Future Research Directions for Accessible Social Media

Organizers of the "Addressing the Accessibility of Social Media" Workshop at CSCW 2019

Abstract

Social media platforms are deeply ingrained in society, and they offer many different spaces for people to engage with others. Unfortunately, accessibility barriers prevent people with disabilities from fully participating in these spaces. Social media users commonly post inaccessible media, including videos without captions (which are important for people who are Deaf or Hard of Hearing) and images without alternative text (descriptions read aloud by screen readers for people who are blind). Users with motor impairments must find workarounds to deal with the complex user interfaces of these platforms, and users with cognitive disabilities may face barriers to composing and sharing information.

We invited accessibility researchers, industry practitioners, and end-users with disabilities to come together at the Computer-Supported Cooperative Work conference (CSCW 2019) to discuss challenges and solutions for improving social media accessibility. Over the course of a day that included two panels and breakout sessions, the workshop attendees outlined four critical future research directions to progress on the path to accessible social media: tooling to support disabled people authoring content, developing more accessible formats/tools for new forms of interaction (e.g, Augmented and Mixed Reality), using communities to distribute accessibility labor, and ensuring machine learning systems are built on representative datasets for disability use-cases.

Introduction

Social media platforms have grown from niche online gathering spaces to public forums for all manner of discourse, yet they are often inaccessible to people with disabilities. The high prevalence of user-generated content such as images, videos, memes, GIFs, and emoji can create accessibility barriers for many users with vision or hearing impairments [1,9,22]. Furthermore, the interfaces for social media platforms are more complex than many standard websites, and people with vision, motor, or cognitive impairments may find them difficult to use.

Recent research has demonstrated that social media platforms have become more inaccessible over time [19], and although some platform features seek to improve this (e.g., the ability to include alternative text descriptions) people are not widely using them [1,7]. In contrast, other high-traffic websites often have alternative text on 40% to 70% of images [2,10]. Social media platforms have the clout and potential impact to substantially improve content accessibility by introducing new standards and approaches. Alternative text for images has mostly remained unchanged but richer representations are possible with modern technology [18]. For instance, Facebook and Instagram have introduced automatic description of images [25], leading to high coverage, although users must have high trust in these automatic systems to add value [17].

While controlling the accessibility of user-generated content is a challenge, social media companies can influence their interface design. In general, blind Facebook users easily navigate the site to post updates and share photos [23], but face other challenges such as searching or creating profiles [3]. Twitter users with physical or communication disabilities who use augmentative and alternative communication (AAC) devices enjoy interacting in close-knit social networks and exhibit different posting and sharing behavior, which might be linked to how they chose to access the platform [12-14]. Facebook users with dyslexia had difficulties with reading content on the platform, but especially with writing content [21], further emphasising the need for accessible interface design, which can improve readability and understanding for disabled people [20]. Evidence of the success of this approach was the design and evaluation of a writing assistance tool to help participants with dyslexia proofread their writing when posting, which increased their confidence [24].

An important reason to improve the accessibility of social media platforms is that people with disabilities often rely on them for critical activism and support. On Twitter, groups of disability advocates have convened around hashtags like #CripTheVote [6] and #HandsOfMyADA [1] to organize political support of disability legislation. Online activism also can be a more accessible outlet for political advocacy than physical marches or protests, so people with a disability may turn to virtual activism more often [15]. Social media platforms are also a place to seek social support online about accessibility barriers and solutions [16]. Campaigns like #SayTheWord use social media to change the public perception and awareness of people with disabilities, calling for a unified identity and a recognition of a critical constituency of people (25% of US adults [5]).

To address these challenges, we organised a CSCW workshop called "Addressing the Accessibility of Social Media" [8]. Our workshop brought together key stakeholders, including researchers and practitioners, to discuss interface and content barriers, and to propose research agendas to address the challenges in this space because social media platforms are an important venue for both daily communication and disability activism. Our workshop participants highlighted future research directions in authoring and encountering content, the intersection of advocacy and activism, and the future of AI and automation on social media.

