Screen Reader Intent Summaries for Faster Navigation

ISEF Category: Systems Software

Subcategory: Human/Machine Interface  ·  Difficulty: Advanced  ·  Setup: University Lab  ·  Time: Full Year

The Hook

A long web page can feel like a maze when you cannot see it. Screen readers help, but they still make you explore line by line. Your project can ask a smarter question, what if the page told you its main jobs up front? That shift could change how fast blind users find what they need.

What Is It?

This project studies a keyboard-only navigation aid for screen readers. The tool scans a long page and turns it into a structured table of intents, which means a short map of what each section is trying to do. Think of it like a table of contents, but with purpose labels instead of just headings.

The key idea is simple. A screen reader reads the web in order, which can be slow when a page is packed with menus, ads, and long paragraphs. Your tool uses a small local language model, which is software that predicts and rewrites text, to summarize the page into a clearer path. Then you compare that path with the default NVDA experience, which is a popular free screen reader used by many blind users.

Why This Is a Good Topic

This is a strong science fair topic because you can test a real usability problem with measurable outcomes. You can compare task time, success rate, error rate, and user ratings between the default screen reader flow and your new interface. The topic connects to accessibility, human-computer interaction, and software design, so the real-world value is easy to explain. You can also learn how to design studies, handle user feedback, and analyze performance data.

Research Questions

• How does a structured table of intents affect task completion time for blind users on long web pages?
• What is the effect of intent summaries on task success rate compared with default NVDA navigation?
• Does the summary interface reduce the number of keyboard actions needed to find a target section?
• To what extent do users rate the summarized interface as easier to use than the default screen reader flow?
• Which page features, such as headings, links, or dense text, most improve or hurt summary accuracy?
• How does the quality of the local small LM summary affect user performance on navigation tasks?
• What is the effect of different intent label styles on user confidence during screen reader navigation?

Basic Materials

• Laptop or desktop computer with keyboard-only navigation support.
• NVDA screen reader installed on Windows.
• Test web pages with long-form content and clear target tasks.
• A local small language model runtime on the computer, such as Ollama or LM Studio.
• Screen recording software for timing and error review.
• Spreadsheet software for logging task times, errors, and survey ratings.
• Consent form and study script for remote participants.

Advanced Materials

• Access to a university computing lab or a powerful local machine for model testing.
• A set of annotated web pages for training or evaluation.
• Python for data collection, interface testing, and analysis.
• ImageJ or other software for viewing recorded interaction traces if needed.
• Statistical software for user study analysis.
• Accessible survey platform for remote participants.
• Version control system for tracking interface changes.

Software & Tools

• NVDA: Provides the baseline screen reader experience you will compare against.
• Ollama: Runs a small local language model on your own computer for offline summarization.
• LM Studio: Lets you test and swap local models while keeping the interface local.
• Python: Helps you automate logging, organize results, and analyze task data.
• R: Supports statistical tests and plots for user study results.

Experiment Steps

1. Define the exact navigation problem you want to solve, such as finding a section, locating a form, or comparing page topics.
2. Choose one page type and one user task set so your results stay focused and measurable.
3. Design the table of intents format, including the labels, order, and keyboard controls.
4. Plan a fair baseline comparison against NVDA defaults, with the same tasks and the same page set.
5. Build a study script that collects time, success, errors, and user ratings without changing the task wording.
6. Decide how you will score summary quality and connect that score to user performance.

Common Pitfalls

• Testing on pages that are too short, which hides the benefit of the intent summary.
• Changing the page set between trials, which makes the NVDA baseline and your tool hard to compare.
• Writing summaries that sound helpful to sighted users but do not match how blind users search with a screen reader.
• Measuring only speed, which misses errors, confusion, and failed task attempts.
• Ignoring remote-study accessibility issues, which can block users from completing the study smoothly.

What Makes This Competitive

A strong version of this project goes beyond a simple faster-or-slower comparison. You would need a careful study design, clear accessibility goals, and a strong analysis of where the summary helps or hurts. The best projects also test more than one page type or intent format, then explain why one works better. If you can connect summary quality to real user performance, your project starts to look like serious human-computer interaction research.

Project Variations

• Test the tool on news articles instead of general long pages to see whether dense editorial text changes the benefit.
• Compare intent summaries generated by two different small local models to see which one preserves navigation cues better.
• Measure whether the interface helps more on pages with many links than on pages with few links.

Learn More

• NV Access NVDA user guide: Learn how the baseline screen reader works, and find it on the official NV Access site.
• WebAIM screen reader usability resources: Read practical accessibility guidance and studies, and find them on WebAIM.
• W3C WAI tutorials and guidelines: Review accessibility standards for structure, headings, and keyboard use, and find them on the W3C Web Accessibility Initiative site.
• ACM Digital Library: Search for peer-reviewed papers on screen reader navigation, summarization, and human-computer interaction.
• PubMed: Search for review articles on assistive technology usability and remote user studies.
• MIT OpenCourseWare Human-Computer Interaction materials: Find lecture notes and course readings on interface design and evaluation.