How Do Language Models Understand Tables? A Mechanistic Analysis of Cell Location

• URL: http://arxiv.org/abs/2602.08548v1
• Date: Mon, 09 Feb 2026 11:47:34 GMT
• Title: How Do Language Models Understand Tables? A Mechanistic Analysis of Cell Location
• Authors: Xuanliang Zhang, Dingzirui Wang, Keyan Xu, Qingfu Zhu, Wanxiang Che,
• Abstract summary: We investigate the process of table understanding by dissecting the atomic task of cell location.<n>We demonstrate that models locate the target cell via an ordinal mechanism that counts discretes to resolve coordinates.<n>We reveal that models generalize to multi-cell location tasks by multiplexing the identical attention heads identified during atomic location.
• Score: 53.68149869349268
• License: http://creativecommons.org/licenses/by-nc-nd/4.0/
• Abstract: While Large Language Models (LLMs) are increasingly deployed for table-related tasks, the internal mechanisms enabling them to process linearized two-dimensional structured tables remain opaque. In this work, we investigate the process of table understanding by dissecting the atomic task of cell location. Through activation patching and complementary interpretability techniques, we delineate the table understanding mechanism into a sequential three-stage pipeline: Semantic Binding, Coordinate Localization, and Information Extraction. We demonstrate that models locate the target cell via an ordinal mechanism that counts discrete delimiters to resolve coordinates. Furthermore, column indices are encoded within a linear subspace that allows for precise steering of model focus through vector arithmetic. Finally, we reveal that models generalize to multi-cell location tasks by multiplexing the identical attention heads identified during atomic location. Our findings provide a comprehensive explanation of table understanding within Transformer architectures.

Related papers

• TableZoomer: A Collaborative Agent Framework for Large-scale Table Question Answering [26.00027389659854]
  TableZoomer is a programming-based agent framework for the table question answering (TQA) task.<n>It introduces three key innovations: (1) replacing the original fully verbalized table with structured table schema to bridge the semantic gap and reduce computational complexity; (2) a query-aware table zooming mechanism that dynamically generates sub-table schema through column selection and entity linking; and (3) a Program-of-Thoughts (PoT) strategy that transforms queries into executable code to mitigate numerical hallucination.
  arXiv  Detail & Related papers  (2025-09-01T09:53:01Z)
• TableRAG: Million-Token Table Understanding with Language Models [53.039560091592215]
  TableRAG is a Retrieval-Augmented Generation (RAG) framework specifically designed for LM-based table understanding.<n>TableRAG leverages query expansion combined with schema and cell retrieval to pinpoint crucial information before providing it to the LMs.<n>Our results demonstrate that TableRAG achieves the highest retrieval quality, leading to the new state-of-the-art performance on large-scale table understanding.
  arXiv  Detail & Related papers  (2024-10-07T04:15:02Z)
• Talking Heads: Understanding Inter-layer Communication in Transformer Language Models [32.2976613483151]
  We analyze a mechanism used in two LMs to selectively inhibit items in a context in one task.<n>We find that models write into low-rank subspaces of the residual stream to represent features which are then read out by later layers.
  arXiv  Detail & Related papers  (2024-06-13T18:12:01Z)
• SEMv2: Table Separation Line Detection Based on Instance Segmentation [96.36188168694781]
  We propose an accurate table structure recognizer, termed SEMv2 (SEM: Split, Embed and Merge) We address the table separation line instance-level discrimination problem and introduce a table separation line detection strategy based on conditional convolution. To comprehensively evaluate the SEMv2, we also present a more challenging dataset for table structure recognition, dubbed iFLYTAB.
  arXiv  Detail & Related papers  (2023-03-08T05:15:01Z)
• TRUST: An Accurate and End-to-End Table structure Recognizer Using Splitting-based Transformers [56.56591337457137]
  We propose an accurate and end-to-end transformer-based table structure recognition method, referred to as TRUST. Transformers are suitable for table structure recognition because of their global computations, perfect memory, and parallel computation. We conduct experiments on several popular benchmarks including PubTabNet and SynthTable, our method achieves new state-of-the-art results.
  arXiv  Detail & Related papers  (2022-08-31T08:33:36Z)
• Split, embed and merge: An accurate table structure recognizer [42.579215135672094]
  We introduce Split, Embed and Merge (SEM) as an accurate table structure recognizer. SEM can achieve an average F-Measure of $96.9%$ on the SciTSR dataset.
  arXiv  Detail & Related papers  (2021-07-12T06:26:19Z)
• TGRNet: A Table Graph Reconstruction Network for Table Structure Recognition [76.06530816349763]
  We propose an end-to-end trainable table graph reconstruction network (TGRNet) for table structure recognition. Specifically, the proposed method has two main branches, a cell detection branch and a cell logical location branch, to jointly predict the spatial location and the logical location of different cells.
  arXiv  Detail & Related papers  (2021-06-20T01:57:05Z)
• TCN: Table Convolutional Network for Web Table Interpretation [52.32515851633981]
  We propose a novel table representation learning approach considering both the intra- and inter-table contextual information. Our method can outperform competitive baselines by +4.8% of F1 for column type prediction and by +4.1% of F1 for column pairwise relation prediction.
  arXiv  Detail & Related papers  (2021-02-17T02:18:10Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.