Impact of Out-of-Vocabulary Words on the Twitter Experience of Blind Users

Title of the Paper

Impact of Out-of-Vocabulary Words on the Twitter Experience of Blind Users

Bibliographic Information

• Subject Area: Social Media, Accessibility Technology, Natural Language Processing
• Keywords: Social Media, Twitter, Out-of-Vocabulary Words (OOV), Blind Users, Visual Impairment, Screen Readers, User Experience

Research Background and Problem

• Issues and Challenges: Non-standard vocabulary on social media (e.g., abbreviations, misspellings, slang, etc.) poses comprehension challenges for blind users, as these words may not be correctly read aloud or semantically translated by screen readers.
• Significance: Social media is a crucial tool for information dissemination and social interaction in modern society. For visually impaired individuals, screen readers are their primary medium. The prevalence of non-standard language may hinder their effective participation in social interactions and information sharing.
• Research Motivation: While extensive studies have focused on the accessibility of visual objects (e.g., images, emojis), there is limited research on how non-standard vocabulary in social media text affects the user experience of blind individuals.

Solution

• Methods and Innovations:
  1. Conducting a study to investigate how current screen readers read various types of non-standard vocabulary on Twitter.
  2. Creating a custom dataset containing 6,717 non-standard words and categorizing them.
  3. Conducting user research to explore how these non-standard words impact the comprehension and interaction behaviors of 15 blind users with tweets.
• Implementation Steps and Techniques:
  1. Dataset Construction: Extracting original tweets and their corresponding standardized versions using a text normalization dataset.
  2. Vocabulary Classification: Dividing non-standard words into 16 subcategories (e.g., abbreviations, misspellings, slang, word merging, etc.).
  3. Screen Reader Analysis: Comparing the performance of the three most popular screen readers—JAWS, NVDA, and VoiceOver—in reading non-standard vocabulary.
  4. User Research: Designing experiments including word recognition tasks and tweet comprehension tasks, recording users' accuracy and their interaction behaviors with screen readers.

Research Findings

• Experimental Results:
  • Current screen readers rarely auto-correct non-standard vocabulary, requiring blind users to rely on their own inference to understand word meanings.
  • Experimental data shows that tweets containing non-standard vocabulary significantly reduce comprehension. Tweets with OOV words had an average comprehension accuracy of 51.33%, far lower than standard tweets (average 90.67%).
  • Screen readers exhibited inconsistent performance in reading certain types of non-standard vocabulary (e.g., abbreviations), potentially increasing users' cognitive load.
  • Users tended to understand non-standard words through repeated listening, manual spelling, and leveraging semantic clues.
• Comparative Analysis:
  • NVDA performed better in some aspects of word spelling accuracy, but overall differences among screen readers were not significant.
  • Users preferred the "spelling" mode of screen readers over full pronunciation.
• Limitations and Constraints:
  • The study was limited to an English academic context and did not cover other languages and cultural backgrounds.
  • User feedback regarding actual social media behavior has not been fully validated.

Limitations and Future Directions

• The research needs to be expanded to multilingual and multicultural contexts to explore the impact of non-standard vocabulary on blind users' experiences in other languages.
• Development of more intelligent solutions is necessary, such as:
  • Text Normalization: Efficient and accurate auto-correction of non-standard vocabulary while preserving users' preferred non-standard terms.
  • Spelling Simulation: Automatically adjusting key types of vocabulary (e.g., slang) to spelling mode.
  • User Control: Developing interactive assistants that allow users to customize the pronunciation or spelling behavior of vocabulary directly within the screen reader interface.
• Further analysis of user behavior regarding non-standard vocabulary usage on Twitter is an important direction for validating user preferences.

Conclusion

This study systematically investigates the impact of non-standard vocabulary on the social media experience of blind users, revealing deficiencies in screen readers' ability to read these words and proposing potential directions for improving blind users' interaction with tweets. This research is not only significant for enhancing existing screen readers but also provides important references for future accessibility studies in social media.