NLCTables: A Dataset for Marrying Natural Language Conditions with Table Discovery

TL;DR

The paper introduces NL-conditional table discovery (nlcTD), addressing the gap where users seek tables relevant to both a query table and an accompanying natural-language condition. It formalizes nlcTD, presents a comprehensive taxonomy of condition types, and introduces nlcTables, a large-scale benchmark with 627 queries, 22,080 candidate tables, and 21,200 ground-truth annotations. A configurable, three-stage dataset construction framework—table preprocessing, query construction via splitting, and ground-truth labeling with LLM-semantics augmentation—enables diverse, scalable evaluation. Six representative baselines are evaluated across NL-only, union, and join scenarios, revealing substantial gaps and guiding future method development for robust NL-conditioned table retrieval. The dataset and baselines are publicly available to accelerate research and real-world adoption of NL-powered table search assistants.

Abstract

With the growing abundance of repositories containing tabular data, discovering relevant tables for in-depth analysis remains a challenging task. Existing table discovery methods primarily retrieve desired tables based on a query table or several vague keywords, leaving users to manually filter large result sets. To address this limitation, we propose a new task: NL-conditional table discovery (nlcTD), where users combine a query table with natural language (NL) requirements to refine search results. To advance research in this area, we present nlcTables, a comprehensive benchmark dataset comprising 627 diverse queries spanning NL-only, union, join, and fuzzy conditions, 22,080 candidate tables, and 21,200 relevance annotations. Our evaluation of six state-of-the-art table discovery methods on nlcTables reveals substantial performance gaps, highlighting the need for advanced techniques to tackle this challenging nlcTD scenario. The dataset, construction framework, and baseline implementations are publicly available at https://github.com/SuDIS-ZJU/nlcTables to foster future research.

Figures (7)

• Figure 1: Table Copilot: a typical nlcTD scenario.
• Figure 2: Illustration of NL-conditional table discovery: Combining the query table with NL conditions (e.g., high-Maths-grade students) enables more precise table retrieval.
• Figure 3: The taxonomy of nlcTD, consisting of 16 NL condition subcategories along with their illustrative examples.
• Figure 4: The three stages of constructing nlcTables: (1) Table Preprocessing: collecting, filtering, and labeling tables; (2) Query Construction: splitting tables vertically and horizontally to create joinable and unionable tables; (3) Ground Truth Generation: generating labels via automatic table splitting with semantic augmentation, and manual SQL-based labeling.
• Figure 5: Comparisons of feasible methods on (left) nlcTables-U and (right) nlcTables-J.