Research map / Large language models / Language, safety and interpretability
Entailment Relation Aware Paraphrase Generation
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 |
|---|---|---|---|
| Add weak entailment-relation supervision during ParaBank pre-training | Paraphrase generator · ParaBank held-out | iBLEU: 17.78 -> 19.06 | Helped |
| Use ERAP-A instead of ERAP-U | ERAP · Recasted SICK test | iBLEU: 9.43 -> 14.43 | Helped |
| Add entailment-aware control to paraphrase generation | ERAP · Recasted SICK human evaluation | relation consistency (%): 40.00 -> 64.00 | Helped |
| Augment SICK NLI training data with ERAP-A equivalence paraphrases instead of Fine-tuned-U equivalence paraphrases | RoBERTa textual entailment classifier · SICK NLI adversarial test | accuracy: 69.72 -> 78.44 | Helped |
| Add paraphrase augmentation rather than training only on original data | RoBERTa textual entailment classifier · SICK NLI adversarial test | accuracy: 83.02 -> 78.44 | Hurt |
| Use only the relation-consistency scorer (ablation) | ERAP-A · Recasted SICK test | R-Consistency: 65.53 -> 96.75 | Helped |
| Use only the relation-consistency scorer (ablation) | ERAP-A · Recasted SICK test | BLEU: 17.20 -> 24.82 | Hurt |
| Add the semantic-similarity scorer to consistency scoring (ablation) | ERAP-A · Recasted SICK test | iBLEU: 12.29 -> 20.24 | Helped |
| Add the semantic-similarity scorer to consistency scoring (ablation) | ERAP-A · Recasted SICK test | Diversity: 58.55 -> 42.05 | Hurt |
| Add the diversity scorer to consistency and similarity scoring (ablation) | ERAP-A · Recasted SICK test | Diversity: 42.05 -> 68.41 | Helped |
| Add the hypothesis-only adversary (ablation) | ERAP-A · Recasted SICK qualitative evaluation | heuristic outputs | Helped |
Closest papers by meaning
- Integrating Linguistic Knowledge to Sentence Paraphrase Generation (AAAI 2020 · similarity 0.88)
- A Task in a Suit and a Tie: Paraphrase Generation with Semantic Augmentation (AAAI 2019 · similarity 0.86)
- Idiomatic Expression Paraphrasing without Strong Supervision (AAAI 2022 · similarity 0.86)
- A Deep Generative Framework for Paraphrase Generation (AAAI 2018 · similarity 0.85)
- Novelty Controlled Paraphrase Generation with Retrieval Augmented Conditional Prompt Tuning (AAAI 2022 · similarity 0.85)
- Latent Paraphrasing: Perturbation on Layers Improves Knowledge Injection in Language Models (NeurIPS 2024 · 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.