Workshop Format

Our workshop was scheduled as a full day event broken up into five sections: an end-user panel, lighting talks from all attendees, Q&A panel with social media company representatives, and breakout groups to identify challenges and generate ideas for future work. A detailed structure of the workshop is available at Accessible Social Media. Below we summarize the discussion at both panels and detail the future research directions identified in the breakout groups.

Participants

The workshop drew over 25 attendees including organizers, panellists, and participants who represented several universities and social media companies. The participants submitted either a position paper, or they could apply via an online form option as a way to maximise the diversity of our workshop attendees. We recognized that relevant stakeholders such as disabled people and industry professionals need to be included in the discussion of social media accessibility and the requirement of a position paper was likely to appeal primarily to an academic audience. Our decision to offer this alternative submission format was validated by the engaging discussion among the academics, disability activists, and industry professionals. The participants presented on a range of relevant topics from disability activists organizing in virtual spaces to empowering adults on the autism spectrum to use social media.

Panel: Perspectives of Success and Barriers

This panel consisted of three individuals who are social media users who self-identified as having disabilities. SIGACCESS sponsored this panel and paid for their travel and appearance.

• Chancey Fleet coordinates technology education programs at the New York Public Library's Andrew Heiskell Braille and Talking Book Library, is president of NFB tech trainers' division, and affiliate at Data & Society, a non-profit research institute (Data Society)
• Rikki Poynter is a streamer, consultant, and public speaker on YouTube. Rikki makes content about deaf awareness, accessibility, and closed captions (Rikki Poynter).
• Nicolas Steenhout is a speaker, trainer and consultant on web accessibility, inclusion and disability. Nicolas also hosts the A11yRules Podcast (a11yrules) where he talks and learns to people working in accessibility (incl.ca).

We will briefly summarize the major themes from the panel and responses from panelists.

Accessing and Authoring Content

All panelists stressed that content on social media should be accessible, yet it often falls far short due to a lack of human-created accessible formats or poor automated systems. Nicolas discussed the lack of alt text online -- a standard that has existed since 1995 but we still struggle to implement well. He noted that it took Twitter many years to add an alt text feature, and how alt text was not available for GIFs (which has been announced since the workshop). Rikki noted that it is difficult to make content accessible, and she tries to add captions, voiceovers, or sign language interpretation to many of her videos. However, it takes time and money to make them accessible, and she is glad to have a sponsor for her voiceovers. Chancey highlighted Youtube, a site she often peruses to find how-to videos, but many lack voiced step-by-step instructions and feature only background music.

The panelists raised concerns with automated systems to make content accessible, as they can lead to bad experiences. Chancey said:

"I worry that automated image detection in social media is being deployed with a vigor that is not commensurate with its effectiveness. I worry that people who have read in the press that we have image [descriptions] on platforms imagine that it's more effective than it is and they may be deterred from contributing their own actually meaningful, vivid image descriptions."" -- Chancey Fleet

Nicolas connected this issue with automated transcription services, noting that he tested 17 different automated services to create transcripts for podcasts, finding that they were all poor. Rikki also has experience using automated speech recognition, primarily to create video captions. She used them on Twitch previously while live-streaming, but after an incident where it incorrectly transcribed her speech as a racist statement, she disabled automatic captioning.

Because content lacks accessible alternatives and automated systems are poor, the panelists reported using several workarounds to access content. Chancey previously tried to engage users on Twitter to ask what the visual content contained, but they frequently did not respond. She instead uses visual interpreting services (e.g., AIRA, BeMyEyes, Despecular). When streaming on Twitch, Rikki has a channel moderator transcribe captions during fast-paced games, although she wishes there was funding to compensate him for the effort. Nicolas and Chancey both shared that social media itself is their primary coping strategy. They have built communities on these sites where they can connect with others, although Nicolas notes that he sometimes must take time away from social media due to harmful ableist perspectives and language.

Education & Advocacy

The panelists discussed the need to educate both social media users and platforms to achieve a more accessible future. Nicolas said he thinks education must happen one person at a time, educating them about the need for accessibility while building strong relationships to cement the importance. However, he noted that this education is often a burden on people with disabilities. Rikki pointed out that this is a major benefit of YouTube -- sharing previously recorded videos and tutorials for accessibility helps avoid having to repeat herself. Chancey believes the correct path is through the platforms, and user education efforts must be led by them. If they build user experiences and support tools that highlight accessibility features, such as algorithms that promote accessible content, users will be more aware and incentivised to invest in accessibility.

