TransExplain-LD: Emotion-Aware and Explainable Machine Translation with Adaptive Latent Depth

Standard neural machine translation systems fail to preserve emotional nuance and cultural context, critical in educational, therapeutic and cross-cultural communication. We present TransExplain-LD, a neuro-symbolic Transformer that (i) dynamically adjusts depth based on emotional complexity, (ii) integrates hybrid embeddings fusing affective lexicons, contextual emotion vectors and knowledge graphs, and (iii) employs template-guided decoding for interpretable output. Optimized with a multi-objective loss balancing translation quality, emotional fidelity and explanation clarity, our framework achieves +2.6 BLEU and +0.24 Emotion Preservation Index (EPI) improvements while reducing computation by 35%. Two novel metrics-EPI and Human Explainability Score (HES)-quantify emotional preservation and translation interpretability. Results across multilingual benchmarks demonstrate that emotion-aware adaptive-depth architectures enhance translation accuracy, fidelity and efficiency simultaneously.