Lightweight markdown format balances table structure and simplicity for AI
LTEM: LLM-Friendly Table-Enhanced Markdown for Table Understanding and Generation
Computation and LanguageArtificial Intelligence
Summary
Tables in reports and papers often have complex layouts that simple formats like Markdown or CSV can't show well, while more detailed formats like HTML or LaTeX are complicated and costly for AI to process. The authors create a way to break down table complexity and build a new simple Markdown version called LTEM that can clearly show common table features. They tested LTEM with several advanced AI models and found it works almost as well as HTML but with fewer errors and less complexity. This helps AI understand and create tables better without extra overhead.
What this means in practice
- •For document processing teams: Use LTEM format to improve AI accuracy when extracting structured data from complex tables in documents without high computational cost.
- •For business intelligence teams: Generate tables with AI using LTEM to keep clear format and reduce errors compared to dense formats like LaTeX in data presentations.
Authors
Eunbin Kim, Hyeonseok Kang, Sangkeun Jung
Abstract
Tables in real-world documents, including scientific papers, financial reports, and statistical summaries, frequently use merged cells, multi-level headers, and sectioned layouts. Yet the lightweight formats LLMs handle best (Markdown, CSV) cannot express these structures, while expressive formats (HTML, LaTeX) impose excessive token cost and parsing complexity. To enable systematic analysis of this trade-off, we introduce a taxonomy decomposing table structure into six atomic features (F1-F6) and corresponding complexity levels (L1-L6), and construct STB (Synthetic Table Benchmark), a controlled benchmark with 1,089 tables systematically distributed across complexity levels. Using this framework, we propose LTEM (LLM-friendly Table-Enhanced Markdown), a formally specified, minimal extension of Markdown that explicitly encodes spans and hierarchical headers. Experiments across five state-of-the-art LLMs show that LTEM achieves accuracy comparable to HTML, offers a favorable accuracy-token trade-off relative to HTML and LaTeX, and produces substantially fewer invalid responses than LaTeX; for table generation, LTEM is comparable to MultiMarkdown though HTML is best. We release our taxonomy, benchmark, and format specification to support future research on LLM-table interaction.