As this discussion of education and advocacy efforts highlighted the immense burden placed on people with disabilities, an audience member asked how communities can better support people with disabilities on social media. Chancey discussed how a platform like Twitter could empower third-party users to write image descriptions when the original author doesn't, possibly with approval before becoming publicly available. An addition of features that support community-generated accessibility would help distribute the work of accessibility online, and create the norm of making other users' content accessible.

Privacy

In response to an audience question, the panelists discussed the issues of disability status privacy on social media. While all three of them are vocal about their disability and advocate for accessibility on some platforms, they had concerns about weaponization of disability status. Nicolas noted that many businesses want to know which website visitors use screen readers as a justification for making something accessible, but these screenreader detections could also be used to exclude blind users from using certain features, giving them a more limited experience. Unfortunately, after the workshop ended, Chancey and others reported that the Facebook Avatar feature used screen reader detection to hide the feature from blind users on iOS devices.

Rikki noted that she is not very vocal about her disability in specific communities, especially certain toxic ones on Twitch. She recounted a case where she couldn't hear a Twitch stream because of low volume, but was concerned about harassment if she asked the streamer to increase the volume in a public chat. Finally, Chancey raised the issue of ad-microtargeting, which has been used on Facebook to illegally exclude people with disabilities from housing advertisements. Ad-microtargeting can often be very opaque, and there are often ways for advertisers to skirt the rules when designing ad targeting campaigns. Platforms need to be more vigilant about how advertisers are including/excluding groups of people from seeing certain ads.

What's exciting?

The panel ended looking to the future, and the panelists recounted what they were excited about.

"I'm really excited about [Automated Systems] and I'm very fearful about it: how can we leverage machine learning and AI to properly do accessibility? I think there are a lot of possibilities out there that will make a difference if we do it properly." -- Nicolas Steenhout

Rikki discussed a particular piece of AI that excites her: Google's Live Transcribe:

It was really nice to have that piece of technology be right there on the table. And as long as not everybody was trying to talk at the same time, it was so good. -- Rikki Poynter

Chancey was excited about social media accessibility being taken more seriously, especially seeing industry allies attend events like our workshop and advocate for projects that increase the accessibility of platforms. Additionally, the next generation of people with disabilities are growing up with more native interactions with technology. Because they have a level of fundamental baseline accessibility they will be able to take for granted, they will be able to imagine, build, and demand better platforms and technology in the future.

Accessing and Authoring Content

Our second panel featured three individuals who work on accessibility at various social media companies:

• Jennison Asuncion is the Head of Accessibility Engineering Evangelism at LinkedIn.
• Jerry Robinson is a User Experience Researcher at Facebook.
• Jess Harvie is a Diversity, Equity and Inclusion Program Manager at Snap, Inc

We invited these industry professionals to talk about their role as advocates for accessibility inside their respective companies, and the work they do to make their platforms more accessible.

Product Development

All three panelists reiterated throughout the panel that accessibility is never finished and is a journey within product development at their respective companies. At LinkedIn, according to Jennison, some products are more mature in their accessibility than others, but they prefer to design inclusively from the beginning as it is hard to add accessibility after a product is built. Jerry added that at Facebook, the product managers are told to consider accessibility from the beginning to the end of the product development cycle. While Snap is early in its accessibility journey, Jess says they aim to design and develop experiences that are not sub-optimal for any individual based on their abilities.

Panelists acknowledged the difficulties of embedding accessibility within product development due to the sheer number of products and speed at which they are launched. LinkedIn pushes out code to their products daily and to their mobile apps weekly, and these updates could lead to broken aspects when viewed under an accessibility lens. As the head of Accessibility Engineering Evangelism, Jennsion has the power to determine the severity and priority of accessibility bugs, but they are also proactive with automated testing. This only catches a small proportion of potential issues, but it ensures the "low hanging fruit" problems are avoided. Jess added that within Snap, teams are busy working on developing new tools, features, and other fixes that accessibility issues are not always prioritized high enough.

