1 of 32

Alt-Text Revolution: Bridging the Image Accessibility Gap on Wikipedia with apps and AI

Angie Cervellera (Wikimedia Argentina)

Fiona Romeo (Culture & Heritage, WMF)

Jazmin Tanner (Apps PM, WMF)

2 of 32

Speakers

Friday, November 17 (16:00-17:00)

Angie

Culture and Open Knowledge Program

Wikimedia Argentina

Fiona

Senior Manager,

Culture & Heritage

Wikimedia Foundation

Jazmin Tanner

Lead Product Manager, Apps

Wikimedia Foundation

3 of 32

Priorities

  • Movement Strategy
    • Recommendation #2 to improve user experience by supporting compliance with the most advanced accessibility guidelines using free and open-source software (WCAG for web, W3C mobile web best practices, etc.).
    • Accessibility Guidelines
  • 23-24 Annual Plan
    • Support the growth of high quality and relevant content within the world’s most linguistically diverse, trusted and comprehensive free knowledge ecosystem by enabling and supporting high quality and accessible experiences.

​

4 of 32

AGENDA

  • Importance of Alt-Text
  • Current state of Alt-text on Wikipedia
  • Wikimedia Argentina’s efforts
  • Suggested Edits Context & Vision
  • Demo & Feedback

5 of 32

Engagement with images

A large scale study of reader interactions with images on Wikipedia

Daniele Rama, Tiziano Piccardi, Miriam Redi & Rossano Schifanella (2022)

6 of 32

Image suggestion, screenshot by RHo (WMF), CC-BY-SA 4.0

Campaign image by Blessing Ojo, Creative Director, Notch Visuals, CC BY-SA 4.0

7 of 32

Why alt text matters

A large scale study of reader interactions with images on Wikipedia

Daniele Rama, Tiziano Piccardi, Miriam Redi & Rossano Schifanella (2022)

8 of 32

Add media (photo/video)

9 of 32

Alt text improves access to information for:

​

  • people with no or low vision

​

  • people with reduced connectivity

10 of 32

All of us.

​

As another form of image description, alt text makes images more findable through search.

11 of 32

How are we doing?

  • Manual of style
  • Alt text activities in Argentina, Ireland, and Poland
  • Audio description projects in Brazil and Estonia

50% of images have captions

10% of the images have alt text

3% have effective alt text

​

12 of 32

"Male and Female Sugar Gliders eating mealworms from a bowl"

"A prism refracting white light into a rainbow on a black background"

"Coloured drawing of a huge octopus rising from the sea and attacking a sailing ship's three masts with its spiraling arms"

"The Boer War Memorial in Dunedin. It is stark white and demonstrates New Zealand's patriotism by showing a man defending his fellow soldier."

"Brando with no shirt"

"Four fairies dance in a circle beside another fairy who faces a human king and queen"

"Photo of a tea bowl, dark- coloured, humble, and asymmetric"

OberonNightSeer at English Wikipedia,

CC BY-SA 3.0

Carl Van Vechten, Public domain

Pink Floyd / Reproduction: Kilyann

Le Hen, Public domain

Pierre Denys deok4 Montfort / Etienne Claude Voysard, Public domain

Carl Van Vechten, Public domain

William Blake, Public domain

Chris 73, CC BY-SA 3.0

13 of 32

"Tools to surface articles and images without accessibility coverage...

Mixed human–AI systems… since it is often easier to edit an existing text than it is to write one from scratch."

Characterizing Image Accessibility on Wikipedia across Languages

E Kreiss, 2023)

14 of 32

Good alt text for images on Wikipedia

​

  • Keep it short and clear
  • Only describe what can be seen in the image
  • Focus on what is relevant to the article
  • Include words and other information from the image
  • Take care when identifying or describing people

15 of 32

Wikimedia Argentina’s efforts

  • Events during the Image Description week (2022) and month (2023)
  • More than 60 people were involved
  • DALAT and TNC Accessible program

16 of 32

Main highlights:

  • Collaboration: Wikipedia’s possibility to generate high quality Alt text together.
  • Human Rights: Alt text is strongly related to human rights by making information more accessible

17 of 32

Main highlights:

  • Research: Wikimedia Commons can help us understand what information to consider in alt text.
  • Be contextual: Same image can have different alt text according to the Wikipedia article where is placed.

