# Schr"odinger's Bat: Diffusion Models Sometimes Generate Polysemous Words in Superposition

arXiv: https://arxiv.org/abs/2211.13095 · HTML: https://mlsanity.com/paper/2211.13095/

## Key facts

- Cited by 8 papers in the map (data through 14 Oct 2026).
- Placed in Image generation, text-to-image and editing, one of 14 directions in Diffusion models.
- Builds on 9 of its 14 references that are in the map.
- First posted on arXiv 23 Nov 2022.

## Placed in

- Diffusion models / [Image generation, text-to-image and editing](https://mlsanity.com/diffusion/image-generation-text-to-image-and-editing/)
- Mechanistic interpretability / [Interpreting and steering diffusion models](https://mlsanity.com/mechinterp/interpreting-and-steering-diffusion-models/)

## Builds on (most cited first, 9 of 9)

- [Learning Transferable Visual Models From Natural Language Supervision](https://mlsanity.com/paper/2103.00020/) (2021)
- [U-Net: Convolutional Networks for Biomedical Image Segmentation](https://mlsanity.com/paper/1505.04597/) (2015)
- [High-Resolution Image Synthesis with Latent Diffusion Models](https://mlsanity.com/paper/2112.10752/) (2021)
- [Hierarchical Text-Conditional Image Generation with CLIP Latents](https://mlsanity.com/paper/2204.06125/) (2022)
- [Man is to Computer Programmer as Woman is to Homemaker? Debiasing Word Embeddings](https://mlsanity.com/paper/1607.06520/) (2016)
- [Toy Models of Superposition](https://mlsanity.com/paper/2209.10652/) (2022)
- [Linear Algebraic Structure of Word Senses, with Applications to Polysemy](https://mlsanity.com/paper/1601.03764/) (2016)
- [DALLE-2 is Seeing Double: Flaws in Word-to-Concept Mapping in Text2Image Models](https://mlsanity.com/paper/2210.10606/) (2022)
- [How to marry a star: probabilistic constraints for meaning in context](https://mlsanity.com/paper/2009.07936/) (2020)

## Built on by (most cited first, 8 of 8)

- [Be Yourself: Bounded Attention for Multi-Subject Text-to-Image Generation](https://mlsanity.com/paper/2403.16990/) (2024)
- [Diffusion Policies for Out-of-Distribution Generalization in Offline Reinforcement Learning](https://mlsanity.com/paper/2307.04726/) (2023)
- [DiffusionPID: Interpreting Diffusion via Partial Information Decomposition](https://mlsanity.com/paper/2406.05191/) (2024)
- [Transferring Visual Attributes from Natural Language to Verified Image Generation](https://mlsanity.com/paper/2305.15026/) (2023)
- [An Exploration of Default Images in Text-to-Image Generation](https://mlsanity.com/paper/2505.09166/) (2025)
- [Enhancing Generalization in Data-free Quantization via Mixup-class Prompting](https://mlsanity.com/paper/2507.21947/) (2025)
- [Un-Doubling Diffusion: LLM-guided Disambiguation of Homonym Duplication](https://mlsanity.com/paper/2509.21262/) (2025)
- [Where did the ambiguity go? Examining how multimodal models interpret polysemous words](https://mlsanity.com/paper/2608.00410/) (2026)

## Abstract (arXiv)

Recent work has shown that despite their impressive capabilities, text-to-image diffusion models such as DALL-E 2 (Ramesh et al., 2022) can display strange behaviours when a prompt contains a word with multiple possible meanings, often generating images containing both senses of the word (Rassin et al., 2022). In this work we seek to put forward a possible explanation of this phenomenon. Using the similar Stable Diffusion model (Rombach et al., 2022), we first show that when given an input that is the sum of encodings of two distinct words, the model can produce an image containing both concepts represented in the sum. We then demonstrate that the CLIP encoder used to encode prompts (Radford et al., 2021) encodes polysemous words as a superposition of meanings, and that using linear algebraic techniques we can edit these representations to influence the senses represented in the generated images. Combining these two findings, we suggest that the homonym duplication phenomenon described by Rassin et al. (2022) is caused by diffusion models producing images representing both of the meanings that are present in superposition in the encoding of a polysemous word.