Weaver : Interweaving SQL and LLM for Table Reasoning

ACL ARR 2025 May Submission7463 Authors

Abstract:

Querying tables with unstructured data is chal-
lenging due to the presence of text (or im-
age), either embedded in the table or in ex-
ternal paragraphs, which traditional SQL strug-
gles to process, especially for tasks requiring
semantic reasoning. While Large Language
Models (LLMs) excel at understanding context,
they face limitations with long input sequences.
Existing approaches that combine SQL and
LLMs typically rely on rigid, predefined work-
flows, limiting their adaptability to complex
queries. To address these issues, we introduce
Weaver , a modular pipeline that dynamically
integrates SQL and LLMs for table-based ques-
tion answering (TableQA). Weaver generates a
flexible, step-by-step plan that combines SQL
for structured data retrieval with LLMs for se-
mantic processing. By decomposing complex
queries into manageable subtasks, Weaver im-
proves accuracy and generalization. Our ex-
periments show that Weaver consistently out-
performs state-of-the-art methods across four
TableQA datasets, reducing both API calls and
error rates.

Paper Type: Long

Research Area: Question Answering

Research Area Keywords: Question Answering, NLP Applications

Contribution Types: NLP engineering experiment

Languages Studied: English

Submission Number: 7463