From Plain Text to Poetic Form: Generating Metrically-Constrained Sanskrit Verses

2025-06-01

AI Summary

Generating classical Sanskrit poetry—strictly adhering to metrical constraints such as Anuṣṭubh—in low-resource, morphologically rich languages remains challenging due to scarce high-quality training data and inadequate modeling of prosodic structure. Method: We introduce the first parallel corpus of Sanskrit metrical poetry; propose a metricality-aware constrained decoding strategy; and design a semantic-prosodic co-fine-tuning paradigm integrating Sanskrit morphological analysis, rule-based prosody modeling, and instruction tuning across multiple open-source and commercial LLMs. Contribution/Results: Our decoding approach achieves >99% metrical compliance. Fine-tuned models significantly outperform baselines in semantic fidelity and stylistic appropriateness (p < 0.01, human evaluation). This work establishes a reusable data resource, methodology, and evaluation framework for structured literary generation in low-resource languages.

Technology Category

• Natural Language Processing: Code Generation / Program Synthesis from Natural Language
• Machine Learning: Large Multimodal Models (LMMs)
• Planning, Routing, and Scheduling: Planning with Language Models

Application Category

• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems

Abstract

Recent advances in large language models (LLMs) have significantly improved natural language generation, including creative tasks like poetry composition. However, most progress remains concentrated in high-resource languages. This raises an important question: Can LLMs be adapted for structured poetic generation in a low-resource, morphologically rich language such as Sanskrit? In this work, we introduce a dataset designed for translating English prose into structured Sanskrit verse, with strict adherence to classical metrical patterns, particularly the Anushtub meter. We evaluate a range of generative models-both open-source and proprietary-under multiple settings. Specifically, we explore constrained decoding strategies and instruction-based fine-tuning tailored to metrical and semantic fidelity. Our decoding approach achieves over 99% accuracy in producing syntactically valid poetic forms, substantially outperforming general-purpose models in meter conformity. Meanwhile, instruction-tuned variants show improved alignment with source meaning and poetic style, as supported by human assessments, albeit with marginal trade-offs in metrical precision.

Problem

• Adapting LLMs for structured Sanskrit poetry generation
• Translating English prose into metrically-constrained Sanskrit verses
• Ensuring metrical and semantic fidelity in low-resource languages

Innovation

• Constrained decoding for metrical accuracy
• Instruction-based fine-tuning for semantic fidelity
• Dataset for English-Sanskrit verse translation