Electronic rings use AI to translate sign language into text

Electronic rings with motion sensors and artificial intelligence can translate sign languages into text. Scientists at Yonsei University are the first to demonstrate it.

Why earlier attempts fell short

For many years, researchers have tried to build a practical sign language translation system. Some used cameras and computer vision, but that worked only under controlled conditions and often made errors when lighting changed. Others developed smart gloves with sensors. However, the gloves trapped heat and moisture, causing discomfort, and did not account for individual differences in hand size and finger length. Most sensors required a wired connection to a computer, restricting hand movement. Even wireless versions usually connected by wire to a single central transmitter.

Seven rings instead of gloves

A team led by Ki Jun Yu developed a new approach: a set of electronic rings, each of which transmits motion data over Bluetooth Low Energy. Using rings instead of gloves made it possible to place sensors flexibly and to account for natural differences in anatomy between people. The wireless link gave the hands complete freedom.

Analysis showed that seven fingers play the main role in sign language, so the team used exactly seven rings. Each is equipped with an accelerometer that tracks both static poses and movements. This is critical because sign languages combine static positions with dynamic transitions between them.

During development, the system ran into a reliability problem. Straight copper contacts broke under frequent bending. The solution was to switch to contacts with a serpentine pattern, which withstand thousands of cycles.

AI recognizes the gestures

The researchers developed a deep learning system that recognizes gestures with remarkable generality. The system learned to work not only with the two people in the training set but also with five completely new people, without retuning for each one.

In experiments, the system recognized 100 words of American Sign Language and 100 words of International Sign with an accuracy of 88.3% and 88.5%, respectively. That is a huge leap forward: earlier attempts were limited to fewer than 50 words in the vocabulary.

Impressively, the system translated not only individual words but also whole sentences during continuous signing. That is exactly what real-time natural conversation requires.

What this means

The long-term goal is to integrate the system into smartphones without special hardware. The rings would send signals to a mobile device, where they would be translated automatically in real time. This would make the technology accessible for everyday communication.

Next steps: extend the system to more people, add more words to the vocabulary, and support Korean Sign Language and other regional variants. The scientists also plan to extend battery life from the current 12 hours to a full day.

One important caveat: the current system translates only hand movements. The next big challenge is adding facial expressions, lip movements and spatial constructions, which are grammatically important in sign languages.