Research map / Large language models / Language, safety and interpretability
CGMH: Constrained Sentence Generation by Metropolis-Hastings Sampling
Ning Miao, Hao Zhou, Lili Mou, Rui Yan, Lei Li
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 |
|---|---|---|---|
| use CGMH rather than GBS for constrained generation | Keywords-to-sentence generation · 1 keyword | NLL: 7.42 -> 7.04 | Improved |
| use CGMH rather than GBS for constrained generation | Keywords-to-sentence generation · 2 keywords | Human: 0.55 -> 0.61 | Improved |
| use CGMH rather than GBS for constrained generation | Keywords-to-sentence generation · 3 keywords | NLL: 8.59 -> 8.26 | Improved |
| use CGMH rather than GBS for constrained generation | Keywords-to-sentence generation · 4 keywords | Human: 0.55 -> 0.65 | Improved |
| add a matching function to the sentence-generation distribution (ablation) | CGMH paraphrase generation · Quora | BLEU-ref: 18.85 -> 20.89 | Improved |
| add keyword matching only instead of KW + WVM (ablation) | CGMH paraphrase generation · Quora | BLEU-ref: 20.89 -> 20.17 | Hurt |
| add average word-embedding similarity to keyword matching (ablation) | CGMH paraphrase generation · Quora | BLEU-ref: 20.89 -> 20.41 | Hurt |
| add skip-thought similarity to keyword matching (ablation) | CGMH paraphrase generation · Quora | NLL: 7.46 -> 7.78 | Hurt |
| use unsupervised CGMH rather than the AMU correction system | Sentence error correction · JFLEG test | GLEU: 44.85 -> 45.5 | Improved |
| use unsupervised CGMH rather than supervised NRL | Sentence error correction · JFLEG test | GLEU: 53.98 -> 45.5 | Hurt |
Closest papers by meaning
- Incorporating Discriminator in Sentence Generation: a Gibbs Sampling Method (AAAI 2018 · similarity 0.88)
- Show Me How To Revise: Improving Lexically Constrained Sentence Generation with XLNet (AAAI 2021 · similarity 0.87)
- Latent Template Induction with Gumbel-CRFs (NeurIPS 2020 · similarity 0.86)
- Controllable Generation via Locally Constrained Resampling (ICLR 2025 · similarity 0.84)
- Provence: efficient and robust context pruning for retrieval-augmented generation (ICLR 2025 · similarity 0.84)
- Principled Gradient-Based MCMC for Conditional Sampling of Text (ICML 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.
