Research map / Large language models / Language, safety and interpretability
Simplify-Then-Translate: Automatic Preprocessing for Black-Box Translation
Sneha Mehta, Bahareh Azarnoush, Boris Chen, Avneesh Saluja, Vinith Misra, Ballav Bihani, Ritwik Kumar
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 |
|---|---|---|---|
| Preprocess source with the in-domain FIGS APP model | Google Translate · FIGS low-resource translation: En-Hu, En-Uk, En-Cs, En-Ro, En-Bg, En-Hi, En-Ms | BLEU: 17.69 -> 18.92; 17.57 -> 18.18; 21.62 -> 22.35; 26.08 -> 27.98; 14.9 -> 16.63; 14.45 -> 15.53; 19.14 -> 20.37 | Improved |
| Preprocess source with the WikiLarge-trained simplification model instead of the in-domain model | Google Translate · FIGS low-resource translation | BLEU: 17.69 -> 11.86; 17.57 -> 12.51; 21.62 -> 16.71; 26.08 -> 21.56; 14.9 -> 12.39; 14.45 -> 11.39; 19.14 -> 13.00 | Hurt |
| Preprocess source with En-Es-trained APP | Google Translate · Open Subtitles En-Ca and En-Pt | BLEU: 27.25 -> 27.84; 25.11 -> 25.5 | Improved |
| Preprocess source with En-Es-trained APP | Google Translate · Open Subtitles En-Bg | BLEU: 23.71 -> 24.68 | Improved |
| Preprocess source with En-Es-trained APP | Google Translate · Open Subtitles En-Hu and En-Ro | BLEU: 7.05 -> 6.28; 26.18 -> 25.03 | Hurt |
| Preprocess source with En-Es-trained APP | Google Translate · Open Subtitles En-Uk | BLEU: 11.73 -> 11.77 | No clear effect |
| Use En-Es rather than En-Ru to generate the APP training corpus | APP simplification model · Turk | SARI: 23.2 -> 26.7 | Improved |
| Use En-Es rather than En-Ru to generate the APP training corpus | APP simplification model · PWKP test set | BLEU: 23.9 -> 32.7 | Improved |
| Use a more easily translated En-Es pair rather than En-Ru for APP corpus generation | APP simplification model · Open Subtitles En-Ca, En-Pt, En-Ro, En-Uk, En-Bg | BLEU: En-Ca: 23.36 -> 27.84; En-Pt: 22.28 -> 25.5; En-Ro: 22.40 -> 25.03; En-Uk: 11.61 -> 11.77; En-Bg: 23.90 -> 24.68 | Mixed |
| Simplify source with APP before translation | Google Translate · FIGS test set | TER: 6.9%, 6.9% and 7.2% reduction for Hungarian, Romanian and Bulgarian | Improved |
| Simplify source with APP before translation | Google Translate · Human evaluation: En-Hu, En-Uk, En-Cs, En-Ro, En-Bg | mean human rating: 2.52 -> 3.11; 3.02 -> 3.61; 2.43 -> 3.12; 2.56 -> 3.0; 2.34 -> 3.33 | Improved |
Closest papers by meaning
- SimulPL: Aligning Human Preferences in Simultaneous Machine Translation (ICLR 2025 · similarity 0.85)
- A Paradigm Shift in Machine Translation: Boosting Translation Performance of Large Language Models (ICLR 2024 · similarity 0.85)
- Exploring the Translation Mechanism of Large Language Models (NeurIPS 2025 · similarity 0.85)
- MT-Ranker: Reference-free machine translation evaluation by inter-system ranking (ICLR 2024 · similarity 0.84)
- LAMPAT: Low-Rank Adaption for Multilingual Paraphrasing Using Adversarial Training (AAAI 2024 · similarity 0.84)
- A Benchmark for Learning to Translate a New Language from One Grammar Book (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.
