QUIETT: Query-Independent Table Transformation for Robust Reasoning

TL;DR

QuIeTT addresses the challenge of irregular real-world tables for table-based reasoning by introducing a query-independent preprocessing framework that converts a raw table $T_{raw}$ into a single, lossless canonical representation $T_C$ suitable for SQL-like queries. The method decouples table normalization from downstream reasoning by generating a reproducible transformation plan and executing it deterministically to produce $T_C$, which can be reused across queries and models. Empirical results across WikiTQ, NQ-Table, SequentialQA, and HiTab show consistent improvements across model families and prompting strategies, with especially strong gains on structurally diverse and unseen queries. This work demonstrates that a robust, reusable table representation significantly enhances QA robustness and efficiency, enabling smaller models to perform competitively when provided with a clean, canonical table input.

Abstract

Real-world tables often exhibit irregular schemas, heterogeneous value formats, and implicit relational structure, which degrade the reliability of downstream table reasoning and question answering. Most existing approaches address these issues in a query-dependent manner, entangling table cleanup with reasoning and thus limiting generalization. We introduce QuIeTT, a query-independent table transformation framework that preprocesses raw tables into a single SQL-ready canonical representation before any test-time queries are observed. QuIeTT performs lossless schema and value normalization, exposes implicit relations, and preserves full provenance via raw table snapshots. By decoupling table transformation from reasoning, QuIeTT enables cleaner, more reliable, and highly efficient querying without modifying downstream models. Experiments on four benchmarks, WikiTQ, HiTab, NQ-Table, and SequentialQA show consistent gains across models and reasoning paradigms, with particularly strong improvements on a challenge set of structurally diverse, unseen questions.

Figures (3)

• Figure 1: Overview of the QuIeTT pipeline.
• Figure 2: QuIeTT framework.
• Figure 3: Table-level performance outcomes after applying QuIeTT. Each table is categorized as a win (improved), no change, or loss (degraded) based on average downstream performance. Results are shown separately for web tables and hierarchical tables.