Story perspectives

USC Develops AI to Translate American Sign Language

9/9/2025

32 2

Summary

Story summary

USC researchers are creating a machine learning model to translate American Sign Language (ASL) into text.
Led by Lee Kezar, the project recognizes ASL's unique syntax as a complex linguistic system.
Limited data for sign languages poses a significant challenge.
The model currently achieves 91% accuracy for isolated signs and 14% for unseen signs' semantics.
Collaboration with the Deaf community is essential for respectful development and future expansion to other sign languages.

1 of 1

Story summary

USC researchers are creating a machine learning model to translate American Sign Language (ASL) into text.
Led by Lee Kezar, the project recognizes ASL's unique syntax as a complex linguistic system.
Limited data for sign languages poses a significant challenge.
The model currently achieves 91% accuracy for isolated signs and 14% for unseen signs' semantics.
Collaboration with the Deaf community is essential for respectful development and future expansion to other sign languages.