Q&A on Any Spreadsheet: Why Cell Classification Hits a Ceiling and What Comes Next

Paper Overview

• Research area: Machine Learning
• Author: Zofia Smoleń
• Published: 2026-09-17
• arXiv: 2609.20732

Summary

In LLM-driven RAG systems, semantic cell annotation improves spreadsheet chunk interpretability and aids answer generation through enriched context rather than through improved retrieval accuracy. The paper proposes a novel framework for splitting any spreadsheet into interpretable chunks using cell role annotation.

Key Points

• The proposed cell-role-annotation chunking framework outperforms the current state of the art, but it faces a hard ceiling.
• Spreadsheets are fundamentally two-dimensional unstructured data: relationships are continuous, and potential cell roles are effectively infinite.
• Classification models are restricted to finite, pre-defined classes, so they cannot perfectly capture this structural nuance — even with human-level annotation.
• The author argues that solving the spreadsheet-to-LLM bottleneck requires moving beyond discrete cell classification.
• The field should instead develop dimensionality-reduction techniques that flatten two-dimensional unstructured spreadsheets directly into one-dimensional unstructured text; such text chunks are easier for downstream RAG systems to understand and generate from.

Original Abstract (excerpt)

> Semantic cell annotation improves chunking interpretability for spreadsheets in LLM-driven RAG systems, aiding answer generation through enriched context rather than improved retrieval accuracy. We propose a novel framework of splitting any spreadsheet into interpretable chunks using cell role annotation. Our framework beats the state of the art, yet it faces a hard ceiling. Spreadsheets are fundamentally two-dimensional unstructured data with continuous relationships and infinite potential cell roles. Because classification models are restricted to finite, pre-defined classes, they cannot perfectly capture this structural nuance, even with human-level annotation. We show that addressing the spreadsheet-to-LLM bottleneck requires moving beyond discrete cell classification. Instead, the fie...

--- *Auto-collected on 2026-09-20*