I Can Do It: Exploring Voice Assistants for Adults with Intellectual Disabilities

Paper Title

I Can Do It: Exploring Voice Assistants for Adults with Intellectual Disabilities

Publication Info

• Topic area: Accessibility and human-computer interaction for individuals with intellectual disabilities.
• Keywords: Voice assistants, intellectual disabilities, accessibility, human-computer interaction, speech recognition, adaptive technology, multimodal interaction, STEAM education, social inclusion, design considerations.

Background and Problem

• Problem / challenge: Existing voice assistants (VAs) are not fully accessible to individuals with intellectual disabilities (ID) due to challenges such as mispronunciations, cognitive difficulties, and limited adaptability to diverse user needs. Research on long-term, real-world VA use by this population is limited.
• Significance: VAs have the potential to reduce barriers to information, communication, and independence for individuals with ID, but their usability and accessibility must be improved to realize these benefits.
• Motivation and related work: Prior studies have shown VAs can support specific-need groups, including older adults and individuals with physical disabilities, but research on their use by individuals with ID is sparse. Existing work often focuses on short-term, structured tasks, leaving gaps in understanding long-term, naturalistic use and the challenges faced by this population.

Solution

• Proposed approach: An eight-week study integrating screen-based VAs (Google Nest Hub and Amazon Echo Show) into a STEAM program for adults with ID, focusing on naturalistic and group-based use.
• Novelty:
  1. Long-term deployment of VAs in real-world group settings for individuals with ID.
  2. Analysis of multimodal interaction data, including voice logs and observational insights.
  3. Identification of key challenges and strategies for VA use by individuals with ID.
  4. Development of six design considerations for more accessible and inclusive VAs.
• Procedure and key techniques:
  • Conducted three phases of data collection: preliminary interviews, deployment interviews, and post-reflection interviews.
  • Trained participants on VA use and allowed open-ended exploration in group settings.
  • Analyzed 260 VA interactions, observational data, and participant/coach feedback.
  • Identified recurring themes and challenges through thematic analysis.

Results

• Concrete findings:
  • Participants initiated 260 interactions, primarily for information retrieval, entertainment, and learning.
  • Correct interactions outnumbered errors (260 correct vs. 35 incorrect), with a 15% error rate for Google Assistant.
  • Pronunciation issues, ambiguous phrasing, and timeout errors were common challenges.
  • Participants showed increased confidence and expanded use over time, with some forming emotional connections to the devices.
• Advantage over baselines:
  • Extended deployment revealed evolving engagement patterns and peer-supported learning, which are underexplored in prior short-term studies.
  • Participants shifted from simple to more complex queries, demonstrating learning and adaptation.
• Experiments / evaluation:
  • Conducted in a disability support organization with 17 participants and 4 coaches.
  • Data sources included VA interaction logs, interviews, and field observations.
  • Activities spanned creative, educational, and social contexts within the STEAM program.
• Limitations and future work:
  • Study setting (structured group environment) may not fully reflect everyday use.
  • Devices were introduced at different times, potentially influencing preferences.
  • Future work should involve participatory design with individuals with ID and explore longer deployments in varied real-world contexts.

Summary

This study explored how adults with intellectual disabilities engaged with voice assistants over an eight-week period in a STEAM program. Participants used VAs for diverse activities, showing increased confidence and expanded use over time, but faced challenges such as mispronunciations, timeout errors, and difficulty processing complex responses. The study identified six design considerations, including auto-correction for atypical speech, extended listening windows, simplified responses, routine support, and multimodal interaction. These findings provide actionable guidance for designing more accessible VAs that support communication, learning, and independence for individuals with ID.