WebQA: Multihop and Multimodal QA

TL;DR

WebQA introduces a new benchmark for multihop, multimodal web-based question answering, highlighting the challenges for current models and emphasizing the need for unified reasoning across visual and textual sources.

Contribution

The paper presents WebQA, a challenging benchmark that combines visual and textual reasoning tasks to advance multimodal question answering models.

Findings

Large models struggle with WebQA's multimodal reasoning.

WebQA's secondary text-only task ensures language understanding is maintained.

Current models find WebQA difficult, indicating room for improvement.

Abstract

Scaling Visual Question Answering (VQA) to the open-domain and multi-hop nature of web searches, requires fundamental advances in visual representation learning, knowledge aggregation, and language generation. In this work, we introduce WebQA, a challenging new benchmark that proves difficult for large-scale state-of-the-art models which lack language groundable visual representations for novel objects and the ability to reason, yet trivial for humans. WebQA mirrors the way humans use the web: 1) Ask a question, 2) Choose sources to aggregate, and 3) Produce a fluent language response. This is the behavior we should be expecting from IoT devices and digital assistants. Existing work prefers to assume that a model can either reason about knowledge in images or in text. WebQA includes a secondary text-only QA task to ensure improved visual performance does not come at the cost of language understanding. Our challenge for the community is to create unified multimodal reasoning models that answer questions regardless of the source modality, moving us closer to digital assistants that not only query language knowledge, but also the richer visual online world.