Confidence-Guided Modular Multilingual Translation Framework (CGMMTF) for Content Scarcity, Ambiguity Resolution and Reordering
Accurate multilingual translation remains challenging due to data imbalance, negative transfer, and uncontrolled error propagation. This paper proposes a Confidence-Guided Modular Multilingual Translation Framework (CGMMTF) to address these limitations. The framework integrates language-aware modular learning with confidence-driven adaptation to improve translation reliability across languages. CGMMTF selectively shares knowledge while preserving language-specific characteristics. Error propagation is reduced through internal consistency checks. Experimental evaluation demonstrates improved performance across standard metrics. The proposed system achieves a Word Error Rate of 0.15, a BLEU score of 91.9, a METEOR score of 0.65 and a translation accuracy of 97.5%. These results indicate enhanced semantic adequacy, robustness and generalization in multilingual translation scenarios.
