A new arXiv preprint, Would you still call this Dax?, takes up a question at the intersection of vision and language: when a model or a person encounters a new visual concept, how do they connect it to language? The authors note that while vision-language models (VLMs) are frequently exposed to novel visual concepts, the mapping of those references to words after exposure is still poorly understood.

The paper appears to investigate this by drawing a parallel between human learners and VLMs, asking whether both groups handle novel visual references in similar ways. The title hints at a core concern: after learning a new name for an object, do VLMs and humans apply it consistently to new instances?

The abstract is brief and does not include specific results, so the full findings are not yet available from this summary alone. Still, the research points to an important gap in how we evaluate VLMs—not just on recognition accuracy, but on the dynamic, open-ended process of learning and using new visual language in context.