Research map / Image generation / Text-to-image diffusion
Controlled Text Generation as Continuous Optimization with Multiple Constraints
Sachin Kumar, Eric Malmi, Aliaksei Severyn, Yulia Tsvetkov
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 |
|---|---|---|---|
| Add cross-lingual similarity constraint during decoding (ablation) | MuCoCO on Marian Transformer fr-en · OpenSubtitles French-English test subset | BLEU: 42.1 -> 42.7 | Improved |
| Add cross-lingual similarity constraint during decoding (ablation) | MuCoCO on Marian Transformer fr-en · OpenSubtitles French-English test subset | BERTScore: 0.932 -> 0.939 | Improved |
| Add formality constraint only (ablation) | MuCoCO on Marian Transformer fr-en · OpenSubtitles French-English test subset | Formality: 0% -> 30% | Improved |
| Add formality constraint only (ablation) | MuCoCO on Marian Transformer fr-en · OpenSubtitles French-English test subset | BLEU: 42.1 -> 37.5 | Hurt |
| Combine formality and cross-lingual similarity constraints (ablation) | MuCoCO on Marian Transformer fr-en · OpenSubtitles French-English test subset | BLEU: 37.5 -> 39.8 | Mixed |
| Use formal constraint only instead of FUDGE (ablation) | MuCoCO on STRAP · GYAFC Entertainment/Music test | Transfer: 85% -> 93% | Improved |
| Use formal constraint only instead of FUDGE (ablation) | MuCoCO on STRAP · GYAFC Entertainment/Music test | WSim (input): 0.71 -> 0.67 | Hurt |
| Add usim semantic-similarity constraint to style-transfer decoding (ablation) | MuCoCO on STRAP · GYAFC Entertainment/Music test | WSim (input): 0.67 -> 0.71 | Improved |
| Add WMD constraint to usim (ablation) | MuCoCO on STRAP · GYAFC Entertainment/Music test | WSim (input): 0.71 -> 0.73 | Improved |
| Add formality constraint alongside usim and WMD (ablation) | MuCoCO on STRAP · GYAFC Entertainment/Music test | Transfer: 87% -> 92% | Improved |
| Tighten formality and similarity thresholds (ablation) | MuCoCO · Style transfer Pareto-front examples | Output fluency | Hurt |
| Use multiple differentiable constraints for requested attributes | MuCoCO paraphrasing model · Yelp review multi-attribute examples | Quantitative evaluation | No clear effect |
Closest papers by meaning
- Syntactic and Semantic Control of Large Language Models via Sequential Monte Carlo (ICLR 2025 · similarity 0.87)
- Controlled Text Generation via Language Model Arithmetic (ICLR 2024 · similarity 0.87)
- Controlled LLM Decoding via Discrete Auto-regressive Biasing (ICLR 2025 · similarity 0.87)
- Successor Features for Efficient Multi-Subject Controlled Text Generation (ICML 2024 · similarity 0.86)
- Principled Gradient-Based MCMC for Conditional Sampling of Text (ICML 2024 · similarity 0.86)
- Guaranteed Generation from Large Language Models (ICLR 2025 · similarity 0.85)
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.
