Research map / Image generation / Autoregressive and latent generation
Controllable Text Generation with Neurally-Decomposed Oracle
Tao Meng, Sidi Lu, Nanyun Peng, Kai-Wei Chang
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 NADO to Seq2seq base model | GPT-2-Large Seq2seq · CommonGen (dev) | BLEU-4: 23.5 -> 30.8 | Improved |
| add NADO to Seq2seq base model | GPT-2-Large Seq2seq · CommonGen (dev) | constraint coverage (%): 87.0 -> 97.1 | Improved |
| add NADO to Seq2seq base model | GPT-2-Large Seq2seq · News (test) | constraint coverage (%): 74.0 -> 91.9 | Improved |
| add NADO to domain-adapted base model with warmup | GPT-2-Large DA · CommonGen (dev) | constraint coverage (%): 79.7 -> 96.1 | Improved |
| add NADO on top of warmup (Seq2seq) (ablation) | Seq2seq + warmup · CommonGen (dev) | BLEU-4: 24.8 -> 30.3 | Improved |
| warmup only versus NADO training only (Seq2seq) (ablation) | Seq2seq base · CommonGen (dev) | BLEU-4: 24.8 -> 30.8 | Improved |
| add warmup before NADO training (Seq2seq) (ablation) | Seq2seq + NADO · CommonGen (dev) | BLEU-4: 30.8 -> 30.3 | Hurt |
| add warmup before NADO training (Seq2seq) (ablation) | Seq2seq + NADO · Yelp Review (test) | constraint coverage (%): 96.7 -> 92.8 | Hurt |
| add NADO to GPT-2 base model | GPT-2 base Seq2seq · CommonGen (dev) | constraint coverage (%): 76.8 -> 85.5 | Improved |
| add NADO to GPT-2 base model | GPT-2 base Seq2seq · CommonGen (dev) | BLEU-4: 22.98 -> 24.01 | Improved |
| add NADO oracle guidance vs FUDGE token-level auxiliary model | MarianMT · Fisher/CALLHOME formal MT | BLEU: 17.96 -> 20.84 | Improved |
| add NADO vs FUDGE | MarianMT · Fisher/CALLHOME formal MT | average formality score: 0.51 -> 0.54 | Improved |
Closest papers by meaning
- DiNADO: Norm-Disentangled Neurally-Decomposed Oracles for Controlling Language Models (ICML 2024 · similarity 0.90)
- Successor Features for Efficient Multi-Subject Controlled Text Generation (ICML 2024 · similarity 0.88)
- Reflection-Window Decoding: Text Generation with Selective Refinement (ICML 2025 · similarity 0.87)
- Improving Open-Ended Text Generation via Adaptive Decoding (ICML 2024 · similarity 0.86)
- Closing the Curious Case of Neural Text Degeneration (ICLR 2024 · similarity 0.86)
- Controlled Text Generation via Language Model Arithmetic (ICLR 2024 · similarity 0.86)
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.