Team Structure

Panelists discussed the different approaches that each of their companies had towards structuring their teams. Jerry and Jennison said that Facebook and LinkedIn both have central teams that focus on accessibility to create tools, prose, and best practices for the company. It's these teams that empower specific product teams to build and maintain accessibility into individual products. Snap does not currently have a separate accessibility team, and accessibility efforts are driven by those within the company who are passionate about taking responsibility for that aspect of the product. Those individuals can have their performance objectives partially-focused on accessibility. Snap is looking at increasing diversity, equity, and inclusion (DEI) expertise amongst all teams that are creating products.

Training

At Facebook and LinkedIn, engineers and other employees are first introduced to accessibility during the onboarding process, and there are other opportunities to learn about accessibility. LinkedIn has a program called "Accessibility Champions", where employees can spend 25% of their time in a quarter learning about accessibility and focusing on those tasks. They also have an "Accessibility Lab Experience" where employees can have hands-on experience with assistive technology to build understanding of the technology and empathy with users. Jennison called on the research community to teach accessibility as part of the computer science curriculum, ensuring that engineers who go to industry have some basic knowledge and interest.

Snap doesn't yet have specific accessibility training, but may in the future. Their philosophy is to train employees to have both resources and empathy for groups different they do not represent. Jennison drives this point home:

"The moments that I think are the most impactful are, whether it's someone who's trying out a piece of technology themselves or if they're watching an end user with a disability or impairment trying to use a piece of technology, above and beyond just like seeing the standards and guidelines and best practices, [. . .] Maybe the person isn't using assistive technology but they're still struggling using a product." That's when people go 'I get it.'" -- Jennison Asuncion

Engaging with Community Feedback

As their companies push out new code with features or bug fixes regularly, they sometimes break accessibility features or add-ons that disabled users rely on. Both LinkedIn and Facebook have some sort of online feedback mechanisms where people can report issues with the platforms, and these will be triaged to the correct person. User research is also a great way to solicit feedback, whether that is in cooperation with a local non-profit organization or field research with end-users. Jerry noted that it is important to get the perspectives of users during the design process, but also to bring product team stakeholders along for field research to ensure they all spend some time participating in user research.

Future Research Directions

After the panel discussions, workshop participants broke into groups organized around topics that surfaced earlier in the day. They were assigned to groups based on their background and interests and encouraged to discuss the relevant challenges for both industry and academia in that topic. Each group presented one challenge they believed to be critically important or understudied. We have summarized their discussions of future research directions below as a means to highlight where industry and academics could advance social media accessibility.

Authoring Content: Automatic or Human Feedback Before Publishing

Within web accessibility, consuming content is often the focus of accessibility efforts, but ensuring that people with disabilities have accessible authoring tools is critical to create equal access on social media platforms. This breakout group focused on discussing the challenges involved in authoring content, and they specifically identified a gap in research and practice in ensuring people with disabilities are confident that their content is presented as intended. People with vision impairments worry about sharing images including content they didn't intend to, such as private information or the wrong image entirely [11]. People with dyslexia are more confident and post more often when they know that their post has been automatically checked [24]. Automated and human review tools should be available so that users with disabilities can seek feedback on their posts before publishing them.

As an example, automated analysis of images and text could ensure the user is posting what they intend. This might alert the user to a mismatch between a caption about a "fun day at the conference!", when the photo instead is of a beach.. It could also pick up on unintended typos, formatting, or capitalizations that may be unintended, helping to ensure the user is presenting the information they mean to. As not all issues can be determined automatically, the user could have the option to share their draft post with either a specific individual or a trusted group of friends, which might be important if the social media post was in a professional setting. Tools that enable automatic or human review of their posts before publishing would ideally lead to a more accessible experience where social media users felt more in control of their feeds.

