Despite efforts to increase the representation of disabled people in AI datasets, accessibility datasets are often annotated by crowdworkers without disability-specific expertise, leading to inconsistent or inaccurate labels. This paper examines these annotation challenges through a case study of annotating speech data from people who stutter (PWS). Given the variability of stuttering and differing views on how it manifests, annotating and transcribing stuttered speech remains difficult, even for trained professionals. Through interviews and co-design workshops with PWS and domain experts, we identify challenges in stuttered speech annotation and develop practices that integrate the lived experiences of PWS into the annotation process. Our findings highlight the value of embodied knowledge in improving dataset quality, while revealing tensions between the complexity of disability experiences and the rigidity of static labels. We conclude with implications for disability-first and multiplicity-aware approaches to data interpretation across the AI pipeline.

Videoconferencing (VC) has become a prominent and normalized mode of professional and personal communication, introducing universally experienced challenges such as reduced non-verbal cues and "Zoom Fatigue." But People who stutter (PWS) encounter these obstacles with extra hurdles as existing VC technologies often rely on assumptions about speech patterns that don’t accommodate stuttering. Leveraging and driven by the unique insights and experiences of PWS on VC, we conducted a two-phase co-design study with PWS to explore and reflect on the design space for inclusive and empowering VC technologies from their perspectives. Our findings present a broad design space for tools that support PWS before, during, and after VC, focusing on aspects such as supporting self-disclosure, educating non-stuttering audiences, and promoting personal reflection for long-term self-growth. While many design ideas by our participants embody universal value to all VC users, some carry an activism approach that proactively disrupts existing communication flows and norms to redistribute the power between stuttering and non-stuttering speakers in VC meetings. This work contributes to a thorough analysis of the design space and empowering PWS to be drivers and designers of inclusive VC experiences.

Existing videoconferencing (VC) technologies are often optimized for productivity and efficiency, with little support for the “soft side” of VC meetings such as empathy, authenticity, belonging, and emotional connections. This paper presents findings from a 15-month long autoethnographic study of VC experiences by the first author, a person who stutters (PWS). Our research shed light on the hidden costs of VC for PWS, uncovering the substantial emotional and cognitive efforts that other meeting attendants are often unaware of. Recognizing the disproportionate burden on PWS to be heard in VC, we propose a set of design implications for a more inclusive communication environment, advocating for shared responsibility among all, including communication technologies, to ensure the inclusion and respect of every voice.

This paper documents the process undertaken by StammerTalk , a grassroots community of Chinese-speaking people who stutter, to autonomously collect and curate stuttered speech data for more inclusive speech AI models. While people with disabilities are often excluded or treated merely as the subjects of AI data collection, our work introduces a new model for disability data collection in which the disability community exerts agency and control over their personal data and data-driven experiences. Our ethnographic data show that community-led data collection not only produces data needed to represent the community in AI systems, but also empowers the community and its members, by embracing - rather than concealing - stuttering and stutterer identity, and strengthening the social bonds of the community. Recognizing the lack of adequate socio-technical infrastructure for community-led, grassroots data collection, we discuss practical challenges, as well as the strategies and factors for communities to succeed in similar endeavors.

This work studies the experiences of people who stutter (PWS) with videoconferencing (VC) and VC technologies. Our interview study with 13 adults who stutter uncovers extra challenges introduced by current VC platforms to people who stutter. While some of the challenges are a direct result of the characteristics of stuttering (e.g. people/systems mistaking pauses as end of turn), a bigger yet less visible challenge comes with the significant amount of emotional and cognitive effort required to manage one’s speech and identity over VC, in which people’s existing communication strategies - such as body language and eye contact - are under-supported and their biggest discomfort - such as seeing oneself stutter - are exacerbated by preset features like self view. Overall, our work sheds light on the structural barriers and the opportunities for PWS to engage and enjoy virtual communications via VC technologies.

People with dyslexia face challenges expressing themselves in writing on social networking sites (SNSs). Such challenges come from not only the technicality of writing, but also the self-representation aspect of sharing and communicating publicly on social networking sites such as Facebook. To empower people with dyslexia-style writing to express them-selves more confidently on SNSs, we designed and implemented Additional Writing Help(AWH) - a writing assistance tool to proofread text produced by users with dyslexia before they post on Facebook. AWH was powered by a neural machine translation (NMT) model that translates dyslexia style to non-dyslexia style writing. We evaluated the performance and the design of AWH through a week-long field study with 19 people with dyslexia and received highly positive feedback. Our field study demonstrated the value of providing better and more extensive writing support on SNSs, and the potential of AI for building a more inclusive Internet.

Recognizing others is a major challenge for people with visual impairments (VIPs) and can hinder engagement in social activities. We present Accessibility Bot, a research prototype bot on Facebook Messenger, that leverages state-of-the-art computer vision and a user's friends' tagged photos on Facebook to help people with visual impairments recognize their friends. Accessibility Bot provides users information about identity and facial expressions and attributes of friends captured by their phone's camera. To guide our design, we interviewed eight VIPs to understand their challenges and needs in social activities. After designing and implementing the bot, we conducted a diary study with six VIPs to study its use in everyday life. While most participants found the Bot helpful, their experience was undermined by perceived low recognition accuracy, difficulty aiming a camera, and lack of knowledge about the phone's status. We discuss these real-world challenges, identify suitable use cases for Accessibility Bot, and distill design implications for future face recognition applications.

