arXiv:2501.06286v3 Announce Type: replace-cross
Abstract: Multi-hop question answering requires a system to identify and integrate evidence distributed across documents, yet large language models remain vulnerable to irrelevant context. We investigate this evidence bottleneck in the English HotpotQA distractor setting and introduce Bactrainus, a modular selector-reader framework that separates paragraph selection, supporting-sentence identification, and answer generation. Optional question decomposition and teacher-generated rationale supervision make it possible to test where additional reasoning structure is useful. The evaluation combines foundation-model screening, controlled context and prompting ablations, parameter-efficient adaptation of Llama 3.1 8B Instruct and Llama 3.1 70B Instruct readers, and integrated selector-reader experiments. Supplying the full candidate context instead of gold supporting facts reduces answer token-overlap F1 by 17-21 points, showing that scale alone does not remove context sensitivity. The largest observed differences are associated with reader adaptation and sentence-level evidence control. The strongest reported configuration obtains 89.01 answer F1 and 79.70 joint F1, whereas decomposition and rationale-supervision variants yield smaller, recipe-dependent changes. These findings support auditable, explicitly supervised evidence interfaces for fixed-candidate multi-hop QA and motivate blind, matched, multi-seed evaluation of the remaining small differences.
Abstract: Multi-hop question answering requires a system to identify and integrate evidence distributed across documents, yet large language models remain vulnerable to irrelevant context. We investigate this evidence bottleneck in the English HotpotQA distractor setting and introduce Bactrainus, a modular selector-reader framework that separates paragraph selection, supporting-sentence identification, and answer generation. Optional question decomposition and teacher-generated rationale supervision make it possible to test where additional reasoning structure is useful. The evaluation combines foundation-model screening, controlled context and prompting ablations, parameter-efficient adaptation of Llama 3.1 8B Instruct and Llama 3.1 70B Instruct readers, and integrated selector-reader experiments. Supplying the full candidate context instead of gold supporting facts reduces answer token-overlap F1 by 17-21 points, showing that scale alone does not remove context sensitivity. The largest observed differences are associated with reader adaptation and sentence-level evidence control. The strongest reported configuration obtains 89.01 answer F1 and 79.70 joint F1, whereas decomposition and rationale-supervision variants yield smaller, recipe-dependent changes. These findings support auditable, explicitly supervised evidence interfaces for fixed-candidate multi-hop QA and motivate blind, matched, multi-seed evaluation of the remaining small differences.
This article has been indexed from cs.AI updates on arXiv.org
Read the original article: