages translation layers or pivot-driven approaches to support multiple languages, combined with a knowledge-graph-based reasoning engine (à la HHH) to maintain accuracy. Evaluation combines technical performance metrics (e.g., accuracy) with user-centric usability indicators as identified in the 2020 scoping review .

Key findings suggest that (1) hybrid knowledge-graph systems like HHH offer strong medical question matching; (2) existing conversational agents in healthcare demonstrate efficacy in varied applications but need systematic evaluation . We outline a workflow: language detection → translation → intent/graph matching → generation → back-translation. Advantages include structured reasoning, multilingual reach, and knowledge maintainability; disadvantages involve complexity and translation reliability. Results show promise in multilingual question-answering effectiveness, yet highlight evaluation gaps. The conclusion affirms the feasibility of multilingual medical chatbots with knowledge-graph NLP, and we recommend future work on low-resource languages, rigorous evaluation in multilingual settings, and integration of speech modalities