Encountering Content: Making Novel Content Formats Accessible

By far the most common activity anyone performs on social media is encountering content from other users, and this breakout group discussed three challenges when encountering content. The first is how to disseminate existing research and standards for basic digital accessibility to social media platforms and ensure they are implemented correctly. The second is developing new content mediums that can be accessible at the time of creation instead of using additional modalities added later (alt text, captions) to make an accessible version. And finally, this group focused on how new mediums in social media are presenting accessibility barriers, namely augmented reality content is not accessible to users with a vision impairment.

While not yet as widespread as images or videos, augmented reality (AR) content has made inroads on many social media platforms, and it is a prime example of a content medium that is being designed without accessibility considered from the start. This group considered how image and video metadata could contain accessible information about AR content. Just as camera information is embedded in photographs, information about the authoring app and AR filters or animations used can be stored alongside visual information. Automated alternative text tools can use that metadata to generate a description of the AR content, no matter what platform the image or video is shared on. The group hopes that accessible AR implementations would result in increased interaction by people with vision impairments, and qualitative evaluations with both senders and receivers of AR content would reveal more accessible communications.

Advocacy & Activism: Communities to Distribute Accessibility Labor

Social media provides a vehicle for organizing disability activism and advocacy, especially for the accessibility of social media itself. This group discussed several challenges, but focused on the labor of accessibility tasks. People with disabilities often must take up the role of educating users on social media platforms about accessibility and advocating for specific pieces of content to be remade in an accessible manner. Instead, how can communities of volunteers and social media followers support the creation of accessible content to distribute that work?

The group suggested that researchers could create and evaluate tools to connect content authors with community members who take on the accessibility labor. If someone posts content on Twitter or streams on Youtube, they might allow anyone to suggest alternative text or captions. To prevent bad actors from taking advantage of this, the content creator could restrict this to a group of trusted individuals or mutual friends. Moderation options could also vary, allowing community-authored accessibility information visible to anyone, only visible with a certain number of upvotes, or requiring approval by the author. If community-based creation of accessible content is successful, authors should see high-quality image descriptions or video captions, and this quality should increase over time as good content is shared more widely.

AI and Automation: Representative Datasets for More Equal AI Models

Artificial intelligence (AI) and machine learning (ML) may have roles in making social media platforms more accessible, and have already been deployed to generate image descriptions and captions automatically [25]. However, many challenges remain to ensure they are accurate and equitable. The datasets these models are trained on were not collected with a disability use-case in mind. When they are deployed, it is difficult for people with disabilities to evaluate when they can be trusted [17], and difficult for developers to define metrics indicating when they are safe to deploy. Finally, there is a concern that AI systems could be weaponized to diagnose/out people with disabilities without their consent based on their behavior on social media [19].

This breakout group focused their discussion on the collection of data to train AI systems. These datasets are often collected without a disability use-case in mind, leading to inaccuracies such as wheelchairs classified as bikes. Instead, institutions must find ways to motivate the collection of datasets to better conform to their use-cases. Possible models for this include asking users to explicitly contribute their data to a collection effort, non-profits and universities working with disability communities to create open-access and representative datasets, and contests or gamifications to incentivise contributions. These datasets, which should be more representative if collected for a disability use-case, can then be synthesized or added to other training datasets to improve the quality of systems like automatic captions for images or videos. If this method of soliciting data contributions to build better AI systems is successful, we could evaluate models trained on this new data alongside prior ones, and see if they handle disability use-cases with better results (e.g., properly labelling wheelchairs). But the true measure of success is whether these controlled and specialized datasets could lead to new innovation in the field of AI and accessibility.

Conclusion

Progress has been made in creating more accessible social media platforms, but issues remain in many aspects, especially in user-generated content on the platforms. Our panelists and workshop participants reiterated that many aspects of accessibility are well-known, yet remain un-implemented. Research in these areas should focus on how to widely disseminate and ensure implementation of accessibility best practices at both the platform and individual level. The role of machine learning and AI in making platforms accessible remains unclear, as it can provide much benefit at the possible expense of accuracy or political will for future changes. As social media changes and evolves, birthing new formats, the design process must incorporate accessibility from the start to guarantee accessible social media for all.

Acknowledgments

Thanks to SIGACCESS and the Trace R&D Center for supporting this workshop. Thank you to our panelists for sharing their knowledge and experiences: Chancey, Rikki, Nicolas, Jennison, Jerry, and Jess . Finally, thank you to all participants for attending and making the workshop a success.

