Making Memes Accessible
Cole Gleason, Amy Pavel, Xingyu Liu, Patrick Carrington, Lydia B. Chilton, Jeffrey P. Bigham · 2019 · Proceedings of the 21st International ACM SIGACCESS Conference on Computers and Accessibility (ASSETS 2019) · doi:10.1145/3308561.3353792
Summary
This paper presents two semi-automatic methods for making internet image macro memes accessible to people with vision impairments: rich alternative text descriptions and audio macro memes. Memes are a pervasive form of online communication, yet they are almost entirely inaccessible to screen reader users because they consist of images with overlaid text and no alternative text. Only 0.1% of images on social media have alt text, and even when meme text is provided as alt text, the critical visual context — the character’s expression, posture, and cultural meaning — is lost. The key insight enabling the approach is that memes are highly repetitive: the same base image template is reused thousands of times with different text overlays. The system works by: (1) recognising whether an image is a meme using Google Cloud Vision API (94.4% accuracy), (2) classifying which meme template it matches using a combined structural similarity and colour histogram score (92.25% accuracy), (3) extracting the overlaid text via OCR (22.1% word error rate, 9.2% character error rate), and (4) inserting the extracted text into a pre-authored alternative text template or audio template for that meme. An authoring interface allows users to create templates with placeholders for top and bottom text. For alt text templates, authors write descriptions that contextualise the visual content around the text placeholders. For audio templates, authors select background sounds from a library and place text-to-speech elements with optional comedic pauses.
Key findings
A user study with 10 blind or low-vision Twitter users (ages 19-53, average 31.8) compared three meme formats: text-only (OCR output), alt text with image description, and audio macro memes. Eight of ten participants preferred the alt text format because the visual description conveyed the emotional tone and character context that made the meme meaningful. As P3 said about the First World Problems meme: "It gives you ‘head in hands, crying’. I could get the emotion, but the reason for the emotion appears in the text." Participants reported higher confidence in understanding memes with alt text (mean 3.95/5) compared to text-only (3.55) and audio (3.52). Accuracy of understanding the joke format was similar across conditions (alt text 63%, audio 70%, text-only 57%), though sample size prevented statistical significance testing. Audio memes raised concerns about universal accessibility — they don’t work with screen readers’ preferred voices and speeds, and are inaccessible to deaf-blind users. Critically, participants expressed strong desire to participate in meme culture: nine said they would share accessible memes, three said they would create memes if given tools, and P1 stated: "I certainly want to be part of the culture." The paper proposes five structured questions for writing meme alt text: Who are the characters? What are they doing? What emotions do they show? Is the image from a recognisable source? Is there anything notable about the background?
Relevance
This paper addresses the social exclusion that arises when a dominant form of online communication is completely inaccessible. Memes are not trivial — they are a primary mode of humour, cultural commentary, and social bonding on platforms where blind users are active participants. The template-based approach is elegant because it leverages memes’ inherent repetitiveness: authoring one good description or audio template for a base meme image can make thousands of instances accessible automatically. For accessibility practitioners, the five structured questions for meme alt text provide a practical framework that goes beyond generic image description guidelines to address the specific communicative function of memes — conveying emotional tone, character context, and the setup-punchline structure that makes them humorous. The finding that participants imagined memes as "low-effort drawings or stick figures" rather than photographs highlights how persistent inaccessibility shapes blind users’ mental models of visual culture. The participant quote — "If there has to be a lot of useless content out there, it ought to be accessible" — captures a powerful principle: accessibility is not just for important or productive content, but for the full spectrum of human communication and culture.
Tags: alternative text · blind · low vision · social media · image accessibility · screen readers · OCR · computer vision · memes · web accessibility · audio description
Standards referenced: WCAG 2.1