The ASL research project at DePaul University is a team that has been working on creating the digital interpreter. Currently, they have a system that takes an english sentence and automatically translates it into a digital avatar that talks in ASL. They have received the funding necessary in order to make software improvements on the tools they are using.

  This project primarily focuses on the open-source program ELAN. Eudico Linguistic ANnotator is the main program translators use to study the ASL signs. With pre-recorded videos of deaf people signing, the researchers have annotated specific features of ASL. Examples are the gloss, nonmanual signals ( body posture, facial expressions, and so on ), and the timing between signs. With that data in place, the animators are able to figure out how to build the internal representations of signs in CAD tools. Therefore the goal of this project is to make their job easier by improving the tools they use.

Phase 1: Understanding the project and planning the scope of the work.

  Upon starting this project, meetings will be necessary to understand what the team is doing and how this summer project will be utilized by them. Furthermore, familiarizing the entire ELAN codebase is required in order to make the changes requested by the team. Finally, setting up the development environment will be done on the computers in the lab.

Phase 2: What improvements to ELAN do the animators need to maximize efficiency in translating signs to digital representations?

  After studying the code of ELAN and meeting with the animators, several areas for improvement have been identified. ELAN is geared mainly for the researchers working on annotating the ASL corpus. As a result, the interface is not that easy to understand and navigate for those new to the project. This phase will absorb most of the project's timeline as the development iterates over changes animators brought up in the first phase. It seems that the biggest task would be improving the search features in ELAN so the animators are able to quickly find the sign they are looking for. In addition to that, the animators need better data than what is currently presented in ELAN to make the translation process less tedious to fine-tune.

Phase 3: Analyzing the ASL corpus to discover patterns that the animators can exploit for their work.

  Another area of improvement to ELAN is to focus on the ASL corpus as a whole. The team has expressed a desire to be able to generate statistics and useful metrics for the entire corpus. The data will be used to identify weak areas in the animations and to further guide the team's work. Currently, ELAN does not have built-in tools to perform this operation but there exists separate tools to do some of the analysis. This phase will consist of integrating those into ELAN and further refining and improving the statistics based on feedback from the team.

Phase 4: Wrapping up the project and finalizing the codebase.

  While working on the project tasks, potential problems/bugs/improvements in the code are logged for follow-up during this phase. As time permits, those will be worked on and documented so others can continue improving ELAN. We hope to publish on our results when the project is done.

Participant: Larwan Berke
  School: Gallaudet University in Washington, DC
  Major: B.S. in Mathematics and B.S. in Postsecondary Education
  Graduation Date: May 2015


Research supervisor: Dr. Rosalee Wolfe
  School: DePaul University in Chicago, IL
  Department: College of Computing and Digital Media

Project sponsored by:
  Distributed Research Experiences for Undergraduates ( DREU )
  The Alliance for Access to Computing Careers ( AccessComputing )