Research map / Architectures and efficiency / RNNs and normalization
Neural Machine Translation with Reconstruction
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 |
|---|---|---|---|
| Add reconstruction objective and reconstruction-based reranking | RNNSearch · All test sets | BLEU: 30.65 -> 31.73 | Helped |
| Add reconstruction objective and reconstruction-based reranking | RNNSearch · All test sets, Beam=100 | BLEU: 27.97 -> 32.94 | Helped |
| Use reconstruction score to rerank a larger candidate set | RNNSearch+Reconstruction · Validation set, Beam=10 to Beam=1000 | BLEU: 36.34; 37.35; 37.77 | Helped |
| Increase beam size with likelihood-only decoding | RNNSearch · Validation set, Beam=10 to Beam=1000 | BLEU: 35.46; 25.80; 1.38 | Hurt |
| Use reconstruction during training without reconstruction reranking at testing (ablation) | RNNSearch · All test sets, Beam=10 | BLEU: 30.65 -> 31.17 | Helped |
| Add reconstruction-based reranking at testing after reconstruction training (ablation) | RNNSearch · All test sets, Beam=10 | BLEU: 31.17 -> 31.73 | Helped |
| Use reconstruction during training without reconstruction reranking at testing (ablation) | RNNSearch · All test sets, Beam=100 | BLEU: 27.97 -> 31.51 | Helped |
| Add reconstruction-based reranking at testing after reconstruction training (ablation) | RNNSearch · All test sets, Beam=100 | BLEU: 31.51 -> 32.94 | Helped |
| Add reconstruction objective | RNNSearch · 200 sampled test sentences | Under-translation source-word percentage: 18.2% -> 16.2% | Helped |
| Add reconstruction objective | RNNSearch · 200 sampled test sentences | Over-translation source-word percentage: 3.9% -> 2.4% | Helped |
| Add reconstruction to coverage-enhanced RNNSearch | RNNSearch+Coverage · All test sets | BLEU: 31.89 -> 33.44 (+1.6) | Helped |
| Add reconstruction to context-gated RNNSearch | RNNSearch+Context gate · All test sets | BLEU: 32.05 -> 33.51 (+1.5) | Helped |
Closest papers by meaning
- Improved Neural Machine Translation with SMT Features (AAAI 2016 · similarity 0.90)
- Neural Machine Translation Advised by Statistical Machine Translation (AAAI 2017 · similarity 0.90)
- Explicit Sentence Compression for Neural Machine Translation (AAAI 2020 · similarity 0.89)
- Asynchronous Bidirectional Decoding for Neural Machine Translation (AAAI 2018 · similarity 0.88)
- Unsupervised Neural Machine Translation (ICLR 2018 · similarity 0.88)
- Tied Transformers: Neural Machine Translation with Shared Encoder and Decoder (AAAI 2019 · 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.