Synthetic Contrastive Reasoning for Multi-Table Q&A
This work addresses the lack of reasoning supervision in multi-table Q&A, providing a method to improve compositional reasoning for LLMs.
The authors constructed a synthetic contrastive reasoning-trace dataset for multi-table Q&A and fine-tuned open-weight LLMs with Contrastive Preference Optimization (CPO), achieving absolute average improvements of 9.7%-16.3% over supervised fine-tuning, with gains up to 21 percentage points on MMQA.
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.