Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

PINGALA: Prosody-Aware Decoding for Sanskrit Poetry Generation

Manoj Balaji Jagadeeshan, Atul Singh, Nallani Chakravartula Sahith, Amrith Krishna, Pawan Goyal

Abstract

Poetry generation in Sanskrit typically requires the verse to be semantically coherent and adhere to strict prosodic rules. In Sanskrit prosody, every line of a verse is typically a fixed length sequence of syllables adhering to prescribed binary patterns of syllable weights. We observe that instead of treating a verse as a monolithic sequence, segmenting them as grouped-lines leads to significant improvement in semantic coherence by 10\% with comparable metrical adherence. Specifically, PINGALA, our proposed decoding approach is designed to encourage every line to have well-formed words and our token selection biases the model towards it by preferring longer tokens. Writing in Sanskrit follows phonemic orthography, hence using a phonetically aware transliteration scheme, SLP1, increased the metrical alignment by 46\% with comparable semantic similarity, for a instruction fine-tuned large language models like Phi-4. We also introduce a new approach for reference-free evaluation using cross-encoders which achieved better alignment with true poetry instances.

PINGALA: Prosody-Aware Decoding for Sanskrit Poetry Generation

Abstract

Poetry generation in Sanskrit typically requires the verse to be semantically coherent and adhere to strict prosodic rules. In Sanskrit prosody, every line of a verse is typically a fixed length sequence of syllables adhering to prescribed binary patterns of syllable weights. We observe that instead of treating a verse as a monolithic sequence, segmenting them as grouped-lines leads to significant improvement in semantic coherence by 10\% with comparable metrical adherence. Specifically, PINGALA, our proposed decoding approach is designed to encourage every line to have well-formed words and our token selection biases the model towards it by preferring longer tokens. Writing in Sanskrit follows phonemic orthography, hence using a phonetically aware transliteration scheme, SLP1, increased the metrical alignment by 46\% with comparable semantic similarity, for a instruction fine-tuned large language models like Phi-4. We also introduce a new approach for reference-free evaluation using cross-encoders which achieved better alignment with true poetry instances.

Paper Structure

This paper contains 46 sections, 21 equations, 4 figures, 14 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Reference Free Evaluation
  3. Syntactic Metrics
  4. Semantic Similarity
  5. Implementation notes.
  6. Cross-Encoder Model Definition
  7. Training Data Construction
  8. Weighted Binary Cross-Entropy Objective
  9. Probability Gap as the Optimization Target
  10. Empirical Outcome
  11. Qualitative Analysis
  12. Method
  13. Fine-tuning
  14. PINGALA
  15. General Definitions
  16. ...and 31 more sections

Figures (4)

  • Figure 1: An infographic comparison of working of PINGALA vs CD from Chandomitra.
  • Figure 2: Impact of constraint weight $\beta$ on syntactic and semantic quality. Semantic quality peaks at $\beta=5$, illustrating the optimal balance between structural enforcement and meaning preservation.
  • Figure 3: Instruction fine-tuning Prompt
  • Figure 4: Phi-4 SLP1 Instruction fine-tuning Prompt