ASL Wiki: An Exploratory Interface for Crowdsourcing ASL Translations
Abraham Glasser, Fyodor Minakov, Danielle Bragg · 2022 · Proceedings of the 24th International ACM SIGACCESS Conference on Computers and Accessibility (ASSETS '22) · doi:10.1145/3517428.3544827
Summary
This paper presents ASL Wiki, a novel bilingual web interface that enables the Deaf and Hard-of-hearing (DHH) community to crowdsource English-to-American Sign Language (ASL) translations of text articles. The system addresses two interconnected problems: the severe lack of informational resources available in signed languages, and the shortage of large, diverse, continuous sign language datasets needed to advance sign language research and technology. Approximately 1 in 6 U.S. adults is Deaf or Hard-of-Hearing, and over 17% of deaf adults have low literacy in English. Because ASL is a completely distinct language from English — not a one-to-one mapping — text-based resources are often insufficient for DHH signers who prefer ASL. Existing sign language resources are limited to dictionaries, vocabulary apps, and a small number of educational materials; most informational content (encyclopedias, news, articles) remains text-only. The ASL Wiki interface provides a side-by-side view with ASL video on the left and English text on the right, synchronized so users can consume content bilingually. In recording mode, contributors select individual English sentences, record themselves signing the ASL translation via webcam, and submit their videos. The interface tracks completion progress, supports multiple signers per sentence, and includes upvote/downvote feedback. For this exploratory work, the platform was seeded with popular English Wikipedia articles across categories like Entertainment, Deaf Culture, Sports, Science, and History. The researchers conducted two studies: a user study with 19 DHH participants who used the interface to both consume and generate content, and a translation quality analysis comparing recordings from four Certified Deaf Interpreters (CDIs) using ASL Wiki versus their standard professional translation setups.
Key findings
The user study revealed strong enthusiasm for the concept: all 19 participants agreed it was helpful to view content in both English and ASL, rating the interface 4.5 out of 5 for usability. Participants self-reported looking at the English text 65% of the time and ASL video 35% when reading the pre-recorded "Caramel" article, with ASL content rated 4.6 out of 5 for understandability. Participants recorded an average of 11 sentences each (202 total across 25 articles), generally choosing topics personally meaningful to them. However, there was a notable gap between enthusiasm for consuming bilingual content (average interest 4.6/5) and willingness to contribute recordings (average 3.6/5), with concerns about being judged for signing quality, camera shyness, and privacy. The translation quality analysis showed that recordings made through ASL Wiki were comparable to those created through professional state-of-the-art setups across all five evaluation dimensions — translation accuracy, linguistic correctness, signing naturalness, recording quality, and signing space captured. Recording quality was actually significantly better through the ASL Wiki interface (p<.005), likely because the built-in webcam recording reduced dependence on interpreters' own equipment. Expert evaluators noted that ASL Wiki translations tended to be more literal, sticking closer to the English text, while standard-setup translations were more interpretive — potentially because the sentence-by-sentence interface discouraged elaboration beyond the source text.
Relevance
This research has significant implications for information accessibility and language equity. It demonstrates a viable approach to scaling the creation of signed language content through community contribution, potentially transforming how DHH individuals access educational materials, news, and general knowledge. For accessibility practitioners, the paper highlights an often-overlooked dimension of digital accessibility: even when text content is technically accessible to screen readers and other assistive technologies, it may not be linguistically accessible to people whose primary language is a signed language rather than a written one. The bilingual interface design — showing both languages simultaneously with synchronized highlighting — offers a model for any application seeking to bridge signed and written languages. The crowdsourcing approach also addresses the critical shortage of sign language datasets needed for developing recognition and translation technologies. Key challenges identified include quality control at scale, contributor privacy concerns around video data, incentivization for sustained participation, and the tension between literal translation and natural ASL interpretation. The platform could be adapted for any pair of signed and written languages, extending its potential impact globally.
Tags: sign language · ASL · Deaf and Hard-of-Hearing · crowdsourcing · bilingual interface · language accessibility · sign language datasets · education