Search papers, labs, and topics across Lattice.
This paper introduces MultAttnAttrib, a training-free method for generating multimodal attributions in long document question answering, addressing the critical need for accurate evidence attribution in AI assistants. By leveraging a model's prefill pass and selected attention heads, the method identifies source evidence efficiently, outperforming existing attribution methods and matching the performance of advanced models like GPT 5.4. Additionally, the authors present MultAttrEval, a novel benchmark dataset designed for fine-grained evaluation of multimodal attributions, enhancing the landscape of attribution research.
MultAttnAttrib achieves superior attribution accuracy while cutting inference latency to one-seventh of traditional prompting methods, revolutionizing multimodal evidence tracing in AI.
As grounded QA systems are increasingly deployed in AI assistants, accurately attributing generated answers to evidence is critical for user trust and model safety. While unimodal attributions have been explored in depth, the multimodal setting remains relatively under-researched. As a result, we introduce MultAttnAttrib, a training-free attribution-generation method that leverages a model's prefill pass, selected attention heads, and calibrated thresholds to locate source evidence within a document. To establish baseline results for the method, we introduce MultAttrEval, a complementary benchmark dataset annotated with fine-grained, ground-truth attributions for answer components grounded in multimodal source documents. To our knowledge, this is the first evaluation dataset designed specifically for multimodal attribution in long-form documents. Experimental results show that MultAttnAttrib consistently outperforms a variety of attribution-generation methods, including several strong prompting-based approaches and matches the latest frontier models such as GPT 5.4. Our method not only substantially improves attribution accuracy for both unimodal and multimodal attribution types, but also produces attributions at up to one-seventh of the direct inference latency compared to prompting on the same base model.
