Legal-aid groups test AI dubbing to expand language access, keep humans in the loop

A panel of legal-aid staff and vendors demonstrated how AI dubbing can speed translations of legal explainer videos but said human review remains essential to avoid harmful errors.

Adam Mustovsky, founder and CEO of Briefly, demonstrated an AI dubbing workflow in which he uploaded a recorded interview and, he said, “within, I don't know, 8 to 10 minutes, it had my Spanish dubbing ready.” He called the speed and scale “a really, important tool for language access,” while noting that a bilingual reviewer should check the automatic output before public use.

Zane, project manager for the eviction right-to-counsel team at the Legal Aid Society of the Middle Tennessee and the Cumberlands, said automated translation can be “90% effective to get that translation done nearly instantaneously,” and that those faster translations are useful in emergencies such as storms when organizations need immediate multilingual messages.

The panel demonstrated the kyjustice.org explainer video used by four legal-aid organizations in Kentucky and described a workflow: subject-matter experts record explanations; transcripts are refined by large language models into plain-language scripts; and automated dubbing produces non-English audio that staff can edit.

Panelists emphasized limits. Zach, senior deputy director of client services at Legal Aid of Middle Tennessee, urged organizations to set review thresholds: if a translation is 80–95% accurate, staff must decide whether the error rate is acceptable for the intended use and set human-in-the-loop processes to catch high-risk mistakes.

Next steps for groups interested in the approach include piloting dubbing on low-risk materials, securing bilingual reviewers for quality control, and tracking which languages and formats reach the most clients. The panel said they will share sample videos and contact information for follow-up.