We designed and deployed automatic alt-text (AAT), a system that applies computer vision technology to identify faces, objects, and themes from photos to generate photo alt-text for screen reader users on Facebook. We designed our system through iterations of prototyping and in-lab user studies. Our lab test participants had a positive reaction to our system and an enhanced experience with Facebook photos. We also evaluated our system through a two-week field study as part of the Facebook iOS app for 9K VoiceOver users. We randomly assigned them into control and test groups and collected two weeks of activity data and their survey feedback. The test group reported that photos on Facebook were easier to interpret and more engaging, and found Facebook more useful in general. Our system demonstrates that artificial intelligence can be used to enhance the experience for visually impaired users on social networking sites (SNSs), while also revealing the challenges with designing automated assistive technology in a SNS context.

Like sighted people, visually impaired people want to share photographs on social networking services, but find it difficult to identify and select photos from their albums. We aimed to address this problem by incorporating state-of-the-art computer-generated descriptions into Facebook's photo-sharing feature. We interviewed 12 visually impaired participants to understand their photo-sharing experiences and designed a photo description feature for the Facebook mobile application. We evaluated this feature with six participants in a seven-day diary study. We found that participants used the descriptions to recall and organize their photos, but they hesitated to upload photos without a sighted person's input. In addition to basic information about photo content, participants wanted to know more details about salient objects and people, and whether the photos reflected their personal aesthetic. We discuss these findings from the lens of self-disclosure and self-presentation theories and propose new computer vision research directions that will better support visual content sharing by visually impaired people.

In this paper, we explore blind people's motivations, challenges, interactions, and experiences with visual content on Social Networking Services (SNSs). We present findings from an interview study of 11 individuals and a survey study of 60 individuals, all with little to no functional vision. Compared to sighted SNS users, our blind participants faced profound accessibility challenges, including the prevalence of photos without sufficient text descriptions. To overcome the challenges, they developed creative strategies, including using a variety of methods to access SNS features (e.g., opening the mobile site on a desktop browser), and inferring photo content from textual cues and social interactions. When strategies failed, participants reached out for help from trusted friends, or avoided certain features. We discuss our findings in the context of CSCW research and SNS accessibility as a design value. We highlight the social significance of photo interactions for blind people and suggest design practices.

Apps are emerging as an important form of on-line content, and they combine aspects of Web usage in interesting ways --- they exhibit a rich temporal structure of user adoption and long-term engagement, and they exist in a broader social ecosystem that helps drive these patterns of adoption and engagement. It has been difficult, however, to study apps in their natural setting since this requires a simultaneous analysis of a large set of popular apps and the underlying social network they inhabit. In this work we address this challenge through an analysis of the collection of apps on Facebook Login, developing a novel framework for analyzing both temporal and social properties. At the temporal level, we develop a retention model that represents a user's tendency to return to an app using a very small parameter set. At the social level, we organize the space of apps along two fundamental axes --- popularity and sociality --- and we show how a user's probability of adopting an app depends both on properties of the local network structure and on the match between the user's attributes, his or her friends' attributes, and the dominant attributes within the app's user population. We also devolop models that show the importance of different feature sets with strong performance in predicting app success.

In this paper we present the first large-scale empirical study of how visually impaired people use online social networks, specifically Facebook. We identify a sample of 50K visually impaired users, and study the activities they perform, the content they produce, and the friendship networks they build on Facebook. We find that visually impaired users participate on Facebook (e.g. status updates, comments, likes) as much as the general population, and receive more feedback (i.e., comments and likes) on average on their content. By analyzing the content produced by visually impaired users, we find that they share their experience and issues related to vision impairment. We also identify distinctive patterns in their language and technology use. We also show that, compared to other users, visually impaired users have smaller social networks, but such differences have decreased over time. Our findings have implications for improving the utility and usability of online social networks for visually impaired users.

In this paper, we consider the natural arrival and departure of users in a social network, and ask whether the dynamics of arrival, which have been studied in some depth, also explain the dynamics of departure, which are not as well studied.

We study several longstanding questions in media communications research, in the context of the microblogging service Twitter, regarding the production, flow, and consumption of information. To do so, we exploit a recently introduced feature of Twitter known as "lists" to distinguish between elite users - by which we mean celebrities, bloggers, and representatives of media outlets and other formal organizations - and ordinary users. Based on this classification, we find a striking concentration of attention on Twitter, in that roughly 50% of URLs consumed are generated by just 20K elite users, where the media produces the most information, but celebrities are the most followed. We also find significant homophily within categories: celebrities listen to celebrities, while bloggers listen to bloggers etc; however, bloggers in general rebroadcast more information than the other categories. Next we re-examine the classical "two-step flow" theory of communications, finding considerable support for it on Twitter. Third, we find that URLs broadcast by different categories of users or containing different types of content exhibit systematically different lifespans. And finally, we examine the attention paid by the different user categories to different news topics.

In this paper, we introduce Z-ring, a fast prefix routing protocol for peer-to-peer overlay networks. Z-ring incorporates cost-efficient membership protocol to achieve fast routing with small maintenance cost. Z-ring achieves routing in logGN steps, where N is the network size and G is the size of a group that can be maintained by a membership protocol with low cost. With G=4096, it translates to one-hop routing for intranet environments (N<4096), two-hop routing for mid-scale internet applications (N<16 million), and three-hop routing for ultra-large Internet applications (N<64 billion). Z-ring maintains good routing success rate under churn and low maintenance cost even at large network size. Its modularized use of the membership protocol also makes it adaptive to dynamic and wide-range network size changes.