Research map / Label-efficient and robust learning / Recognition and long-tail
Sequence-to-Sequence Learning with Latent Neural Grammars
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 phrase-level copy to Neural QCFG (ablation) | Neural QCFG · StylePTB hard active to passive | BLEU-4: 0.472 -> 0.662 | Helped |
| Add phrase-level copy to Neural QCFG (ablation) | Neural QCFG · StylePTB hard adjective emphasis | BLEU-4: 0.000 -> 0.316 | Helped |
| Add phrase-level copy to Neural QCFG (ablation) | Neural QCFG · StylePTB hard verb emphasis | BLEU-4: 0.073 -> 0.319 | Helped |
| Add a bidirectional LSTM to contextualize source embeddings (ablation) | Neural QCFG · English-French MT test set | BLEU: 23.5 -> 26.8 | Helped |
| Add a bidirectional LSTM to contextualize source embeddings (ablation) | Neural QCFG · English-French daxy test | accuracy: 100% -> 75.0% | Hurt |
| Constrain grammar rules so target hierarchy respects source hierarchy | Neural QCFG · SCAN compositional splits | accuracy | Helped |
| Manually disallow incorrect root-level splits during decoding | Neural QCFG · SCAN | accuracy: 1%-2% | Helped |
| Use neural QCFG instead of a well-tuned Transformer | English-French MT · English-French MT test set | BLEU: 30.4 vs 23.5 | Hurt |
Closest papers by meaning
- Learning to Recombine and Resample Data For Compositional Generalization (ICLR 2021 · similarity 0.89)
- Generalization without Systematicity: On the Compositional Skills of Sequence-to-Sequence Recurrent Networks (ICML 2018 · similarity 0.88)
- Structured Reordering for Modeling Latent Alignments in Sequence Transduction (NeurIPS 2021 · similarity 0.88)
- Improving Sequence-to-Sequence Learning via Optimal Transport (ICLR 2019 · similarity 0.87)
- The Missing Alignment Link of In-context Learning on Sequences (ICML 2025 · similarity 0.86)
- Neural-Symbolic Recursive Machine for Systematic Generalization (ICLR 2024 · similarity 0.86)
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.