Latent Bridges for Multi-Table Question Answering
This work addresses the challenge of integrating relational databases with LLMs for question answering, offering an efficient and principled approach that preserves LLM reasoning capabilities.
GRAB introduces a constructor-encoder-bridge pipeline that uses a graph encoder and latent tokens to provide LLMs with compact structural representations of relational data, achieving significant improvements in multi-table question answering while keeping the LLM frozen.
We introduce GRAB, a constructor-encoder-bridge pipeline for table question answering. Our method lifts relational data into an heterogeneous graph, encodes it via message passing, and transfers the signals to an LLM through a small set of query-conditioned latent tokens. This provides the LLM with a compact, task-relevant structural representation together with the flattened text. Crucially, the LLM remains strictly frozen to preserve its general reasoning capabilities; we train only the lightweight graph encoder and latent bridge (91M parameters), allowing the entire pipeline to be trained efficiently. Our pipeline significantly improves performance on relational Question Answering, with the largest gains in demanding multi-table settings, offering an efficient, principled way to connect relational deep learning with LLMs.