ASL Educators' Perspectives on AI for Enhancing Student Learning in American Sign Language Education
Saad Hassan, Laleh Nourian, Caluã de Lacerda Pataca, Michelle M Olson, Toni D'aurio, Kanupriya Agarwal, Syeda Mah Noor Asad, Garreth W. Tigwell, Matt Huenerfauth · 2026 · CHI '26: Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems · doi:10.1145/3772318.3791928
Summary
This CHI 2026 paper from a multi-institution US team (Tulane, Rochester Institute of Technology, Birmingham City University) investigates how AI could support American Sign Language (ASL) education — and centres, for the first time in this literature, the perspectives of the Deaf educators who teach it. ASL is the third most-studied language in US higher education with 107,000+ enrolments annually, but pedagogy remains fragmented, most research to date has focused on ASL learners, and the AI systems under development have been designed largely without input from the Deaf educators who would deploy, moderate, and grade with them. The authors conducted formative 40-60 minute Zoom interviews with 11 Deaf and 1 hearing ASL instructor at US universities, followed by two focus groups with six Deaf educators after a structured AI primer, using reflexive thematic analysis (Braun and Clarke). Participants averaged 14 years of teaching experience, ages 31-67, at a mix of R1, R2, and Primarily Undergraduate Institutions. The paper surfaces five themes — AI integration must align with institutional curricula and pedagogical approaches; tools must adapt to ASL\'s linguistic diversity and nuances (regional variation, Black ASL, non-manual grammar); technology barriers and differential institutional access affect educational equity; AI feedback tools must be instructor-moderated and tailored to learner level; and virtual conversational partners require warmth, cultural adeptness, customisation, and support for natural turn-taking — producing eleven concrete design insights tagged as near-term or aspirational.
Key findings
Educators expressed cautious optimism: AI could genuinely reduce their workload (22 minutes per student per test, 19 minutes per video assignment is typical) and offer students feedback outside limited class hours, but only if instructors remain the authority over grading, rubrics, and curricular alignment. Participants were explicit that AI should not auto-generate rubrics or replace human judgment — it is a "textbook tool" or "answer key" under instructor oversight. Linguistic diversity emerged as a central concern: current AI is English-centric and risks encoding ASL as glossed English, ignoring regional variation, Black ASL, and non-manual grammar. Educators called for Deaf-led data collection, inclusion of ASLTA and linguistic stakeholders in development, and avoidance of dialect homogenisation. On feedback tools, participants asked for instructor-moderated output, learner-customised feedback (L1 vs L2 learners, proficiency level, regional accent), time-stamped textual and visual feedback, and correct-performance demonstrations rather than error-flagging alone — with the corrective-signing demo flagged as aspirational given current sign-generation limits. For virtual conversational partners, educators wanted warm and personable characters, customisable signing speed and appearance, diverse signers reflecting Deaf community variation, 20-30 minute conversations matching college attention spans, and good turn-taking with appropriate interruption patterns. Equity concerns were strong: AI could widen the gap between well-resourced universities and under-resourced schools, and paywalled AI tools could replicate existing access barriers in higher education.
Relevance
This paper is immediately useful for anyone building AI-powered sign-language tools, video-feedback platforms, or signing avatars — and more broadly, it is a model for how to centre disabled experts in the design of technologies that will affect their communities. The eleven design insights split cleanly into near-term recommendations (instructor-moderated feedback, time-stamped textual/visual feedback, warm conversational personas, customisable avatar attributes, attention-span-matched durations, natural turn-taking) and aspirational goals requiring further model capability (learner-context-aware feedback, correct-signing demonstrations from learner errors, Deaf-cultural-diversity-reflecting avatars, story-driven conversational practice, specialised interpreting scenarios). Practitioners should note the cultural-taxation point — relying on Deaf educators as unpaid cultural consultants is not a sustainable design practice, and the paper makes a concrete case for paid Deaf-led data curation and ASLTA involvement. Limitations include the predominantly White/Caucasian sample (8 of 12), the absence of BASL-specialist or Latinx-Deaf educator perspectives, the online-only focus-group format, and that participants reflected on AI tools via demonstrations rather than hands-on use. The paper complements the same lab's earlier work on learner-facing sign-language technologies and extends it to a cross-stakeholder agenda for equitable AI in sign-language pedagogy.
Tags: American Sign Language · ASL · sign language education · deaf educators · AI in education · signing avatars · conversational agents · feedback tools · deaf culture · generative AI · educator perspectives · higher education
Standards referenced: World-Readiness Standards for Learning Languages