Language-Conditioned Visual Grounding with CLIP Multilingual

TL;DR

This study investigates the performance gaps in multilingual vision-language models, revealing that the text branch causes most issues, especially in low-resource languages, and highlights the importance of spatial alignment for effective grounding.

Contribution

The paper introduces a dense multilingual CLIP probe to isolate the source of performance gaps, demonstrating the impact of language resource levels and model scaling on visual grounding accuracy.

Findings

Low-resource languages show significant deficits in the text branch.

Scaling the visual encoder widens the gap for some languages but improves others.

Spatial misalignment, not signal collapse, is the main failure mode.

Abstract

Multilingual vision-language models exhibit systematic performance gaps across languages, but the mechanism remains ambiguous: cross-language divergence could arise from the visual encoder, the text branch, or their interaction. We resolve this ambiguity through a dense multilingual CLIP probe in which the visual encoder is held identical across thirteen typologically diverse languages and only the XLM-RoBERTa text branch varies. We evaluate two CLIP architectures spanning a 7x visual-encoder scale gap (XLM-R base + ViT-B/32, ~87M visual parameters; XLM-R large + ViT-H/14, ~632M) on 11 concepts and 210 images, and quantify cross-language agreement via cluster-mask IoU, top-percentile IoU, and Spearman rank correlation against an English reference (n=2,310 paired observations per language). Three findings emerge. First, low-resource languages (Arabic, Basque, Luxembourgish) incur a structural penalty at both backbone scales (Wilcoxon HR>LR p<10^-300; cluster-mask IoU gap +0.114 at base, +0.143 at large), isolating the deficit to the text branch. Second, scaling the encoder 7x widens the gap for structural failure cases (Basque {\Delta}=-0.056, Luxembourgish {\Delta}=-0.076) while improving Arabic ({\Delta}=+0.033), separating corpus-coverage from tokeniser-fertility failures. Third, peak similarity is preserved across languages (mean ratio 0.94 at large scale) while cluster-mask IoU drops sharply, identifying spatial misalignment, not signal collapse, as the dominant failure mode. At 3.4-3.9 Wh per 1,000 queries, dense-CLIP grounding is competitive with high-throughput inference budgets, positioning it as a practical substrate for energy-aware multilingual deployment.