Table Question Answering
Answering natural language questions over tables by selecting cells and aggregations or by generating SQL-like reasoning.
Tier Grades
Tier A: Frontier AI labs
feasible
Google and Microsoft published TAPAS and TAPEX.
Tier B: Indonesian AI labs
feasible
BERT-size table models are small; Indonesian table data is not in the sources.
Tier C: Institution units such as ITB
feasible
TAPAS came from a 5-author team.
Tier D: Organisations such as SEMAI (PoT, PoC, PoV)
feasible
Questions over client spreadsheets are a common proof-of-concept request; BERT-size TAPAS checkpoints are released.
Where this comes from
Tiers: a) frontier AI labs, b) Indonesian AI labs, c) institution units such as ITB, d) organisations such as SEMAI delivering proof of technology, concept and value. Each rationale cites the anchors below.
Anchors
- ComputeTAPAS extends BERT to encode tables; size and compute are not in the abstract.
- DataTAPAS: joint pretraining on text segments and tables crawled from Wikipedia, tested on three semantic parsing datasets including SQA.
- Team size (proxy)Proxy only (author count of an exemplar paper, not a headcount): 5 authors on TAPAS.
- Time horizonMonths to a year.
- Shelf-life signalsGeneral LLMs with tool use answering table questions through SQL, Table pretraining recipes (TAPAS, TAPEX) used as baselines
- Pipeline roleTier A publishes table pretraining; Tier C builds Indonesian table question sets; Tier D answers questions over client tables.
Where this comes from
Team size is a proxy (author counts of exemplar papers), not a headcount.
Who and What
Links
R&D TopicTable Question Answering
Models
TAPASGoogle, tabular
TAPEXMicrosoft AI, tabularTAPAS base fine-tuned on WTQGoogle, tabularTAPEX base fine-tuned on WikiSQLMicrosoft AI, tabular
TAPASGoogle, tabularTAPEXMicrosoft AI, tabularTAPAS base fine-tuned on WTQGoogle, tabularTAPEX base fine-tuned on WikiSQLMicrosoft AI, tabularUse Casesnone recorded
Who Runs Itnone recorded
Papers
- Compositional Semantic Parsing on Semi-Structured TablesPanupong Pasupat et al., arXiv preprint, 2015 founding
- Spider: A Large-Scale Human-Labeled Dataset for Complex and Cross-Domain Semantic Parsing and Text-to-SQL TaskTao Yu et al., EMNLP 2018 (per arXiv comment), 2018 founding
- Can LLM Already Serve as A Database Interface? A BIg Bench for Large-Scale Database Grounded Text-to-SQLsJinyang Li et al., NeurIPS 2023 (per arXiv comment), 2023 state of the art
- TableLlama: Towards Open Large Generalist Models for TablesTianshu Zhang et al., NAACL 2024 (per arXiv comment), 2023 state of the art
- A Survey of Table Reasoning with Large Language ModelsXuanliang Zhang et al., arXiv preprint, 2024 survey
Where this comes from
The papers to read first: where the direction began, its current best, and a survey when one exists.
Sources
- arxiv.org/abs/2004.02349 https://arxiv.org/abs/2004.02349
To cite this page: Project AURION, AI Use-Cases Research & Innovation, "Table Question Answering", https://aurion.semai.id/topics/table-question-answering/, records of October 2026.