Reimagining Sign Language Technologies: Analyzing Translation Work of Chinese Deaf Online Content Creators
Xinru Tang, Anne Marie Piper · 2026 · Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems (CHI '26) · doi:10.1145/3772318.3790624
Summary
Tang and Piper investigate the translation practices of thirteen deaf Chinese online content creators who produce sign language videos for Kuaishou, Bilibili, Douyin, WeChat, and Xiaohongshu, reaching audiences that range from thousands to nearly a quarter million followers. The study is motivated by sharp community skepticism toward emerging sign language AI — avatars, automatic translation systems, and tools like Google's SignGemma — which deaf communities have criticized for misrepresenting signed communication and undermining hard-won linguistic rights. Rather than critiquing these technologies from the outside, the authors center deaf-led translation work as empirical grounding for reimagining system design. Using semi-structured interviews conducted December 2022 to May 2023 (1-4 hours each, in written Chinese, spoken Mandarin, or sign language via professional interpreters) and reflexive thematic analysis, the researchers examined how creators move between Chinese (written and spoken), Chinese Sign Language and its regional variants, Signed Chinese, gloss, fingerspelling, mouthing, visual vernacular, classifiers, captions, and AI-generated speech. The research context is distinctive: China lacks a standardized national sign language, has multiple competing signing systems, and most deaf students have limited access to formal CSL instruction. Creators in this landscape therefore function as everyday translators whose work spans linguistic, multimodal, and cultural dimensions. The authors frame their findings through the sociolinguistic concept of (trans)languaging — treating communication as a fluid blending of linguistic and non-linguistic resources rather than discrete switching between fixed languages.
Key findings
Three themes define the creators' translation work. First, creators design with multilingualism and multiculturalism in mind from the start, not as an afterthought: they sign while writing on whiteboards during live streams, mix CSL variants to reach regional audiences, and adapt register across deaf viewers with wildly different education backgrounds (the paper notes that over 90% of the ~10,000 DHH signers surveyed in a prior study found televised Signed Chinese interpretation confusing). Second, creators scaffold meaning-making through layered strategies — straightforward signing paired with 'deeper' captions, role-playing and analogies for mental-health concepts, visual-spatial classifiers for abstract terms like 'turning point' or 'space capsule,' and culturally-rooted translations (e.g., signing 'curly hair' for the Starbucks logo rather than fingerspelling). Third, creators actively negotiate politics: balancing accuracy against audience engagement when translating the Civil Code, choosing between Signed Chinese and CSL depending on whether they are teaching hearing viewers or promoting deaf awareness, and navigating criticism that signed songs or CODA-style content distort deaf culture to appeal to hearing audiences. Appendix A catalogs observed strategies across gloss, written and spoken Chinese, AI-generated speech, CSL variants, Signed Chinese, classifiers, mouthing, images, narratives, context-setting, and repetition.
Relevance
For accessibility practitioners building or evaluating sign language technologies, this paper is a direct rebuttal to the dominant interpreting model that frames translation as one-to-one conversion between a sign language and a written/spoken language. The authors argue that real deaf translation work is a languaging activity — emergent, multimodal, multicultural, and shaped by deliberate political choices — and that systems built on the interpreting model will inevitably flatten it. Concrete design implications: support multilingual captioning and visual captioning as part of signed output; treat sign language variation as a feature rather than noise to be standardized; provide multiple translation suggestions rather than single authoritative outputs; and invest in community-driven language resources rather than large uniform datasets. Limitations: the research team is hearing and non-signing (translation was mediated by interpreters and one bilingual co-author), participants skew educated and urban, and reception by audiences was not examined. Still, the work provides a rare empirical foundation for deaf-led reimagining of sign language AI.
Tags: sign language · deaf accessibility · sign language translation · content creators · translanguaging · Chinese Sign Language · deaf culture · AI accessibility · multimodal communication · video accessibility