Enhancing Digital Accessibility in Government: Leveraging AI for Public Feedback Management

Artificial Intelligence (AI) technologies have numerous applications in the fields of Accessibility and Public Services. These two areas frequently overlap due to legal requirements and the need to ensure that all community members can benefit from government programs. In this article, we will explore how AI is applied at the intersection of these domains.

The issues in managing and making public feedback available in an accessible manner

Citizen input is crucial for governments looking to tailor policies and programs for public welfare. Actively seeking public opinion before and after policy launches, gathering ideas for new programs, and addressing general and specific public needs help ensure that citizens' interests are at the forefront of government decisions. Governments worldwide actively seek public feedback while designing and implementing public programs.

For example, many municipalities in Ontario have established programs to solicit public opinion through public meetings, surveys, and other methods. These municipalities provide various ways for public participation, including public speaking opportunities, emails, and online forms. Municipalities publish the feedback they receive in public forums as part of their commitment to transparency and participation. Additionally, the Accessibility for Ontarians with Disabilities Act (AODA) requires that this material be made available in accessible formats.

Many municipalities publish this feedback in PDF format because it is widely accessible and easy to distribute. However, collecting and publishing all the feedback can be time-consuming and labor-intensive. Ensuring that the PDF is accessible involves carefully tagging the file, adding alt text to images for those who are visually impaired, maintaining the correct reading order, and checking for sufficient color contrast, among other requirements. This process is primarily manual and can take up valuable staff time.

Leveraging AI to manage public feedback

An AI-enabled system can simplify the process and reduce manual work, making it easier for the municipality to execute publishing tasks. The system can:

1. Review public feedback, opinions, and questions from various media channels, including emails and online forms.
2. Analyze the material for potential misuse using Natural Language Processing (NLP).
3. Utilize an AI-powered engine to automatically apply tags to PDF content.
4. Employ an AI-enabled engine to generate contextual alt text for images.
5. Use an AI-driven contrast checker to identify potential contrast violations in text and graphics.
6. Present the output in a human interface for inspection.

Let us look at all the steps in detail

• Reading public feedback: This is the initial stage of data entry into the system. Various protocols allow the system to retrieve data from emails (SMTP) and online forms. This data is then stored in the system's memory for further processing.
• Checking for potential abuse: A key aspect of any government document is maintaining respectful output. To achieve this, a Natural Language Processing (NLP) engine will analyze the content for potential abusive language and flag any concerns in the human interface, which will be discussed later.
• Automatic tagging: The content will be formatted as a basic PDF, and an AI-based engine will automatically apply appropriate tags, such as paragraph, heading, figure, etc. This tagging structure will be visible in the human interface for decision-making.
• Generating contextual alt text for images: The public feedback will consist of text and images, which may range from simple to complex. The system will include a sub-system designed to generate alt text for these images using techniques from computer vision and Natural Language Processing (NLP). A pre-trained model will be used for generating alt text, which can be further refined to adhere to the specific language of the organization. Context is crucial when describing images; therefore, the AI model will consider the overall content to determine whether an image is informative or decorative, subsequently suggesting appropriate alt text for informative images. This sub-system will be pre-trained to generate alt text and will consider:
  • Existing data, particularly images and their corresponding alt text, to understand the organization's current style of alt text generation, including word choice and tone.
  • Existing document data to comprehend the context and identify how images were previously categorized as decorative or informative.
  • Based on this information, the AI tool will enhance its training to generate contextually appropriate alt text for images, tailored to the organization’s style.
• Color contrast flagging: PDFs must adhere to color contrast standards for both text and non-text graphics. Unlike on the web, there are currently no automated tools for flagging color contrast issues in PDFs. One application of computer vision and machine learning (ML) technologies is scanning user interfaces to identify potential contrast problems. Using this approach, a subsystem within the main system will scan the PDF with computer vision techniques to flag any potential color contrast issues.
• Human interface: In our rapidly advancing AI age, technology aims to complement human intelligence and efforts rather than replace them. As AI can still make errors, reviewing the output from the steps above before it is finalized in the PDF is essential. The human interface will provide options to:
  • Retain or discard content flagged for abusive language,
  • Evaluate and accept or modify the PDF tagging,
  • Accept, reject, or modify the suggested alt text for images, and
  • Act on flagged color contrast issues.

Based on the human actions, the system will then generate a final PDF document, which will be accessible to everyone.

Conclusion

By automating tasks such as reading public feedback, checking for abuse, tagging content, generating context-sensitive alt text, and flagging color contrast issues, public sector agencies can streamline their workflows and reduce manual labor. It remains essential for humans to conduct final quality checks and make necessary adjustments, ensuring the content is accurate and respectful. This thoughtful application of AI optimizes staff time and improves the accessibility of public documents, making them more inclusive and informative for all community members. The combination of AI and accessibility illustrates how technology can enhance human efforts, leading to more efficient and accessible public services.

Author Details

Vaibhav Saraf, Accessibility SME and Quality Engineering Lead, Infosys Public Services

Vaibhav is a seasoned Accessibility Professional with extensive expertise in WCAG standards (2.0/2.1/2.2) and WAI-ARIA. With a strong programming background, he is adept at identifying and resolving accessibility challenges across various platforms. Vaibhav has a deep understanding of assistive technologies and has successfully led and delivered numerous projects focused on accessibility for both web and mobile applications (Android and iOS).

With over five years of experience complying with AODA, ADA, and Section 508, Vaibhav has effectively managed accessibility initiatives for a range of projects, including design systems, rebranding efforts, native mobile applications, web applications, enterprise solutions, and accessibility research. He is committed to ensuring that all deliverables are of the highest quality and adhere to accessibility standards.

Sakshi Sood, Quality Engineering Lead, Infosys Public Services

Sakshi is a Quality Assurance expert at Infosys Public Services, where she plays a vital role in ensuring the successful completion of technology programs for public sector agencies, aligning them with established objectives. She has a keen interest in accessibility and actively shares her insights about strategies organizations can implement to develop accessible and inclusive systems.