Research map / Large language models / Retrieval and knowledge
Same Content, Different Representations: A Controlled Study for Table QA
Yue Zhang, Seiji Maekawa, Nikita Bhutani
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 |
|---|---|---|---|
| represent table semi-structured (verbalized) instead of structured | XiYan · RePairTQA all splits | accuracy (%): 69.55 -> 24.08 | Hurt |
| represent table semi-structured instead of structured | LLM-NL2SQL · RePairTQA all splits | accuracy (%): 69.14 -> 38.65 | Hurt |
| represent table semi-structured instead of structured | GPT-4o · RePairTQA all splits | accuracy (%): 45.37 -> 41.93 | Hurt |
| represent table semi-structured instead of structured | Weaver · RePairTQA all splits | accuracy (%): 62.19 -> 57.70 | Hurt |
| use long tables instead of short | GPT-4o · RePairTQA | accuracy (%): 53.26 -> 27.98 | Hurt |
| use long tables instead of short | H-STAR · RePairTQA | accuracy (%): 53.59 -> 12.19 | Hurt |
| use noisy (incomplete) schema instead of clean | LLM-NL2SQL · RePairTQA lookup | accuracy (%): 74.74 -> 20.12 | Hurt |
| use noisy schema instead of clean | H-STAR · RePairTQA lookup | accuracy (%): 77.90 -> 84.91 | Improved |
| verbalize only query-related columns (ablation) | GPT-4o · RePairTQA S3 semi-structured | accuracy (%): 43.26 -> 46.45 | Improved |
| verbalize more columns (1 -> 6 random) (ablation) | GPT-4o · RePairTQA S3 semi-structured | accuracy (%): 44.68 -> 42.20 | Hurt |
| widen tables (narrow to wide) | GPT-4o · BIRD-based splits structured | accuracy (%): 52.75 -> 27.91 | Hurt |
| replace GPT-4o with Gemini-2.5-Pro | LLM · RePairTQA S1 | structured vs semi-structured gap: 92.63 -> 88.42 | No clear effect |
Closest papers by meaning
- TableBench: A Comprehensive and Complex Benchmark for Table Question Answering (AAAI 2025 · similarity 0.90)
- Text2Analysis: A Benchmark of Table Question Answering with Advanced Data Analysis and Unclear Queries (AAAI 2024 · similarity 0.88)
- CABINET: Content Relevance-based Noise Reduction for Table Question Answering (ICLR 2024 · similarity 0.88)
- TrustUQA: A Trustful Framework for Unified Structured Data Question Answering (AAAI 2025 · similarity 0.88)
- MMTU: A Massive Multi-Task Table Understanding and Reasoning Benchmark (NeurIPS 2025 · similarity 0.87)
- Chain-of-Table: Evolving Tables in the Reasoning Chain for Table Understanding (ICLR 2024 · 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.
