Research map / Architectures and efficiency / RNNs and normalization
Neural Machine Translation with Byte-Level Subwords
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 a 2K BBPE vocabulary instead of 32K BPE | T_big En-De translation · WMT14 En-De test | BLEU: 28.36 -> 28.78 | No clear effect |
| Use a 2K BBPE vocabulary instead of 32K BPE | T_base En-De translation · WMT14 En-De test | BLEU: 27.31 -> 26.98 | Hurt |
| Use 4K BBPE instead of 16K BPE | T_big Japanese-English translation · Ja-En KFTT test | BLEU: 24.08 -> 23.88 | Hurt |
| Use 4K BBPE instead of 16K BPE | T_big Japanese-English translation · Ja-En TED test | BLEU: 18.34 -> 19.0 | Helped |
| Use 4K BBPE instead of 16K BPE | T_big Japanese-English translation · Ja-En JESC test | BLEU: 17.89 -> 17.93 | No clear effect |
| Use 4K BBPE instead of 16K BPE | T_big Japanese-English translation · Ja-En All test | BLEU: 19.14 -> 19.58 | Helped |
| Use 2K BBPE instead of 16K BPE | T_base many-to-English translation · X-En All test | BLEU: 29.62 -> 29.91 | Helped |
| Use BBPE instead of character tokens | T_base many-to-English translation · X-En All test | BLEU: 30.75 -> 29.91 | Hurt |
| Add embedding contextualization with a BiGRU (ablation) | BBPE on X-En · X-En validation | BLEU: over 4% | Helped |
| Transfer the X-En BBPE model and finetune embeddings and encoder | T_flores BBPE 4K+ · Si-En test | BLEU: 7.1 -> 9.0; 7.2 -> 9.0 | Helped |
| Finetune the transferred model's encoder only rather than embeddings and encoder (ablation) | T_flores BBPE 4K+ transfer · Si-En test | BLEU: 9.0 -> 8.3 | Hurt |
| Finetune encoder and decoder rather than embeddings and encoder (ablation) | T_flores BBPE 4K+ transfer · Si-En test | BLEU: 9.0 -> 8.1 | Hurt |
Closest papers by meaning
- Multilingual Neural Machine Translation With Soft Decoupled Encoding (ICLR 2019 · similarity 0.87)
- Neural Machine Translation with Joint Representation (AAAI 2020 · similarity 0.85)
- From Fully Trained to Fully Random Embeddings: Improving Neural Machine Translation with Compact Word Embedding Tables (AAAI 2022 · similarity 0.85)
- Unsupervised Neural Machine Translation (ICLR 2018 · similarity 0.84)
- Evaluating the Cross-Lingual Effectiveness of Massively Multilingual Neural Machine Translation (AAAI 2020 · similarity 0.84)
- MMTEB: Massive Multilingual Text Embedding Benchmark (ICLR 2025 · similarity 0.83)
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.