English

RSAT: Structured Attribution Makes Small Language Models Faithful Table Reasoners

Computation and Language 2026-05-08 v2 Artificial Intelligence Information Retrieval Machine Learning

Abstract

When a language model answers a table question, users have no way to verify which cells informed which reasoning steps. We introduce RSAT, a method that trains small language models (SLMs, 1-8B) to produce step-by-step reasoning with cell-level citations grounded in table evidence. Phase 1 (SFT) teaches a structured JSON output format from verified reasoning traces. Phase 2 (GRPO) optimizes a composite reward centered on NLI-based faithfulness, alongside citation validity and parsimony. Across six models from two families-Qwen 2.5 (1.5B/3B/7B) and Llama 3 (1B/3B/8B)-RSAT improves faithfulness 3.7×\times over SFT alone (0.224\rightarrow0.826), with near-perfect citation validity (0.992). Post-hoc attribution collapses below 13% format success, confirming that attribution must be integrated into reasoning, not retrofitted. Ablations show the faithfulness reward is essential: removing it drops faithfulness from 0.97 to 0.03.

Keywords

Cite

@article{arxiv.2605.00199,
  title  = {RSAT: Structured Attribution Makes Small Language Models Faithful Table Reasoners},
  author = {Jugal Gajjar and Kamalasankari Subramaniakuppusamy},
  journal= {arXiv preprint arXiv:2605.00199},
  year   = {2026}
}

Comments

8 pages, 8 tables, 9 figures, and a 3-page Appendix. Accepted at the SURGeLLM Workshop at ACL 2026 and will be included in the proceedings

R2 v1 2026-07-01T12:44:29.134Z