Research map / Architectures and efficiency / Attention and transformers
Improving Sequence-to-Sequence Constituency Parsing
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 |
|---|---|---|---|
| deterministic attention on top-down linearization vs probabilistic attention (Vinyals), without POS/external embeddings | seq2seq parser · PTB section 23 | F1: 88.3 -> 90.1 | Helped |
| top-down instead of bottom-up linearization with deterministic attention (vs Ma et al.) | deterministic-attention parser · PTB section 23 | F1: 88.6 -> 90.1 | Helped |
| remove POS tags and pretrained word embeddings (ablation) | proposed parser · PTB section 23 | F1: 91.2 -> 90.1 | Hurt |
| ensemble of five models (ablation) | proposed parser · PTB section 23 | F1: 91.2 -> 92.3 | Helped |
| richer context function (DA1 phrase plus semi-phrase boundaries vs DA3 single word) (ablation) | proposed parser · PTB dev | F1: 90.3 -> 90.2 | No clear effect |
| increase beam size from 1 to 80 (ablation) | proposed parser · PTB dev | F1: 89.8 -> 90.3 | Helped |
| larger beam beyond 80 (ablation) | proposed parser · PTB dev | F1 | No clear effect |
| top-down deterministic attention vs fine-grained probabilistic attention (Liu and Zhang) | seq2seq parser · PTB section 23 | F1: 90.5 -> 91.2 | Helped |
| top-down vs bottom-up deterministic attention, coordination errors | parser · PTB error analysis | coordination errors: 379 -> 225 | Helped |
Closest papers by meaning
- Deterministic Attention for Sequence-to-Sequence Constituent Parsing (AAAI 2017 · similarity 0.91)
- Tree-Structured Attention with Hierarchical Accumulation (ICLR 2020 · similarity 0.88)
- Ensemble Distillation for Unsupervised Constituency Parsing (ICLR 2024 · similarity 0.87)
- Improving Unsupervised Constituency Parsing via Maximizing Semantic Information (ICLR 2025 · similarity 0.87)
- Dynamic Chunking for End-to-End Hierarchical Sequence Modeling (ICLR 2026 · similarity 0.86)
- ZETA: Leveraging $Z$-order Curves for Efficient Top-$k$ Attention (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.