Research map / Large language models / Instruction and fine-tuning
PoeTone: A Framework for Constrained Generation of Structured Chinese Songci with LLMs
Zhan Qu, Shuzhou Yuan, Michael Färber
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Generate-Critic best-of-N SFT with rule-based critic | Qwen3-8B · Songci formal conformity (one-shot) | Formal Conformity Score (%): +5.88 | Improved |
| Generate-Critic SFT | Qwen3-8B · Songci formal conformity (chain-of-thought) | Formal Conformity Score (%): -3.11 | Hurt |
| Generate-Critic SFT | Qwen3-8B · Songci formal conformity (instruction) | Formal Conformity Score (%): +0.33 | No clear effect |
| Generate-Critic SFT | Llama-3.1-8B-Instruct · Songci formal conformity (all prompts) | Formal Conformity Score (%): +1.11 to +1.63 | Improved |
| Generate-Critic SFT | Mistral-Small-Instruct-2409 · Songci formal conformity (one-shot) | Formal Conformity Score (%): +3.40 | Improved |
| Generate-Critic SFT | Mistral-Small-Instruct-2409 · Songci formal conformity (instruction and CoT) | Formal Conformity Score (%): -1.46 (instruction), -1.13 (CoT) | Hurt |
| use chain-of-thought prompting instead of zero-shot (ablation) | GPT-4o · Songci formal conformity | Formal Conformity Score (%): 78.62 -> 71.11 | Hurt |
| use one-shot example instead of zero-shot (ablation) | Llama-3.3-70B-Instruct · Songci formal conformity | Formal Conformity Score (%): 35.88 -> 63.07 | Improved |
| use one-shot example instead of zero-shot (ablation) | GPT-4o · Songci formal conformity | Formal Conformity Score (%): 78.62 -> 76.59 | Hurt |
| use completion prompt instead of zero-shot (ablation) | ERNIE 4.5 Turbo · Songci formal conformity | Formal Conformity Score (%): 77.36 -> 82.38 | Improved |
| explicit rule instructions instead of one-shot (ablation) | Qwen3-32B · Songci formal conformity | Formal Conformity Score (%): 68.55 -> 56.51 | Hurt |
| higher formal conformity (Qwen3-32B) versus poetic quality | Qwen3-32B · Songci poetic quality | human Overall Quality (1-5): 2.50 (vs DeepSeek-R1-0528-Qwen3-8B 2.85) | Hurt |
Closest papers by meaning
- PoetryDiffusion: Towards Joint Semantic and Metrical Manipulation in Poetry Generation (AAAI 2024 · similarity 0.89)
- WenMind: A Comprehensive Benchmark for Evaluating Large Language Models in Chinese Classical Literature and Language Arts (NeurIPS 2024 · similarity 0.88)
- WritingBench: A Comprehensive Benchmark for Generative Writing (NeurIPS 2025 · similarity 0.86)
- LongGenBench: Benchmarking Long-Form Generation in Long Context LLMs (ICLR 2025 · similarity 0.86)
- Turning Up the Heat: Min-p Sampling for Creative and Coherent LLM Outputs (ICLR 2025 · similarity 0.85)
- Sheared LLaMA: Accelerating Language Model Pre-training via Structured Pruning (ICLR 2024 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
