mlsanityData to 14 Oct 2026

Synthetic Contrastive Reasoning for Multi-Table Q&A

arXiv 2606.05382 ↗First posted 3 Jun 2026cs.AIcited by 0

On the map: LLM reasoning → Knowledge- and tool-augmented reasoning

Abstract (arXiv)

Multi-table question answering requires models to retrieve relevant evidence, link schemas, and perform compositional reasoning across relational tables. Existing multi-table Q&A resources typically provide questions and final answers but lack reasoning supervision that explains how answers are derived. To address this gap, we construct a synthetic contrastive reasoning-trace dataset for MMQA by generating validated positive traces and plausible negative traces with heterogeneous LLMs. We then use the resulting preference pairs to fine-tune open-weight LLMs with Contrastive Preference Optimization (CPO). Across Qwen3-14B, Mistral-8B, and Llama-3.1-8B, CPO achieves absolute average improvements over Q&A supervised fine-tuning ranging from 9.7%-16.3%, with gains up to 21 percentage points on MMQA. Ablations show that heterogeneous positive and negative trace generators strengthen the contrastive signal, and automated as well as human evaluations indicate that the generated pairs are largely faithful, coherent, and meaningfully contrastive.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Placed in Knowledge- and tool-augmented reasoning, one of 17 directions in LLM reasoning. Builds on 15 of its 33 references that are in the map. First posted on arXiv 3 Jun 2026.

Builds on

15 of 33 references.

Closest ancestors · 3

Same direction, newest first

  1. 2025MMTU: A Massive Multi-Task Table Understanding and Reasoning Benchmark5 Jun
  2. 2025Reasoning-Table: Exploring Reinforcement Learning for Table Reasoning2 Jun
  3. 2025Table-R1: Inference-Time Scaling for Table Reasoning29 May

Other references · 12

  1. 2023LLaMA: Open and Efficient Foundation Language Modelscited by 14,132
  2. 2024GPT-4o System Cardcited by 4,588
  3. 2023Qwen Technical Reportcited by 3,344
  4. 2018Spider: A Large-Scale Human-Labeled Dataset for Complex and Cross-Domain Semantic Parsing and Text-to-SQL Taskcited by 1,139
  5. 2020TaBERT: Pretraining for Joint Understanding of Textual and Tabular Datacited by 398
  6. 2024Contrastive Preference Optimization: Pushing the Boundaries of LLM Performance in Machine Translationcited by 394
  7. 2024Direct Preference Optimization with an Offsetcited by 122
  8. 2023Contrastive Preference Learning: Learning from Human Feedback without RLcited by 79
  9. 2024TableGPT2: A Large Multimodal Model with Tabular Data Integrationcited by 72
  10. 2023Contrastive Chain-of-Thought Promptingcited by 52
  11. 2023MultiTabQA: Generating Tabular Answers for Multi-Table Question Answeringcited by 32
  12. 2024TQA-Bench: Evaluating LLMs for Multi-Table Question Answeringcited by 17

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2026MATA: Multi-Agent Framework for Reliable and Flexible Table Question Answering
  2. 2023Contrastive Decoding Improves Reasoning in Large Language Models
  3. 2026Evolving from Lessons: Skill-Augmented Table Graph Reasoning for Operation-wise Table Question Answering
  4. 2026TraceBack: Multi-Agent Decomposition for Fine-Grained Table Attribution
  5. 2024Seek and Solve Reasoning for Table Question Answering
  6. 2025Plugging Schema Graph into Multi-Table QA: A Human-Guided Framework for Reducing LLM Reliance
  7. 2026QUIETT: Query-Independent Table Transformation for Robust Reasoning
  8. 2026ReasonTabQA: A Comprehensive Benchmark for Table Question Answering from Real World Industrial Scenarios