UpstreamQA: A Modular Framework for Explicit Reasoning on Video Question Answering Tasks
Jason Nguyen, Ameet Rao, Alexander Chang, Ishaan Kumar, Erin Tan · 2026
Read UpstreamQA: A Modular Framework for Explicit Reasoning on Video Question Answering Tasks by Jason Nguyen, Ameet Rao, Alexander Chang, Ishaan Kumar, Erin Tan (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2604-23145.txt.
Description
Nguyen and four colleagues build UpstreamQA, a modular setup that makes video question answering reason out loud instead of hiding steps inside large multimodal models. Multimodal reasoning models first identify objects and sketch scene context, then pass those traces to downstream models that answer the questions. Tests on OpenEQA and NExTQA with two reasoning models and two answering models show clear gains in accuracy and clarity. Gains fade when the baseline already scores high, so explicit reasoning helps most where the base model still struggles.
Catalog record
| CREATOR | Jason Nguyen, Ameet Rao, Alexander Chang, Ishaan Kumar, Erin Tan |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 32191 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2604.23145