Abstract:Structured tabular data exist across nearly all fields. Reasoning task over these data aims to answer questions or determine the truthiness of hypothesis sentences by understanding the semantic meaning of a table. While previous works have devoted significant efforts to the tabular reasoning task, they always assume there are sufficient labeled data. However, constructing reasoning samples over tables (and related text) is labor-intensive, especially when the reasoning process is complex. When labeled data is insufficient, the performance of models will suffer an unendurable decline. In this paper, we propose a unified framework for unsupervised complex tabular reasoning (UCTR), which generates sufficient and diverse synthetic data with complex logic for tabular reasoning tasks, assuming no human-annotated data at all. Specifically, we first utilize a random sampling strategy to collect diverse programs of different types and execute them on tables based on a "Program-Executor" module. To bridge the gap between the programs and natural language sentences, we design a powerful "NL-Generator" module to generate natural language sentences with complex logic from these programs. Since a table often occurs with its surrounding texts, we further propose novel "Table-to-Text" and "Text-to-Table" operators to handle joint table-text reasoning scenarios. This way, we can adequately exploit the unlabeled table resources to obtain a well-performed reasoning model under an unsupervised setting. Our experiments cover different tasks (question answering and fact verification) and different domains (general and specific), showing that our unsupervised methods can achieve at most 93% performance compared to supervised models. The impressive performance demonstrates that UCTR can significantly reduce the dependence on manual annotation. Moreover, we also find that it can substantially boost the supervised performance in low-resourced domains as a data augmentation technique.

Automated Reasoning Based on Tableau

PoTable: Programming Standardly on Table-based Reasoning Like a Human Analyst

CRT-QA: A Dataset of Complex Reasoning Question Answering over Tabular Data

Toward a Unified Framework for Unsupervised Complex Tabular Reasoning.

Beyond Embeddings: The Promise of Visual Table in Visual Reasoning

On A Semi-Automatic Method for Generating Composition Tables

MATATA: A weakly-supervised MAthematical Tool-Assisted reasoning for Tabular Applications

Learning Guided Automated Reasoning: A Brief Survey

ART: Rule Based Future-Inference Deduction

TART: An Open-Source Tool-Augmented Framework for Explainable Table-based Reasoning

Optimization Techniques for Unsupervised Complex Table Reasoning via Self-Training Framework

TabMoE: A General Framework for Diverse Table-Based Reasoning with Mixture-of-Experts

VisTR: Visualizations as Representations for Time-series Table Reasoning

A Tableau Algorithm for Term Logic of Generic Sentences

Chain-of-Table: Evolving Tables in the Reasoning Chain for Table Understanding

Reasoning in artificial intelligence:advances and challenges

Tablepedia: Automating PDF Table Reading in an Experimental Evidence Exploration and Analytic System

Interactive Visual Reasoning under Uncertainty

Research on Tableau Algorithm for Horn-Extended DL

A Survey of Table Reasoning with Large Language Models

Take A Step Back: Rethinking the Two Stages in Visual Reasoning