Research map / Label-efficient and robust learning / Recognition and long-tail
Non-Monotonic Latent Alignments for CTC-Based Non-Autoregressive 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 |
|---|---|---|---|
| Fine-tune CTC with non-monotonic 2-gram matching (ablation) | CTC · WMT14 En-De test | BLEU: 26.34 -> 27.57 | Helped |
| Fine-tune CTC with non-monotonic 2-gram matching (ablation) | CTC · WMT14 En-De test | METEOR: 53.15 -> 54.50 | Helped |
| Fine-tune SCTC with non-monotonic matching objectives (ablation) | SCTC · WMT14 En-De test | BLEU: 25.26 -> 26.79 | Helped |
| Use 2-gram matching instead of 1-gram matching (ablation) | SCTC · WMT14 En-De test | BLEU: 26.22 -> 26.79 | Helped |
| Increase matching order from 2-gram to 3-gram or 4-gram (ablation) | SCTC · WMT14 En-De test | BLEU: 26.79 -> 26.66; 26.79 -> 26.62 | Hurt |
| Use CTC rather than SCTC pretraining (ablation) | 2-gram NMLA · WMT14 En-De test | BLEU: 26.79 -> 27.57 | Helped |
| Mix 1-gram and 2-gram matching instead of using only 2-gram matching (ablation) | CTC · WMT14 En-De validation | BLEU: 25.90 -> 25.74; 25.90 -> 25.66; 25.90 -> 25.51; 25.90 -> 25.42 | Hurt |
| Combine different n-gram objectives by arithmetic or geometric average (ablation) | CTC and SCTC · WMT14 En-De validation | BLEU: 24.94; 24.88 | Hurt |
| Add beam search and a 4-gram language model | CTC w/ NMLA · WMT14 En-De test | BLEU: 27.57 -> 28.35 | Helped |
| Fine-tune DDRS with NMLA | DDRS · WMT14 En-De test | BLEU: 27.18 -> 28.02 | Helped |
| Fine-tune DDRS with NMLA | DDRS · WMT16 En-Ro test | BLEU: 34.42 -> 34.73 | Helped |
| Fine-tune DDRS with NMLA | DDRS · WMT16 Ro-En test | BLEU: 34.31 -> 34.76 | Helped |
Closest papers by meaning
- Aligned Cross Entropy for Non-Autoregressive Machine Translation (ICML 2020 · similarity 0.90)
- AEQA-NAT : Adaptive End-to-end Quantization Alignment Training Framework for Non-autoregressive Machine Translation (ICML 2025 · similarity 0.90)
- Non-Autoregressive Neural Machine Translation with Enhanced Decoder Input (AAAI 2019 · similarity 0.89)
- Leveraging Importance Sampling to Detach Alignment Modules from Large Language Models (NeurIPS 2025 · similarity 0.86)
- Exploring Intrinsic Alignments Within Text Corpus (AAAI 2025 · similarity 0.86)
- Decoding-Time Language Model Alignment with Multiple Objectives (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.