2026-10-21: From Accessible to Usable: A PhD Journey in Non-Visual Web Interaction

By -

Figure 1: With my committee after the defense on July 23, 2026. From left: Dr. Michael L. Nelson, me, Dr. Michele C. Weigle and Dr. Vikas Ashok. Dr. Hae-Na Lee joined remotely.

In Fall 2019, I moved from Dhaka to Norfolk to begin my PhD in Computer Science at Old Dominion University. I introduced myself in a WS-DL blog post in August 2020. I came in with a background in machine learning applied to human behavior, from my master's at American International University–Bangladesh and my undergraduate work at East West University. What I wanted from a PhD was to build practical systems that people use, and to study how they use them.

In January 2020, I joined Dr. Vikas Ashok's group, part of the Web Science and Digital Libraries (WS-DL) Research Group in the Department of Computer Science. Dr. Ashok's Human-Centered Computing Lab works on accessible computing: making computing devices usable for people with visual disabilities, so they benefit from the same technological advances as sighted users, at the intersection of human-computer interaction and artificial intelligence. That was the right place for what I wanted to do, and web accessibility has been my focus ever since. It became my dissertation, "Proxy Interfaces for Enhancing Non-Visual Web Usability."


Reachable Is Not Usable

Accessibility research and practice have focused on making content reachable: standards such as WCAG, checklists, compliance tooling. That work matters, and it has largely succeeded, but a page that passes those checks can still be very hard to use.

Sighted users take in a page's structure before they read a word of it. Layout, grouping, and visual cues tell them where things are and what to skip, and that costs them almost no effort. A screen reader delivers the same page linearly, one element at a time, as speech. Everything the layout was doing has to be rebuilt in the user's head as they go. The user has to remember more, loses their place more easily, and struggles with tasks that are trivial for someone who can see the page.

A blind user can reach every element on a compliant page and still give up before finishing a task.

The Research Challenge

My research asked one question: if we cannot fix every website, can we put something between the website and the screen reader that makes the page usable?

The idea of placing a proxy between a page and a screen reader is not new: earlier work used transcoding proxies to adapt pages for non-visual access. I call my version of that layer a proxy interface. It finds the semantically meaningful units on a page, reorganizes them into a stable representation aligned with how a screen reader is actually used, and exposes them through predictable interaction. It does not depend on how the page was authored.

This work is guided by three research questions.

RQ1: How can proxy interfaces support efficient non-visual navigation in arbitrary websites for blind screen reader users?

Blind users move linearly through long layouts to reach the section they want, and repeat that work on every visit. I built sTag, a browser extension that lets a user tag the page segments that matter to them and generates a personal Table-of-Contents overlay from those tags, independent of the page's markup. sTag also learns from user-created TOCs and reuses them to generate TOCs automatically for similar pages.

In a study with 15 blind participants, sTag significantly reduced both navigation effort and task time. Two results I had not predicted: 13 of the 15 participants used a single input modality instead of switching between keyboard and pointer, and six asked for a hierarchical TOC instead of a flat list so they could assign priority levels to their skip links. This work was published at ACM SAC 2021: "Semantic Table-of-Contents for Efficient Web Screen Reading," and I presented it at the 2021 WS-DL Research Expo.

RQ2: How can proxy interfaces improve non-visual interaction with structured web data, such as product lists or other data records requiring comparison?

Product listings, job results and classifieds are comparison tasks, and the information needed is split between the records themselves and the auxiliary segments around them: filters, sort options, page links. With a screen reader, every trip between those regions costs a traversal of the page.

I built InSupport, which uses machine learning to detect coherent data-record segments and reorganizes the auxiliary segments into a simplified proxy interface with natural-language querying, reachable instantly from anywhere on the page.

I evaluated it with two populations. With 14 blind participants using JAWS, average task time on the first task fell from 780.6 seconds with the screen reader alone and 540.9 seconds with the SaIL baseline to 288.6 seconds with InSupport; input actions fell from 611.9 and 395.9 to 187.2; and SUS rose from 53.9 and 68.9 to 86.1 (one-way ANOVA, F = 22.865, p < 0.0001). With 16 low-vision participants using screen magnifiers, with acuity from 20/100 to 20/700, SUS rose from 48.8 with the magnifier and 63.3 with space compaction to 84.2 with InSupport (F = 33.45, p < 0.001).

Twelve of the 14 blind participants valued having the auxiliary segments somewhere separate. As P4 put it: "I only have to remember the layout of content in each item of the results, and not the entire webpage." This work was published at ACM IUI 2022: "InSupport: Proxy Interface for Enabling Efficient Non-Visual Interaction with Web Data Records," and extended to cover both populations in ACM TiiS 2023: "Enabling Efficient Web Data-Record Interaction for People with Visual Impairments via Proxy Interfaces."


