Seeing Is Not Sharing: Some Vision-Language Models Overestimate Common Ground in Asymmetric Dialogue

2026-06-30

AI Summary

This study addresses a critical limitation in vision-language models: their tendency to conflate shared perceptual access with genuine mutual understanding in asymmetric dialogue settings. Leveraging the HCRC MapTask dataset, the authors formulate a referential expression matching task that systematically manipulates map visibility and conversational context to evaluate models’ ability to infer true common ground. The work reveals, for the first time, a systematic bias in prominent models (e.g., Qwen3-VL-8B-Instruct) to treat task-relevant map content—regardless of modality—as evidence of shared knowledge, stemming from their neglect of grounding processes in dialogue history. Through multimodal inputs, calibration analyses, and referential chain tracing across five models, experiments demonstrate that while providing real maps improves overall performance, it leads to significant overestimation of alignment, thereby reducing accuracy in misaligned scenarios—a bias attributable to content relevance rather than the visual modality per se.

Abstract

In collaborative dialogue, shared perception does not guarantee shared interpretation. Mutual understanding must be established through interaction. We investigate whether vision-language models (VLMs) can distinguish what could be shared from what has been shared between dialogue participants through grounding. We formulate this as an interpretation-matching task on 13,077 annotated reference expressions from HCRC MapTask dialogues, and evaluate VLMs under systematically controlled manipulations of dialogue context and map-information access. Our results show that providing authentic map images improves overall performance but shifts models toward over-predicting alignment. Textual descriptions of the same map content reproduce this bias, while non-informative images suppress alignment predictions entirely, indicating that the bias is driven by task-relevant map content, not the visual channel. This improvement comes at the cost of degraded accuracy on non-aligned cases. Calibration analysis and reference-chain tracking further suggest that models rely on static referential cues on the maps rather than tracking how grounding unfolds through dialogue history. We observe these patterns most clearly in Qwen3-VL-8B-Instruct and, to varying degrees, in four additional models from two architecture families. In models that exhibit the bias, map content, whether presented visually or textually, is treated as evidence of mutual understanding, conflating potential with established common ground.

Problem

• vision-language models
• common ground
• asymmetric dialogue
• reference interpretation
• grounding

Innovation

• vision-language models
• common ground
• dialogue grounding
• reference interpretation
• asymmetric dialogue