Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Evidence-Guided Schema Normalization for Temporal Tabular Reasoning

Ashish Thanga, Vibhu Dixit, Abhilash Shankarampeta, Vivek Gupta

TL;DR

Temporal reasoning over evolving semi-structured tables is hard for LLMs. The authors recast the problem as text-to-SQL by generating normalized 3NF schemas from infobox timelines, populating a database, and then generating SQL queries guided by the schema. They show that schema quality can outweigh model capacity, achieving 80.39 EM on TransientTables with a Gemini-based config and demonstrating strong cross-model portability. The work highlights the value of hybrid symbolic-systems approaches, where principled schema design and executable queries unlock robust temporal QA beyond raw JSON-table reasoning.

Abstract

Temporal reasoning over evolving semi-structured tables poses a challenge to current QA systems. We propose a SQL-based approach that involves (1) generating a 3NF schema from Wikipedia infoboxes, (2) generating SQL queries, and (3) query execution. Our central finding challenges model scaling assumptions: the quality of schema design has a greater impact on QA precision than model capacity. We establish three evidence-based principles: normalization that preserves context, semantic naming that reduces ambiguity, and consistent temporal anchoring. Our best configuration (Gemini 2.5 Flash schema + Gemini-2.0-Flash queries) achieves 80.39 EM, a 16.8\% improvement over the baseline (68.89 EM).

Evidence-Guided Schema Normalization for Temporal Tabular Reasoning

TL;DR

Temporal reasoning over evolving semi-structured tables is hard for LLMs. The authors recast the problem as text-to-SQL by generating normalized 3NF schemas from infobox timelines, populating a database, and then generating SQL queries guided by the schema. They show that schema quality can outweigh model capacity, achieving 80.39 EM on TransientTables with a Gemini-based config and demonstrating strong cross-model portability. The work highlights the value of hybrid symbolic-systems approaches, where principled schema design and executable queries unlock robust temporal QA beyond raw JSON-table reasoning.

Abstract

Temporal reasoning over evolving semi-structured tables poses a challenge to current QA systems. We propose a SQL-based approach that involves (1) generating a 3NF schema from Wikipedia infoboxes, (2) generating SQL queries, and (3) query execution. Our central finding challenges model scaling assumptions: the quality of schema design has a greater impact on QA precision than model capacity. We establish three evidence-based principles: normalization that preserves context, semantic naming that reduces ambiguity, and consistent temporal anchoring. Our best configuration (Gemini 2.5 Flash schema + Gemini-2.0-Flash queries) achieves 80.39 EM, a 16.8\% improvement over the baseline (68.89 EM).

Paper Structure

This paper contains 82 sections, 1 figure, 8 tables.

Table of Contents

  1. Introduction
  2. The TransientTables task:
  3. Methodology
  4. Normalized Schema Generation
  5. Schema Tables (database) population
  6. Schema-Guided SQL Generation
  7. Few-shot Examples Construction.
  8. Domain-specific Prompting.
  9. Query Generation and Evaluation.
  10. Experiments
  11. Normalized Schema Generation LLMs:
  12. SQL Query Generation LLMs:
  13. Automatic Schema Tables Population:
  14. Few-Shot Examples:
  15. Results and Analysis
  16. ...and 67 more sections

Figures (1)

  • Figure 1: Our three-stage pipeline: (1) Schema Generation: convert evolving infoboxes into normalized 3NF database schemas. (2) Schema Population: parse and clean JSON data to populate SQLite databases. (3) SQL Generation & Execution: Schema-guided prompting generates SQL queries, which are executed to produce answers.