I presented the paper at the 2022 WS-DL Research Expo that April, and blogged about InSupport in July 2023.

InSupport was also the subject of my PhD candidacy exam in July 2023.


RQ3: How can proxy interfaces enhance blind users' ability to understand, follow, and participate in online multi-party discussions where context and reply relationships are critical?

Before building anything here, I needed to know what the actual barriers were, so I ran semi-structured interviews with 20 blind screen reader users about their experience of online discussions.

Most participants (18 out of 20) said it was difficult to join a conversation already in progress, because of the listening effort required to work out where the discussion had gone. Sixteen out of 20 said their experience depended heavily on the type of discussion, with live threads far worse than asynchronous ones, and 16 said they did not feel productive or effective in online discussions at all. Participants preferred longer, context-rich posts, because those restate key details and reduce the need to jump back through a thread. What they asked for was summarization on demand, explicit relational cues, cleaner text for listening, and structure they could navigate selectively.

This work was published in the International Journal of Human–Computer Interaction in 2025: "Understanding Online Discussion Experiences of Blind Screen Reader Users." I also blogged about it: "Beyond Alt-Text: Surprising Truths About How Blind Users Experience Online Discussions."

I turned the interview findings into five design goals and realized them as Standpoint, a browser extension that generates conversational context over a discussion thread on demand, anchored to the reader's current position and computed entirely on the reader's own device. It produces position-scoped summaries and explicit relation labels over the thread structure.

The technical evaluation showed the generated structure was reliable while the relation labels were uneven; I traced that to prompt design and corrected it there. In a pilot with five blind users, all five navigated large threads with less time and effort, and described a shift from grasping a discussion's mood to grasping its argument structure. They checked the relation labels, which they could verify, and were more cautious about trusting the AI-generated summaries. A large-scale controlled evaluation is the primary next step.

The Proxy Interface Pattern

The three systems are instances of one pattern. Each identifies structure that the page does not expose to a screen reader and turns it into something the user can work with:
  • sTag identifies user-defined page segments and turns them into a reusable navigation structure
  • InSupport identifies data-record and auxiliary segments and turns them into a queryable comparison interface
  • Standpoint identifies argumentative structure in a thread and turns it into position-scoped summaries and relation labels
None of them requires the website to change.



Figure 2: One proxy layer between the webpage and the screen reader, applied three times in the dissertation: to page structure (Chapter 3, RQ1), web data records (Chapter 4, RQ2) and conversational meaning (Chapter 5, RQ3).

Publications

Four of these papers make up the dissertation work across the three research questions above. Standpoint, the system built for RQ3, is currently under submission. The rest came out of collaborations in the lab.

The dissertation
Collaborations
The Hypertext paper was a departure from accessibility work: detecting human smuggling advertisements on social media.


Teaching

I was a graduate teaching assistant at ODU from Fall 2019 through Spring 2026, for Principles of Programming Languages (CS 355), Introduction to Theoretical Computer Science (CS 390), Computers in Society (CS 300), Introduction to Computer Science with Python (CS 115), Data Visualization (CS 625, graduate), Introduction to Programming with Python (CS 153) and Introduction to Programming with Java (CS 151). I taught the Python section of ODU's Data Science Boot Camp in 2022 and 2024.

I completed ODU's GTA Instructor Institute in 2021 and the Preparing Future Faculty Certificate through the Career Pathways Program in 2025. I also judged the Virginia State Science and Engineering Fair in 2024.


Service and Community

I reviewed for more than 30 venues during the PhD, including ACM TOCHI, ACM TACCESS, ACM TKDD, the International Journal on Digital Libraries, Universal Access in the Information Society, and the CHI, CSCW, ASSETS, DIS, IUI, EICS, ICMI, IDC, IMX, CUI, ETRA, NordiCHI, OzCHI, ISMAR, IEEE VR, ICWSM and CogSci conferences.

I served as Associate Chair for ACM CSCW 2025 (Research and Posters), ACM CSCW 2024 (Posters) and ACM CHI PLAY 2024 (Work in Progress), and on program committees for ACM CHI 2025 (Late-Breaking Work), ACM ETRA 2025 (Short Papers), ICWSM 2024, ACM ICMI 2023 and ACM EICS 2023.

I was a student volunteer at ACM ASSETS from 2022 to 2024, and at CHI PLAY 2023, DIS 2022 and 2023 and RecSys 2022.

