# Designing for Harm Reduction: Communication Repair for Multicultural Users' Voice Interactions

arXiv: https://arxiv.org/abs/2403.00265 · HTML: https://mlsanity.com/paper/2403.00265/

## Key facts

- Cited by 6 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 1 Mar 2024.

## Built on by (most cited first, 6 of 6)

- [NormAd: A Framework for Measuring the Cultural Adaptability of Large Language Models](https://mlsanity.com/paper/2404.12464/) (2024)
- [Voices Unheard: NLP Resources and Models for Yor`ub'a Regional Dialects](https://mlsanity.com/paper/2406.19564/) (2024)
- [Mind the Gesture: Evaluating AI Sensitivity to Culturally Offensive Non-Verbal Gestures](https://mlsanity.com/paper/2502.17710/) (2025)
- [Sound Check: Auditing Audio Datasets](https://mlsanity.com/paper/2410.13114/) (2024)
- [A Case Against Implicit Standards: Homophone Normalization in Machine Translation for Languages that use the Ge'ez Script](https://mlsanity.com/paper/2507.15142/) (2025)
- [A Cross Community Agenda for Speech AI](https://mlsanity.com/paper/2609.13168/) (2026)

## Abstract (arXiv)

Voice assistants' inability to serve people-of-color and non-native English speakers has largely been documented as a quality-of-service harm. However, little work has investigated what downstream harms propagate from this poor service. How does poor usability materially manifest and affect users' lives? And what interaction designs might help users recover from these effects? We identify 6 downstream harms that propagate from quality-of-service harms in voice assistants. Through interviews and design activities with 16 multicultural participants, we unveil these 6 harms, outline how multicultural users uniquely personify their voice assistant, and suggest how these harms and personifications may affect their interactions. Lastly, we employ techniques from psychology on communication repair to contribute suggestions for harm-reducing repair that may be implemented in voice technologies. Our communication repair strategies include: identity affirmations (intermittent frequency), cultural sensitivity, and blame redirection. This work shows potential for a harm-repair framework to positively influence voice interactions.