mlsanityData to 14 Oct 2026

SoS: Analysis of Surface over Semantics in Multilingual Text-To-Image Generation

arXiv 2601.16803 ↗First posted 23 Jan 2026cs.CL cs.AIcited by 0

Not on the map. Why?

Abstract (arXiv)

Text-to-image (T2I) models are increasingly employed by users worldwide. However, prior research has pointed to the high sensitivity of T2I towards particular input languages - when faced with languages other than English (i.e., different surface forms of the same prompt), T2I models often produce culturally stereotypical depictions, prioritizing the surface over the prompt's semantics. Yet a comprehensive analysis of this behavior, which we dub Surface-over-Semantics (SoS), is missing. We present the first analysis of T2I models' SoS tendencies. To this end, we create a set of prompts covering 171 cultural identities, translated into 14 languages, and use it to prompt seven T2I models. To quantify SoS tendencies across models, languages, and cultures, we introduce a novel measure and analyze how the tendencies we identify manifest visually. We show that all but one model exhibit strong surface-level tendency in at least two languages, with this effect intensifying across the layers of T2I text encoders. Moreover, these surface tendencies frequently correlate with stereotypical visual depictions.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Builds on 22 of its 39 references that are in the map. First posted on arXiv 23 Jan 2026.

Builds on

22 of 39 references.

References · 22

  1. 2021Learning Transferable Visual Models From Natural Language Supervisioncited by 33,744
  2. 2023SDXL: Improving Latent Diffusion Models for High-Resolution Image Synthesiscited by 4,356
  3. 2021Zero-Shot Text-to-Image Generationcited by 4,288
  4. 2024Scaling Rectified Flow Transformers for High-Resolution Image Synthesiscited by 4,181
  5. 2021CLIPScore: A Reference-free Evaluation Metric for Image Captioningcited by 2,424
  6. 2020A Primer in BERTology: What we know about how BERT workscited by 815
  7. 2022Easily Accessible Text-to-Image Generation Amplifies Demographic Stereotypes at Large Scalecited by 262
  8. 2023Having Beer after Prayer? Measuring Cultural Bias in Large Language Modelscited by 153
  9. 2023Social Biases through the Text-to-Image Generation Lenscited by 127
  10. 2023Kandinsky: an Improved Text-to-Image Synthesis with Image Prior and Latent Diffusioncited by 114
  11. 2024INCLUDE: Evaluating Multilingual Language Understanding with Regional Knowledgecited by 86
  12. 2023SeeGULL: A Stereotype Benchmark with Broad Geo-Cultural Coverage Leveraging Generative Modelscited by 66
10 more
  1. 2024Survey of Bias In Text-to-Image Generation: Definition, Evaluation, and Mitigationcited by 66
  2. 2024MBBQ: A Dataset for Cross-Lingual Comparison of Stereotypes in Generative LLMscited by 35
  3. 2022Speaking Multiple Languages Affects the Moral Bias of Language Modelscited by 34
  4. 2023AltDiffusion: A Multilingual Text-to-Image Diffusion Modelcited by 30
  5. 2023Stereotypes and Smut: The (Mis)representation of Non-cisgender Identities by Text-to-Image Modelscited by 27
  6. 2024ViSAGe: A Global-Scale Analysis of Visual Stereotypes in Text-to-Image Generationcited by 27
  7. 2024Kandinsky 3: Text-to-Image Synthesis for Multifunctional Generative Frameworkcited by 25
  8. 2023Mitigating stereotypical biases in text to image generative systemscited by 19
  9. 2024Multilingual Text-to-Image Generation Magnifies Gender Stereotypes and Prompt Engineering May Not Help Youcited by 15
  10. 2023A Multidimensional Analysis of Social Biases in Vision Transformerscited by 8

Built on by

Not cited yet.