References

1. Brooke E. Auxier, Cody L. Buntain, Paul Jaeger, Jennifer Golbeck, and Hernisa Kacorri. 2019. #HandsOfMyADA: A Twitter Response to the ADA Education and Reform Act. Proc. CHI ’19. ACM, New York, NY, USA, Article 527, 12 pages. htps://doi.org/10.1145/3290605.3300757
2. Jeffrey P. Bigham, Ryan S. Kaminsky, Richard E. Ladner, Oscar M. Danielsson, and Gordon L. Hempton. 2006. WebInSight: making web images accessible. Proc. ASSETS’06. ACM, New York, NY, USA, 181–188. DOI:https://doi.org/10.1145/1168987.1169018
3. Julian Brinkley and Nasseh Tabrizi. 2017. A Desktop Usability Evaluation of the Facebook Mobile Interface using the JAWS Screen Reader with Blind Users. Proc. of the Human Factors and Ergonomics Society Annual Meeting 61, 1 (2017), 828–832. htps://doi.org/10.1177/1541931213601699
4. Maria Claudia Buzzi, Marina Buzzi, and Barbara Leporini. 2011. Web 2.0: Twitter and the blind. Proc. CHItaly’11. ACM, New York, NY, USA, 151–156. DOI:https://doi.org/10.1145/2037296.2037333
5. Centers for Disease Control and Prevention. 2019. CDC: 1 in 4 US adults live with a disability. (2019). Retrieved June 27, 2019 from htps://www.cdc.gov/media/releases/2018/p0816-disability.html.
6. Amber Ferguson. 2016. The #CripTheVote Movement Is Bringing Disability Rights To The 2016 Election. (2016). Retrieved August 20, 2018 from https://www.huffingtonpost.com/entry/cripthevote-movement-2016-election_us_57279637e4b0f309baf177bd
7. Cole Gleason, Patrick Carrington, Cameron Cassidy, Meredith Ringel Morris, Kris M. Kitani, and Jeffrey P. Bigham. 2019. “It’s almost like they’re trying to hide it”: How User-Provided Image Descriptions Have Failed to Make Twitter Accessible. Proc. WWW ’19. ACM, New York, NY, USA, 549–559. DOI:https://doi.org/10.1145/3308558.3313605
8. Cole Gleason, Patrick Carrington, Lydia B. Chilton, Benjamin M. Gorman, Hernisa Kacorri, Andrés Monroy-Hernández, Meredith Ringel Morris, Garreth W. Tigwell, and Shaomei Wu. 2019. Addressing the Accessibility of Social Media. In Proc. Companion CSCW ’19. ACM, New York, NY, USA, 474–479. DOI:https://doi.org/10.1145/3311957.3359439
9. Cole Gleason, Amy Pavel, Xingyu Liu, Patrick Carrington, Lydia B. Chilton, and Jeffrey P. Bigham. 2019. Making Memes Accessible. Proc. ASSETS ’19. ACM, New York, NY, USA, 367–376. DOI:https://doi.org/10.1145/3308561.3353792
10. Darren Guinness, Edward Cutrell, and Meredith Ringel Morris. 2018. Caption Crawler: Enabling Reusable Alternative Text Descriptions using Reverse Image Search. Proc. CHI ’18. ACM, New York, NY, USA, Paper 518, 1–11. DOI:https://doi.org/10.1145/3173574.3174092
11. Danna Gurari, Qing Li, Chi Lin, Yinan Zhao, Anhong Guo, Abigale Stangl, and Jeffrey P Bigham. 2019. VizWiz-Priv: A Dataset for Recognizing the Presence and Purpose of Private Visual Information in Images Taken by Blind People. 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) 00: 939–948. https://doi.org/10.1109/cvpr.2019.00103
12. Bronwyn Hemsley, Stephen Dann, Stuart Palmer, Meredith Allan, and Susan Balandin. 2015. “We definitely need an audience”: experiences of Twitter, Twitter networks and tweet content in adults with severe communication disabilities who use augmentative and alternative communication (AAC). Disability and rehabilitation 37, 17 (2015), 1531–1542.
