Key points are not available for this paper at this time.
Multi-hop question answering (QA) is a challenging task that requires complex reasoning over multiple documents. Existing works commonly introduce techniques such as graph modeling and question decomposition to explore precise intermediate results of multi-hop reasoning, leading to complexity growth and error accumulation. In this paper, we propose FE2H, a simple yet effective framework without extra tasks to address these problems. FE2H is based on our key observation that a standard fine-tuned pre-trained language model (PLM) for QA could achieve strong performance once the input context could be encoded by PLM without truncation. Specifically, a novel two-stage document selector is pro-posed to generate sufficient context while avoiding input truncation. Additionally, an enhanced reader trained with a two-stage strategy is devised to further boost the performance. Extensive experiments on the popular multi-hop QA benchmark HotpotQA show that despite its simplicity, FE2H achieves competitive results compared to state-of-the-art methods.
Li et al. (Fri,) studied this question.