Un-Doubling Diffusion: LLM-guided Disambiguation of Homonym Duplication

2025-09-25

AI Summary

Homographs—words with identical spelling but distinct semantics—induce “homograph duplication” in text-to-image diffusion models, causing semantic confusion and generating semantically incoherent images. More critically, English-centric prompting pipelines require non-English prompts to undergo translation preprocessing, often erroneously mapping unambiguous source-language terms to English homographs, thereby introducing translation-induced semantic distortion. To address this, we propose an LLM-guided, context-aware prompt expansion method that enables fine-grained semantic disambiguation. We introduce the Homograph Repetition Rate (HRR), a novel quantitative metric, to systematically characterize and mitigate translation-triggered semantic duplication for the first time. Our evaluation integrates vision-language model (VLM)-based automated assessment with human validation, achieving significant HRR reduction and markedly improving both semantic fidelity and cross-lingual consistency of generated images.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: Prompt Engineering / Prompting

Application Category

• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Homonyms are words with identical spelling but distinct meanings, which pose challenges for many generative models. When a homonym appears in a prompt, diffusion models may generate multiple senses of the word simultaneously, which is known as homonym duplication. This issue is further complicated by an Anglocentric bias, which includes an additional translation step before the text-to-image model pipeline. As a result, even words that are not homonymous in the original language may become homonyms and lose their meaning after translation into English. In this paper, we introduce a method for measuring duplication rates and conduct evaluations of different diffusion models using both automatic evaluation utilizing Vision-Language Models (VLM) and human evaluation. Additionally, we investigate methods to mitigate the homonym duplication problem through prompt expansion, demonstrating that this approach also effectively reduces duplication related to Anglocentric bias. The code for the automatic evaluation pipeline is publicly available.

Problem

• Homonyms cause diffusion models to generate multiple meanings simultaneously
• Anglocentric bias creates artificial homonyms through translation issues
• Measuring and reducing homonym duplication in text-to-image generation

Innovation

• LLM-guided disambiguation of homonym duplication
• Measuring duplication rates with automatic VLM evaluation
• Mitigating duplication through prompt expansion techniques