Research map / Label-efficient and robust learning / Recognition and long-tail
Translating Pro-Drop Languages With Reconstruction Models
Longyue Wang, Zhaopeng Tu, Shuming Shi, Tong Zhang, Yvette Graham, Qun Liu
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 |
|---|---|---|---|
| Add encoder-side and decoder-side reconstructors of DP-labelled source | NMT with DP-labelled training data · Chinese–English dialogue | BLEU: 32.67 -> 35.08 | Improved |
| Add encoder-side reconstruction | NMT with DP-labelled training data · Chinese–English dialogue | BLEU: 32.67 -> 33.67 | Improved |
| Add decoder-side reconstruction | NMT with DP-labelled training data · Chinese–English dialogue | BLEU: 32.67 -> 33.48 | Improved |
| Train on explicitly DP-labelled source sentences | NMT baseline · Chinese–English dialogue | BLEU: 31.80 -> 32.67 | Improved |
| Add encoder-side and decoder-side DP reconstruction | NMT with DP-labelled training data · Japanese–English | BLEU: 20.55 -> 21.84 | Improved |
| Do not use reconstruction during testing; retain it during training (ablation) | DP reconstruction with encoder and decoder reconstructors · Chinese–English dialogue | BLEU: 32.67 -> 34.02 | Improved |
| Reconstruct the original source instead of the DP-labelled source (ablation) | NMT reconstruction with encoder and decoder reconstructors · Chinese–English dialogue | BLEU: 35.08 -> 33.25 | Hurt |
| Use manual rather than automatic DP labels at test time (ablation) | DP reconstruction with encoder and decoder reconstructors · Chinese–English dialogue | BLEU: 35.08 -> 38.38 | Improved |
| Increase the hidden layer size to 2100 without adding DP reconstruction (ablation) | Baseline (+DPs) · Chinese–English dialogue | BLEU: 32.67 -> 32.00 | No clear effect |
| Use a three-layer encoder and decoder instead of reconstruction (ablation) | Baseline (+DPs) · Chinese–English dialogue | BLEU: 33.36 vs. 35.08 | Hurt |
Closest papers by meaning
- Neural Machine Translation with Reconstruction (AAAI 2017 · similarity 0.87)
- Dual Transfer Learning for Neural Machine Translation with Marginal Distribution Regularization (AAAI 2018 · similarity 0.87)
- Joint Training for Neural Machine Translation Models with Monolingual Data (AAAI 2018 · similarity 0.87)
- Improved Neural Machine Translation with SMT Features (AAAI 2016 · similarity 0.87)
- Neural Machine Translation with Joint Representation (AAAI 2020 · similarity 0.87)
- Evaluating the Cross-Lingual Effectiveness of Massively Multilingual Neural Machine Translation (AAAI 2020 · similarity 0.87)
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.
