ARCHIVE: A New Framework for Accurate and Efficient Ambiguity Detection in Open-Domain QA
So, there's this new system called ARCHIVE, which stands for Ambiguity Recognition via Cascaded Hypothesis Inspection and Conflict Verification. It’s aimed at improving how open-domain question answering systems spot ambiguity. This research was shared in a paper on arXiv (ID: 2608.03177) on August 26, 2024. One of the main issues they address is how existing methods often mix up answer variety with ambiguity, leading to incorrect predictions. ARCHIVE figures out ambiguity by finding logical conflicts, deciding that a query is ambiguous if its valid answers can't fit together under one understanding. They also created QuireQA, a dataset with 4,703 queries to help evaluate ambiguity detection while focusing on efficient processing to cut down on unnecessary computational costs.
Key facts
- ARCHIVE is a framework for ambiguity detection in open-domain QA.
- It detects ambiguity via logical conflict among valid answers.
- It uses a lightweight early-exit encoder and a conflict reasoning module.
- An invariance objective enhances robustness to noisy answer sets.
- QuireQA is a new benchmark with 4,703 queries.
- The paper is available on arXiv with ID 2608.03177.
- The announcement type is 'new' and the paper was posted on August 26, 2024.
- Existing methods conflate answer diversity with ambiguity.
Entities
Institutions
- arXiv