Abstract
While semantic communication systems outperform traditional ones, most current research focuses on a single language, overlooking multilingual contexts. This letter proposes two approaches to extend the Text Semantic Communication System (TSC) to a Multilingual Text Semantic Communication System (MTSC). The first employs centralized learning on a hybrid dataset, processing multilingual texts with composite word-sequence indices. The second utilizes federated learning to aggregate linguistic features while preserving user data privacy. To assess the MTSC system, we introduce the Multi-Bilingual Evaluation Understudy (MBLEU) score. Experimental results show that the MTSC can extend the TSC without increasing model size, with federated learning achieving superior multilingual performance while protecting data privacy.