Australian Sign Language Fingerspelling Generation

Inclusive Technologies

This project is supported by the Telematics Trust. The project brings together the Department of Human-Centred Computing and the School of Languages, Literature, Cultures and Linguistics at Monash University.

Background

Australia faces a shortage of Australian Sign Language interpreters leaving Deaf people isolated and unable to access critical services. Improving Australian Sign Language training is key to ensuring more people pass interpreting exams and enter the workforce as confident and diverse communicators. Most signers in the Australian Sign Language-using community use fingerspelling and signs as part of their communication, where, on average, signing includes about 10% fingerspelling. Therefore, fingerspelling is a core component of Australian Sign Language, used to render proper names and other English words for which there is no lexicalised sign or to clarify a message if the addressee does not understand a sign. Fingerspelling also forms many lexical signs where initials or other letters in the English word form part of how the sign is made.

Aims

The project aims to design and develop technologies for the automatic generation of Australian Sign Language fingerspelling. The project will generate knowledge by creating the first dataset of high-quality motion-captured 3D fingerspellings and the transitions between them, enabling further advancements in this research area. The project outcomes will also include the invention of the first Australian Sign Language fingerspelling generator capable of synthesizing all 26 letters (A-Z) of the English alphabet and the associated transitions, representing a substantial advancement towards automated Australian Sign Language translation. The generator will be made freely available on Auslan Signbank: Australia’s premiere Australian Sign Language website and dictionary.

Investigators

Steering Committee

Contributors

Outcomes

Dataset

We are in process of recording the first dataset of high-quality motion-captured 3D Australian Sign Language fingerspelling. The participants (including both left- and right-handed) fingerspell a specifically designed word listFrequency of character pairs in English languageList of common affixes in English language at a natural pace in our motion capture studio. The data is recorded through a Vicon motion capture system and a pair of Manus OptiTrack Gloves.

Web Application

The WebGL application will allow for the conversion of a text prompt into an animated, interactive 3D model fingerspelling the prompt. The application will be deployed on the Auslan Signbank: Australia’s premiere Australian Sign Language website and dictionary. Below is a prototype of the application. Type a word in the search box and press ↵ Return to convert into 3D fingerspelling (use to zoom, pan and rotate).

Smartphone Application

The iOS application will allow for the conversion of a text prompt into an animated, interactive 3D model fingerspelling the prompt. The application will be deployed on the Apple App Store.

Application mockup (image source: Andre Ky Pham)