Research map / Vision-language and multimodal / Human and embodied modeling
GraPPa: Grammar-Augmented Pre-Training for Table Semantic 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 |
|---|---|---|---|
| pre-train with MLM+SSP instead of MLM alone (ablation) | Grappa encoder with the task-specific parser · Spider | exact set match accuracy: 71.1(+1.4) -> 73.4(+3.7) dev; - -> 69.6(+4.0) test | Helped |
| pre-train with MLM+SSP instead of SSP alone (ablation) | Grappa encoder with the task-specific parser · Spider | exact set match accuracy: 73.6(+3.9) -> 73.4(+3.7) dev; 67.7(+2.1) -> 69.6(+4.0) test | Mixed |
| pre-train with MLM+SSP instead of MLM alone (ablation) | Grappa encoder with the task-specific parser · Fully supervised WikiSQL | execution accuracy: 91.4 -> 91.2 dev; 90.7 -> 90.8 test | No clear effect |
| pre-train with MLM+SSP instead of SSP alone (ablation) | Grappa encoder with the task-specific parser · Fully supervised WikiSQL | execution accuracy: 91.2 -> 91.2 dev; 90.7 -> 90.8 test | No clear effect |
| pre-train with MLM+SSP instead of MLM alone (ablation) | Grappa encoder with the task-specific parser · WikiTableQuestions | accuracy: 51.5(+7.8) -> 51.9(+8.2) dev; 51.7(+7.2) -> 52.7(+8.2) test | Helped |
| pre-train with MLM+SSP instead of SSP alone (ablation) | Grappa encoder with the task-specific parser · WikiTableQuestions | accuracy: 51.2(+7.5) -> 51.9(+8.2) dev; 51.1(+6.6) -> 52.7(+8.2) test | Helped |
| pre-train with MLM+SSP instead of MLM alone (ablation) | Grappa encoder with the task-specific parser · Weakly supervised WikiSQL | execution accuracy: 83.3 (+3.9) -> 85.9 (+6.5) dev; 83.5 (+4.2) -> 84.7 (+5.4) test | Helped |
| pre-train with MLM+SSP instead of SSP alone (ablation) | Grappa encoder with the task-specific parser · Weakly supervised WikiSQL | execution accuracy: 83.5(+4.1) -> 85.9 (+6.5) dev; 83.7 (+4.4) -> 84.7 (+5.4) test | Helped |
| add MLM regularization on human table-language examples to synthetic SSP pre-training (ablation) | Grappa · table semantic parsing downstream tasks | performance | Helped |
| extend WikiSQL training to 10k examples using Grappa (MLM+SSP) instead of RoBERTa (ablation) | SQLova · Fully supervised WikiSQL 10k | execution accuracy: 79.6 -> 82.3 dev; 79.2 -> 82.2 test | Helped |
| use Grappa (MLM+SSP) instead of RoBERTa with 10% of training data (ablation) | WikiTableQuestions parser · WikiTableQuestions 10% | accuracy: 37.3 -> 40.4 dev; 38.1 -> 42.0 test | Helped |
| continue pre-training for longer rather than stopping early (ablation) | Grappa · Spider and WikiTableQuestions | downstream performance | Hurt |
Closest papers by meaning
- NeedleInATable: Exploring Long-Context Capability of Large Language Models towards Long-Structured Tables (NeurIPS 2025 · similarity 0.87)
- HeGTa: Leveraging Heterogeneous Graph-enhanced Large Language Models for Few-shot Complex Table Understanding (AAAI 2025 · similarity 0.86)
- Table as a Modality for Large Language Models (NeurIPS 2025 · similarity 0.85)
- Filling Memory Gaps: Enhancing Continual Semantic Parsing via SQL Syntax Variance-Guided LLMs Without Real Data Replay (AAAI 2025 · similarity 0.85)
- TableBench: A Comprehensive and Complex Benchmark for Table Question Answering (AAAI 2025 · similarity 0.85)
- Improving Unsupervised Constituency Parsing via Maximizing Semantic Information (ICLR 2025 · 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.