13. Amanda Hynan, Juliet Goldbart, and Janice Murray. 2015. A grounded theory of Internet and social media use by young people who use augmentative and alternative communication (AAC). Disability and Rehabilitation 37, 17 (2015), 1559–1575. htps://doi.org/10.3109/09638288.2015.1056387
14. Amanda Hynan, Janice Murray, and Juliet Goldbart. 2014. ‘Happy and excited’: Perceptions of using digital technology and social media by young people who use augmentative and alternative communication. Child Language Teaching and Therapy 30, 2 (2014), 175–186. https://doi.org/10.1177/0265659013519258
15. Hanlin Li, Disha Bora, Sagar Salvi, and Erin Brady. 2018. Slacktivists or Activists? Identity Work in the Virtual Disability March. Proc. CHI ’18. ACM, New York, NY, USA, Paper 225, 1–13. DOI:https://doi.org/10.1145/3173574.3173799
16. Hanlin Li and Erin Brady. 2016. #accessibilityFail: Categorizing Shared Photographs of Physical Accessibility Problems. Proc. ASSETS ’16. ACM, New York, NY, USA, 277–278. DOI:https://doi.org/10.1145/2982142.2982186
17. Haley MacLeod, Cynthia L. Bennett, Meredith Ringel Morris, and Edward Cutrell. 2017. Understanding Blind People’s Experiences with Computer-Generated Captions of Social Media Images. Proc. CHI ’17. ACM, New York, NY, USA, 5988–5999. DOI:https://doi.org/10.1145/3025453.3025814
18. Meredith Ringel Morris, Jazette Johnson, Cynthia L. Bennett, and Edward Cutrell. 2018. Rich Representations of Visual Content for Screen Reader Users. Proc. CHI ’18. ACM, New York, NY, USA, Paper 59, 1–11. DOI:https://doi.org/10.1145/3173574.3173633
19. Meredith Ringel Morris, Annuska Zolyomi, Catherine Yao, Sina Bahram, Jeffrey P. Bigham, and Shaun K. Kane. 2016. “With most of it being pictures now, I rarely use it”: Understanding Twitter’s Evolving Accessibility to Blind Users. Proc. CHI ’16. ACM, New York, NY, USA, 5506–5516. DOI:https://doi.org/10.1145/2858036.2858116
20. Imke Niediek. 2016. Don’t write it, picture it! Accessible Information by graphic signs. Proc. DSAI’16. ACM, New York, NY, USA, 188–193. DOI:https://doi.org/10.1145/3019943.3019971
21. Lindsay Reynolds and Shamoei Wu. 2018. “I’m Never Happy with What I Write”: Challenges and strategies of people with dyslexia on social media. Proc. ICWSM’18. htps://www.aaai.org/ocs/index.php/ICWSM/ICWSM18/paper/view/17812
22. Garreth W. Tigwell, Benjamin M. Gorman, and Rachel Menzies. 2020. Emoji Accessibility for Visually Impaired People. Proc. CHI ’20. ACM, New York, NY, USA, 1–14. DOI:https://doi.org/10.1145/3313831.3376267
23. Shaomei Wu and Lada A. Adamic. 2014. Visually impaired users on an online social network. Proc. CHI ’14. ACM, New York, NY, USA, 3133–3142. DOI:https://doi.org/10.1145/2556288.2557415
24. Shaomei Wu, Lindsay Reynolds, Xian Li, and Francisco Guzmán. 2019. Design and Evaluation of a Social Media Writing Support Tool for People with Dyslexia. Proc. CHI ’19. ACM, New York, NY, USA, Paper 516, 1–14. DOI:https://doi.org/10.1145/3290605.3300746
25. Shaomei Wu, Jeffrey Wieland, Omid Farivar, and Julie Schiller. 2017. Automatic Alt-text: Computer-generated Image Descriptions for Blind Users on a Social Network Service. Proc. CSCW ’17. ACM, New York, NY, USA, 1180–1192. DOI:https://doi.org/10.1145/2998181.2998364

About the Authors