Research map / Image generation / Text-to-image diffusion
Relation-Constrained Decoding for Text Generation
Xiang Chen, Zhixian Yang, Xiaojun Wan
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 |
|---|---|---|---|
| remove probability surgery (ablation) | RESEAL · English-EWT-Dep dependency placement (Stanza) | LC: 79.26 -> 70.91 | Improved |
| replace RG-Top-K with original DBA beam allocation (ablation) | RESEAL · English-EWT-Dep dependency placement (Stanza) | LC: 79.26 -> 74.27 | Improved |
| allocate beams by lexical plus relation constraint count (word+rel) (ablation) | RESEAL · English-EWT-Dep dependency placement (Stanza) | LC: 79.26 -> 74.51 | Improved |
| set lambda to 0 (no probability surgery) (ablation) | RESEAL · English-EWT-Dep dependency placement | LC | Improved |
| increase lambda above 4 (ablation) | RESEAL · English-EWT-Dep dependency placement | LC | No clear effect |
| apply RESEAL on top of DBA decoding | DBA · English-EWT-Dep dependency placement (Stanza) | LC: 67.39 -> 79.26 | Improved |
| choose words during decoding using parse feedback instead of reranking | Rerank · English-EWT-Dep dependency placement (Stanza) | LC: 74.86 -> 79.26 | Improved |
| add RESEAL with predicted constraints | BART · Gigaword | ROUGE-1: 50.14 -> 50.73 | Improved |
| add RESEAL with predicted constraints | BART · MSR-ATC | ROUGE-1: 40.39 -> 43.77 | Improved |
| use gold instead of predicted dependency constraints (ablation) | BART+RESEAL · MSR-ATC | ROUGE-1: 43.77 -> 69.42 | Improved |
| add RESEAL | Seq2Seq · WebEdit | SARI: 73.74 -> 78.33 | Improved |
| add RESEAL | Seq2Seq · WebEdit | BLEU-4: 82.96 -> 84.12 | Improved |
Closest papers by meaning
- Controlled LLM Decoding via Discrete Auto-regressive Biasing (ICLR 2025 · similarity 0.85)
- Guiding LLMs The Right Way: Fast, Non-Invasive Constrained Generation (ICML 2024 · similarity 0.84)
- Enhancing Low-Resource Relation Representations through Multi-View Decoupling (AAAI 2024 · similarity 0.84)
- Improving Open-Ended Text Generation via Adaptive Decoding (ICML 2024 · similarity 0.84)
- Reflection-Window Decoding: Text Generation with Selective Refinement (ICML 2025 · similarity 0.83)
- Language Model Decoding as Direct Metrics Optimization (ICLR 2024 · similarity 0.83)
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.
