---
title: "Machines that read tables and charts (research idea)"
description: "Document understanding, table question answering and chart understanding appear on 17 recent papers, all in AI journals, with 7 in the last quarter against about 3 per quarter before. Financial information lives in tables, footnotes and charts that models still misread. How accurate are current models on real filings, and which errors would mislead an analyst?"
url: https://scholar4ai.com/ideas/reading-tables-and-charts
markdown_url: https://scholar4ai.com/ideas/reading-tables-and-charts.md
license: CC BY-NC 4.0 (https://creativecommons.org/licenses/by-nc/4.0/)
share_text: "Machines that read tables and charts (research idea): Document understanding, table question answering and chart understanding appear on 17 recent papers, all in AI journals, with 7 in the last quarter against about 3 per quarter before."
date: "2026-10-23"
updated: "2026-10-23"
status: "seed"
tags: ["ai", "finance", "documents"]
keywords: ["document understanding", "table question answering", "chart understanding"]
---

# Machines that read tables and charts

Document understanding, table question answering and chart understanding appear on 17 recent papers, all in AI journals, with 7 in the last quarter against about 3 per quarter before. Financial information lives in tables, footnotes and charts that models still misread. How accurate are current models on real filings, and which errors would mislead an analyst?

## What prompted this

Since October 2025, 17 papers carry a document-understanding, table-question-answering, chart-understanding, document-layout or visually-rich-document keyword, all in AI journals: "document understanding" on 5, "table question answering" on 4, "document image analysis" on 3, "chart understanding" on 2. Seven were published in July–September 2026 against about 3 per quarter earlier.

## Questions

- On questions that require combining a table, a footnote and a chart from the same annual report, how often do multimodal models get the number right, and how often do they confidently get it wrong?
- Which document features cause errors (merged cells, units in headers, negative numbers in parentheses, restated columns)?
- Would a layer that extracts tables into structured data before the model reads them remove most of the errors?

## Data that could help

Annual reports with XBRL-tagged values as ground truth, public financial document question sets (FinQA, TAT-QA), and chart datasets built from earnings presentations.

## Next step

Build 500 questions whose answers are XBRL-tagged values in annual reports and measure the accuracy of three multimodal models on the PDF pages.

## Related research activity

*The keywords this note follows, in the tracked finance, accounting, economics, management, operations research and AI journals as of 2026-10-26*

| Keyword | Papers, last 30 days | Rank | Papers per month, Oct 23 – Oct 26* | Papers, last 36 months |
|---|---:|---:|---|---:|
| document understanding | 0 | – | 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 1, 0, 0, 1, 0, 1, 0, 0, 1, 1, 1, 1, 0, 0, 0, 2, 0, 1, 0 | 12 |
| table question answering | 0 | – | 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 1, 0, 0, 1, 1, 1, 0, 0 | 7 |
| chart understanding | 0 | – | 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 1, 0, 0 | 3 |

## Cite this page

Scholar4AI. (2026). Machines that read tables and charts. Journal Research Rankings. Retrieved October 23, 2026, from https://scholar4ai.com/ideas/reading-tables-and-charts

```bibtex
@misc{scholar4ai-ideas-reading-tables-and-charts,
  author = {{Scholar4AI}},
  title = {Machines that read tables and charts},
  year = {2026},
  howpublished = {\url{https://scholar4ai.com/ideas/reading-tables-and-charts}},
  urldate = {2026-10-23},
  note = {Journal Research Rankings; data from OpenAlex (CC0); licensed CC BY-NC 4.0}
}
```

License: CC BY-NC 4.0 (https://creativecommons.org/licenses/by-nc/4.0/).