New Framework Improves Ambiguity Detection in Open-Domain QA

Accurately determining if a user query is ambiguous is crucial for open-domain Question Answering (QA) systems, as misclassification can lead to incorrect answers or unnecessary clarifications. Existing methods often confuse answer diversity with true ambiguity, resulting in inaccurate predictions and inefficient processing. This research introduces ARCHIVE (Ambiguity Recognition via Cascaded Hypothesis Inspection and Conflict Verification), a novel framework designed to address these limitations.ARCHIVE detects ambiguity by identifying logical conflicts among potential answers, meaning a query is ambiguous if its valid answers cannot all be true under a single interpretation. The framework employs a lightweight early-exit encoder for straightforward cases and a more sophisticated conflict reasoning module for complex scenarios. Evaluated on the new QuireQA benchmark, ARCHIVE significantly outperforms competitors, improving F1 scores for both ambiguous and unambiguous queries while operating up to 16 times faster.