Overview / Usage

Sign language examinations usually consist of two components: a written component and a practical component. The written component tests the exam candidate’s theoretical knowledge on sign language and issues related to it, such as ethics surrounding deaf people. In the practical component the candidate is expected to exhibit his knowledge on sign language by physically doing different hand signs and gestures. In most cases, these examinations are carried out in a traditional way, where the evaluation is done manually by a person. This obviously comes with a lot of different challenges and obstacles. It requires a lot of effort and time from both the candidate and evaluator. It brings about many logistical complications as well. All of this results in the entire process being unnecessarily costly and time consuming. A complete automatization of these examinations seems like the logical thing to do, but it has proven to be a challenging task. If the examination consisted purely of multiple-choice questions, it would be relatively simple to create an application which can automatically evaluate it. This would be done by simply comparing the submitted answers to a pre-stored answer. But if the examination consists of descriptive questions, such as ‘how can we solve this issue’, the task of automatically evaluating it becomes a little more complicated. In the case of descriptive questions there isn’t a single fixed answer which we could store in advance and compare the submitted answer to. So, we need to find a workaround to this problem. The other major challenge that we face when attempting to automatize these examinations is the evaluation of the practical component of the exam. Since it’s physical hand signs and gestures that is evaluated here, the computer is supposed to take them in as visual input and decide whether they are correct. Historically, computers have not been particularly good at these kinds of tasks. However, the recent revolution in the field of Artificial Intelligence, especially machine learning, has made it possible for us to create a solution for the problems mentioned above, which is what we will be attempting to do in this project. The system will be in the form of an application, which will assess both the written and the practical test. It will make use of machine learning technologies and techniques such as Natural Language Processing, Artificial Neural Network, etc.

Repository

https://github.com/ANair97/ASLD-system

Collaborators