See or Recall: A Sanity Check for the Role of Vision in Solving Visualization Question Answer Tasks with Multimodal LLMs (2504.09809v2)

Abstract: Recent developments in multimodal LLMs (MLLM) have equipped LLMs to reason about vision and language jointly. This permits MLLMs to both perceive and answer questions about data visualization across a variety of designs and tasks. Applying MLLMs to a broad range of visualization tasks requires us to properly evaluate their capabilities, and the most common way to conduct evaluation is through measuring a model's visualization reasoning capability, analogous to how we would evaluate human understanding of visualizations (e.g., visualization literacy). However, we found that in the context of visualization question answering (VisQA), how an MLLM perceives and reasons about visualizations can be fundamentally different from how humans approach the same problem. During the evaluation, even without visualization, the model could correctly answer a substantial portion of the visualization test questions, regardless of whether any selection options were provided. We hypothesize that the vast amount of knowledge encoded in the LLM permits factual recall that supersedes the need to seek information from the visual signal. It raises concerns that the current VisQA evaluation may not fully capture the models' visualization reasoning capabilities. To address this, we propose a comprehensive sanity check framework that integrates a rule-based decision tree and a sanity check table to disentangle the effects of "seeing" (visual processing) and "recall" (reliance on prior knowledge). This validates VisQA datasets for evaluation, highlighting where models are truly "seeing", positively or negatively affected by the factual recall, or relying on inductive biases for question answering. Our study underscores the need for careful consideration in designing future visualization understanding studies when utilizing MLLMs.