Research map / Large language models / Retrieval and knowledge
Novelty Controlled Paraphrase Generation with Retrieval Augmented Conditional Prompt Tuning
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Retrieve semantically similar examples with kNN rather than random examples (ablation) | RAPT · QQP 50K | iBLEU: 6.22 -> 10.61 | Helped |
| Retrieve semantically similar examples with kNN rather than random examples (ablation) | RAPT · ParaSCI ACL | iBLEU: 0.21 -> 8.13 | Helped |
| Condition generation on High novelty rather than using unconditioned RAPT | NC-RAPT · QQP 140K | iBLEU: 14.33 -> 15.05 | Helped |
| Condition generation on High novelty rather than using unconditioned RAPT | NC-RAPT · QQP 50K | Self-TER: 44.99 -> 58.27 | Helped |
| Condition generation on High novelty rather than using unconditioned RAPT | NC-RAPT · ParaSCI ACL | iBLEU: 8.13 -> 9.44 | Helped |
| Condition generation on High novelty rather than using unconditioned RAPT | NC-RAPT · MSRPC | iBLEU: 7.60 -> 7.70 | Helped |
| Condition generation on Low novelty rather than using unconditioned RAPT | NC-RAPT · QQP 50K | BLEU: 31.61 -> 31.35 | Helped |
| Use randomly initialized prompt embeddings instead of vocabulary-embedding initialization (ablation) | Prompt Tuning · QQP 50K | iBLEU: 4.85 -> 4.42 | Hurt |
| Use BiLSTM-associated prefix and infix tokens (ablation) | Prompt Tuning · QQP 50K | iBLEU: 4.85 -> 2.66 | Hurt |
| Apply transformed prompt parameters in every GPT2 layer (ablation) | Prompt Tuning · QQP 50K | Self-TER: 34.23 -> 43.98 | Helped |
| Train separate prefix-infix parameters for every layer (ablation) | Prompt Tuning · QQP 50K | iBLEU: 4.85 -> 4.63 | No clear effect |
Closest papers by meaning
- A Deep Generative Framework for Paraphrase Generation (AAAI 2018 · similarity 0.87)
- Integrating Linguistic Knowledge to Sentence Paraphrase Generation (AAAI 2020 · similarity 0.87)
- The Devil is in the Prompts: Retrieval-Augmented Prompt Optimization for Text-to-Video Generation (CVPR 2025 · similarity 0.86)
- CriSPO: Multi-Aspect Critique-Suggestion-guided Automatic Prompt Optimization for Text Generation (AAAI 2025 · similarity 0.85)
- Retrieval is Accurate Generation (ICLR 2024 · similarity 0.84)
- Evoke: Evoking Critical Thinking Abilities in LLMs via Reviewer-Author Prompt Editing (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.