Research map / Architectures and efficiency / RNNs and normalization
Scaling Laws for Multilingual 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 |
|---|---|---|---|
| Change language-pair sampling weights | Multilingual NMT models · High-resource language pairs across tested directions | scaling exponent and irreducible cross-entropy loss | No clear effect |
| Change language-pair sampling weights | Multilingual NMT models · High-resource language pairs across tested directions | multiplicative scaling-law factor β | Mixed |
| Use the joint scaling law with shared exponent and irreducible loss across weights | Multilingual NMT models · In-domain and out-of-domain translation test sets | cross-entropy loss prediction | Helped |
| Train on related rather than unrelated target languages | En→XX multilingual models · En→{De, Fr} versus En→{De, Zh} | effective capacity interaction | No clear effect |
| Translate multiple source languages into English rather than translating out of English | Two-task multilingual NMT models · {De, Zh}→En versus En→XX | effective capacity synergy | Helped |
| Assign a language pair 5% of training weight in a multilingual model rather than training a single-task model with 5% of the multilingual model's capacity | En→{De, Zh} multilingual model · German translation | effective parameters: more than 3× | Helped |
| Assign a language pair 5% of training weight in a multilingual model rather than training a single-task model with 5% of the multilingual model's capacity | {De, Zh}→En multilingual model · Translation into English | effective parameters: more than 6× | Helped |
| Estimate the effective-parameter fraction and use it with the single-language-pair scaling law | Multilingual NMT models · En→{De, Zh} performance trade-off frontier | frontier prediction accuracy | Helped |
| Use cross-entropy scaling observations to model generated translation quality | Multilingual NMT models · ChrF and BLEURT evaluation | translation-quality frontier prediction | Helped |
Closest papers by meaning
- A Hitchhiker's Guide to Scaling Law Estimation (ICML 2025 · similarity 0.90)
- When Scaling Meets LLM Finetuning: The Effect of Data, Model and Finetuning Method (ICLR 2024 · similarity 0.90)
- Language models scale reliably with over-training and on downstream tasks (ICLR 2025 · similarity 0.90)
- OWLS: Scaling Laws for Multilingual Speech Recognition and Translation Models (ICML 2025 · similarity 0.90)
- A Paradigm Shift in Machine Translation: Boosting Translation Performance of Large Language Models (ICLR 2024 · similarity 0.89)
- Scaling Laws for Robust Comparison of Open Foundation Language-Vision Models and Datasets (NeurIPS 2025 · similarity 0.89)
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.