Sorenson's AI ASL push

Sorenson Communications said on April 16 that it has built two artificial intelligence proofs of concept for American Sign Language, including one for live back-and-forth conversations. (sorenson.com) One system turns pre-recorded video or text documents into American Sign Language using a human-like avatar. The other recognizes American Sign Language and converts it into English text in real time for exchanges between a signer and a non-signer. (sorenson.com) Sorenson said the tools are aimed at routine situations where a live interpreter is not practical or available, including retail interactions and workplace communications. The company’s public product page says it is developing translation between signed and spoken languages, including American Sign Language and English. (sorenson.com 1) (sorenson.com 2) Sign-language translation software does two different jobs: reading a signer’s movements and facial cues, and producing signed output that another person can understand. Sorenson’s announcement covers both directions, which is broader than a single captioning or recognition tool. (sorenson.com) (sciencedirect.com) That arrives as companies and researchers push harder on automated sign-language workflows, with Sorenson already tying its effort to prior acquisitions in avatar and translation technology. A 2025 industry report on those deals said Sorenson bought OmniBridge and Hand Talk to expand AI-based sign-language translation. (hearingreview.com) (omnibridge.ai) Researchers have kept warning that sign languages carry meaning through hand shape, movement, body position, and facial expression, which makes automation harder than word-for-word text translation. A recent review of machine learning in sign language said variability, limited annotated data, and model generalization remain central problems. (sciencedirect.com) (springer.com) Recent technical papers have focused on that missing layer. A 2026 conference paper on real-time sign recognition said facial expressions are still under-studied even though they are needed for proper interpretation, and a 2024 preprint proposed adding sentiment and facial-expression generation to improve signed output. (springer.com) (arxiv.org) Some researchers are explicitly arguing for human-in-the-loop systems instead of fully automated ones. An IEEE paper on adaptive sign-language translation proposed continuous teacher and interpreter feedback, and the April 2026 SLxAI Summit framed the choice as one between automated translation and hybrid access models depending on the setting. (ieeexplore.ieee.org) (slxai.org) Gallaudet University said last week that sign language and artificial intelligence are now converging fast enough to warrant dedicated meetings among researchers, companies, and Deaf community advocates. Sorenson’s demos put one of the biggest interpreter-service providers directly into that race. (gallaudet.edu) (sorenson.com)