TRACE
Paper at the GroundLM 2026 workshop, co-located with EMNLP 2026, on the gap between finding the right paper and producing an answer a scorer can verify.
Accepted at the first workshop on Grounding Language Models: Learning Faithfully and Efficiently (GroundLM 2026), co-located with EMNLP 2026. Written with Divya Godara.
Finding the relevant paper is not the same as producing a verifiable answer from it. The LitTraceQA task wants a canonical paper identifier, evidence pinned to a page or an object, and a typed answer the evaluator can match. The paper calls the distance between those two things the grounding contract gap, and TRACE is the system built to close it: retrieval grouped by the question's target, evidence located independently of the answer, table values read as a structure rather than as text, and validation that fails closed instead of guessing.
The preprint is on arXiv.
