# Context Generation Improves Open Domain Question Answering

**Authors:** Dan Su, Mostofa Patwary, Shrimai Prabhumoye, Peng Xu, Ryan Prenger, Mohammad Shoeybi, Pascale Fung, Anima Anandkumar, Bryan Catanzaro  
**arXiv:** [2210.06349](https://arxiv.org/abs/2210.06349)  
**Date:** 2022-10-12  
**Summary page:** https://www.summarizepaper.com/en/paper/2210.06349v2/

## Key points

- The paper proposes a two-stage framework for closed-book question answering (QA)
- The approach utilizes pretrained language models (LMs) to exploit stored knowledge
- A coarse-to-fine approach is used to extract relevant knowledge and provide answers
- Experimental results show significant improvement over previous closed-book QA methods
- Achieves an exact matching accuracy of 68.6% on three QA benchmarks
- Performs on par with open-book methods that use external knowledge sources
- Authored by Dan Su, Mostofa Patwary, Shrimai Prabhumoye, Peng Xu, Ryan Prenger, Mohammad Shoeybi, Pascale Fung, Anima Anandkumar, and Bryan Catanzaro
- Accepted at EACL2023 conference

## Summary

The paper titled "Context Generation Improves Open Domain Question Answering" addresses the challenge of closed-book question answering (QA) by proposing a two-stage framework that effectively exploits stored knowledge in pretrained language models (LMs). The proposed approach utilizes a coarse-to-fine approach to extract relevant knowledge and provide answers, overcoming the limitations of previous methods. Experimental results on three QA benchmarks demonstrate significant improvement over previous closed-book QA methods, achieving an exact matching accuracy of 68.6%. This method also performs on par with open-book methods that utilize external knowledge sources, showcasing its potential for hybrid models. The paper was authored by Dan Su, Mostofa Patwary, Shrimai Prabhumoye, Peng Xu, Ryan Prenger, Mohammad Shoeybi, Pascale Fung, Anima Anandkumar, and Bryan Catanzaro and has been accepted at EACL2023 conference.