Research map / Architectures and efficiency / RNNs and normalization
PARABANK: Monolingual Bitext Generation and Sentential Paraphrasing via Lexically-Constrained Neural Machine Translation
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 |
|---|---|---|---|
| Use the unconstrained ParaBank decoding setup rather than the ParaNMT baseline | Czech-English paraphrase generation · CzEng, average reference lengths | semantic similarity: 73.39 -> 79.63* | Helped |
| Negatively constrain the 3rd-lowest-IDF token (ablation) | ParaBank Czech-English generation · CzEng, average reference lengths | lexical diversity (modified BLEU): 27.48 -> 26.88 | Helped |
| Negatively constrain the 3rd-lowest-IDF token (ablation) | ParaBank Czech-English generation · CzEng, average reference lengths | semantic similarity: 73.39 -> 76.07 | Helped |
| Negatively constrain the 3 lowest-IDF tokens (ablation) | ParaBank Czech-English generation · CzEng, average reference lengths | semantic similarity: 79.63 -> 66.74 | Hurt |
| Negatively constrain the 3 lowest-IDF tokens (ablation) | ParaBank Czech-English generation · CzEng, average reference lengths | lexical diversity (modified BLEU): 27.48 -> 14.70* | Helped |
| Require the 3rd-highest-IDF token (ablation) | ParaBank Czech-English generation · CzEng, average reference lengths | semantic similarity: 79.63 -> 78.81 | No clear effect |
| Negatively constrain the first 3 tokens (ablation) | ParaBank Czech-English generation · CzEng paraphrases | meaningful and grammatical (%): 73.25 -> 82.50* | Helped |
| Use PPDB reverse-entailment lexical constraints (ablation) | ParaBank French-English generation · Giga paraphrases | meaningful and grammatical (%): 80.00* -> 52.46 | Hurt |
| Train an English-English rewriting model on ParaBank and decode without constraints instead of bilingual generation | ParaBank System 28 · CzEng monolingual rewriting, 100-sentence sample | semantic similarity: 81.91 (std. 3.11) -> 86.25 (std. 3.94) | Helped |
| Train an English-English rewriting model on ParaBank and decode without constraints instead of bilingual generation | ParaBank System 28 · CzEng monolingual rewriting, 100-sentence sample | lexical diversity (modified BLEU): 33.90 (std. 11.58) -> 44.36 (std. 13.69) | Hurt |
Closest papers by meaning
- Monolingual Transfer Learning via Bilingual Translators for Style-Sensitive Paraphrase Generation (AAAI 2020 · similarity 0.88)
- Integrating Linguistic Knowledge to Sentence Paraphrase Generation (AAAI 2020 · similarity 0.87)
- LAMPAT: Low-Rank Adaption for Multilingual Paraphrasing Using Adversarial Training (AAAI 2024 · similarity 0.87)
- Paraphrase Diversification Using Counterfactual Debiasing (AAAI 2019 · similarity 0.87)
- A Paradigm Shift in Machine Translation: Boosting Translation Performance of Large Language Models (ICLR 2024 · similarity 0.84)
- Seeking Neural Nuggets: Knowledge Transfer in Large Language Models from a Parametric Perspective (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.