Communications of the ACM
A Machine Can Learn to Identify Sign Languages
A computer program is learning to identify the sign language of signers.
Credit: Radboud University Nijmegen
A Ph.D. student at the Max Planck Institute for Psycholinguistics has teamed up with the machine-learning group at Radboud University to train a computer program to quickly identify the sign language of signers.
Binyam Gebrekidan Gebre first addressed the problem of automatic language recognition, training a program using a video of signers working in six sign languages. The program is able to distinguish between the languages with an accuracy rate of 84 percent. "This is a big success rate, given the fact that the machine learned to do so from four signers per language only," Gebre says. "This accuracy will improve when we feed the program with more data. We solved this by generating a dictionary of pixel patterns that appear in the videos and then matched that to those patterns in named languages."
He also addressed the recognition of role-taking in conversations by a computer and the meaningful part of a gesture.
The research could be used for transcribing videos of signed stories or for translating signed languages into spoken or written languages in real time. Video search engine projects also stand to benefit from the work.
From Radboud University Nijmegen
View Full Article
Abstracts Copyright © 2015 Information Inc., Bethesda, Maryland, USA
No entries found