ACM named me a Certified Reviewer in 2025 and invited me to the Distinguished Reviewer Board of ACM Transactions on Computer-Human Interaction in January 2026. In 2022 the department awarded me the Dr. Hussein Abdel-Wahab Memorial Fellowship.

I also blogged throughout the PhD, on my own work and on other people's. One I keep coming back to is the Global Accessibility Awareness Day piece on whether web archives are usable for blind readers, which asks my dissertation's question about web archives, a core research area of WS-DL.


The Defense

I defended on July 23, 2026. My committee was:
I presented the three systems and the argument that holds them together: that they are one pattern applied at three levels (page structure, data records and conversation), not three separate tools. Then I took questions, and then they sent me out of the room to wait while they decided.

My defense slides are embedded below.


The full dissertation is available from ODU Digital Commons.



Figure 3: Explaining sTag (RQ1) at the defense. The slide shows how a user's tags on an Expedia flight search become a Table-of-Contents overlay.


Being called back into the room took only a few seconds, but I had imagined that moment for years. I remember the room more than the words: my advisor, colleagues from the lab, people who had watched this work change shape several times over. We took photographs afterwards. Somebody called me Dr. Ferdous, and it took me a while to realize they were talking to me.


Figure 4: "Congratulation!!! Dr Javedul" on the screen after the defense.


Lessons Learned

1. Do not measure yourself against whoever is closest to finishing

For a long stretch I compared my progress to whoever in the group was nearest to defending, which did not help me. Some years produce three papers and some produce none, and from inside a year with no papers there is no way to tell whether the next result is three months away or much further. The more useful comparison is with your own progress from the year before.

2. Life does not pause for the degree

Six months after I arrived, the pandemic closed everything, and as an international student I went through it on the other side of the world from my family. During the PhD I lost five people close to me. The hardest was my mother, in 2021. Travel was shut down, and I could not go home. Grieving from that far away was the most difficult part of the PhD. If you are an international student, work out early who you can lean on here, because you will need them sooner than you expect.

3. Keep the people at home close

I moved from Dhaka to Norfolk on my own, and my family supported work they were never in the room for. Any finish date you give them will probably move. Call more often than you feel like calling, and do not wait until you have good news.

4. Do the unglamorous community work, and do it early

I started reviewing because somebody asked me to, and I spent three years volunteering at conferences where the job was often carrying things across a hall. It does not look like much on a CV, but it is how I met most of the people I know in this field. A PhD can be isolating, and serving the community is one of the easiest ways to stay connected.

5. Find out which part of the job you want before you have to choose

I was a teaching assistant for almost the entire degree. Somewhere in the middle it stopped being the thing I did for funding and became the thing I wanted to do, and I would not have known that without years of doing it. That is what led me to a teaching position at Ottawa University.

Looking Forward

Teaching at Ottawa University

In August 2026 I joined Ottawa University as a Visiting Assistant Professor in the School of Arts and Sciences. This term I am teaching Introduction to Programming (ITS/CST 16163), Introduction to Computer Science (CST 20363), Introduction to Database (ITS 20163) and Microsoft Excel (ITS 25008).

After years of running lab sections, review sessions and office hours as a teaching assistant, having my own courses is the part of this move I most wanted. The Preparing Future Faculty program is where I settled on this direction.

Where the research goes

I want to keep working on the same thing: moving accessibility from access to usability. AI now mediates a lot of what people read online, and in my Standpoint pilot, blind readers were cautious about AI summaries and checked the parts they could verify. My next research direction is trust and user control in AI-mediated reading for blind users.

Staying in the community

I intend to keep reviewing and serving on program committees. CHI, CSCW and ASSETS are the main venues for this research, and I want to keep contributing to them.

Acknowledgment

I am deeply grateful to my advisor, Dr. Vikas Ashok, for his guidance, his patience through the long stretches when I had nothing to show him, and his belief in my abilities. His mentorship shaped this dissertation and the researcher I have become.

I am equally grateful to Dr. Michele C. Weigle, Dr. Michael L. Nelson and Dr. Hae-Na Lee, for their thoughtful feedback, encouragement and support at every stage of this work, and for the care they gave to improving it.

I gratefully acknowledge the Web Science and Digital Libraries group and the Human-Centered Computing Lab, whose discussions improved every idea in this work, and my friends and colleagues, who stood by me and kept me grounded through the hardest stretches.

I thank every study participant who shared their time and experience with me. This work does not exist without them.

Above all, I owe the greatest debt to my family. Their love, sacrifices, prayers and constant encouragement kept me going when the work felt overwhelming. This degree is theirs too.



-- Md Javedul Ferdous (@jaf_ferdous)

Comments

Post a Comment