SLIP: A Sanskrit Linguistic Intelligence Pipeline for Enhanced Neural Machine Translation of Classical Texts
N Biraja Isac, Himansu Das · IEEE Access · 2025
The translation of classical Sanskrit texts presents unique challenges due to the language's complex morpho-phonological structure, rich grammatical system, and cultural-philosophical depth. This paper introduces the Sanskrit Linguistic Intelligence Pipeline (SLIP), a comprehensive framework that integrates traditional Sanskrit grammatical analysis with modern neural machine translation architectures to enhance translation quality for classical texts. SLIP incorporates morphological feature extraction (verb tenses, case endings, prefixes), phonological analysis (vowel ratios, syllable patterns, sandhi rules), and structural processing (compound word detection, sentence analysis) specifically designed for Sanskrit linguistic characteristics. The paper evaluates SLIP across multiple neural architectures including enhanced transformers, MBART, and retrieval-augmented generation (RAG) systems using diverse evaluation metrics (BLEU, ROUGE, METEOR, BERTScore) on Bhagavad Gita verses spanning different thematic categories. The results demonstrate substantial improvements: SLIP-enhanced transformers achieve a 10.20% relative improvement in BERT F1-score (57.63% to 63.51%) with statistical significance, while RAG integration yields remarkable gains of up to 41.8% in BLEU-4 scores. Comparative analysis reveals that domain-specific linguistic intelligence significantly outperforms generic large language models, with enhanced transformers showing +5.88 BERT F1 improvement compared to MBART's -7.12 decline. Thematic analysis indicates variable effectiveness across philosophical concepts, with austerity practices showing the highest improvement (+7.23%) compared to food classifications (+4.73%). These findings demonstrate that Sanskrit-specific morpho-phonological feature engineering provides superior translation quality compared to generic multilingual approaches, offering a paradigm for preserving semantic accuracy and cultural nuances in automated translation of ancient languages. The SLIP framework addresses critical limitations in existing approaches and provides a foundation for broader digitization efforts of classical Indian literature.