​

18 of 32

How we edit alt text now

​

​

​

​

​

19 of 32

Jazmin Tanner, Lead PM of Mobile Apps

Suggested Edits on iOS

20 of 32

What are Suggested Edits?

  • Suggested edits presents opportunities for small but vital contributions to Wikipedia.
  • Presently on Android
    • Article Descriptions & Translate
    • Image Captions & Translate
    • Edit Patrol
    • Image Tags
    • Image Recommendations
  • Presently on Web
    • Add a link & Reference
    • Copy Edit
    • Image Recommendations

​

21 of 32

Android’s Impact

Edit Type

% of Edits

Revert Rate

Suggested Edits

73%

4.37%

Non Suggested Edits

29.63%

18.9%

22 of 32

"Tools to surface articles and images without accessibility coverage...

Mixed human–AI systems… since it is often easier to edit an existing text than it is to write one from scratch."

Characterizing Image Accessibility on Wikipedia across Languages

E Kreiss, 2023)

23 of 32

Goals of Suggested Edits on iOS

By July 2024

  • Increase unreverted mobile contributions from iOS by 10%
  • 20,000 articles enhanced using Suggested Edits

By Jan 2024

  • 71% of edits have a score of 3 or higher out of 5
  • Average edit per unique user 7 or higher
  • 60% of users report satisfaction with tool

24 of 32

Using VoiceOver, users can navigate Wikipedia by voice and gesture

​

Users who are colorblind, have contrast sensitivity, or other similar visual issues will find that the Wikipedia iOS app is WCAG AA compliant, features smart color inversion, and has several different appearance themes

​

Those with less-than-perfect eyesight can take advantage of dynamic type, which increases text size across the entire app, not just for articles.

​

Wikipedia iOS app named an Editor’s Choice

25 of 32

Suggested Edits:�Add Alt-Text

​

Long-term vision

26 of 32

Machine Assisted Article Descriptions (MAAD)

  • Most popular Suggested Edit released in 2017
  • Requests for interventions to improve quality
  • EPFL shared LLM that generates article descriptions in 2021
  • AB tested suggestions in the app
  • Patrollers provided quality scores for edits across several languages

27 of 32

MAAD Impact

Edit Type

Published Edits

Unique editors

Avg Quality Score

% of edits would be rewritten

% of edit would revert

7 day return rate

Machine Accepted Unmodified

895

149

4.3

25%

2.3%

24.1%

Machine Accepted Modified

552

118

4.1

38.4%

5.7%

​

Human Generated

6370

(2363 exposed to machine suggestion)

345

4.2

25.8%

15%

22.6%

28 of 32

Nov

2023

Dec

2023

Jan

2023

Engineer

Iterate

Evaluate

Consultation

Build prototype

Get feedback on prototype

​

Evaluate edits with accessibility specialists

Improve feature or pivot to a different Suggested edits type

Oct

2023

Mar.

2024

Release

Release to select communities

Feb

2023

Test

Get feedback

29 of 32

CONNECT WITH US.

MediaWiki:

[[Wikimedia_Apps/iOS_Suggested_edits_project]]

​

MetaWiki:

[[Wikimedia_Apps/Newsletter]]

30 of 32

Demo & Test Instructions

  1. Visit https://bit.ly/3suF1g4 to download the test build
  2. Tap the gear icon in the top right part of the screen after going through onboarding
  3. Click Login and sign in or create an account
  4. Tap Account once logged in
  5. Tap Suggested Edits
  6. Review feature onboarding
  7. Start Editing!
  8. Provide feedback at anytime via the survey accessed in the overflow menu

31 of 32

32 of 32