Abstract

While there are many Deaf or Hard of Hearing (DHH) individuals with excellent reading literacy, there are also some DHH individuals who have lower English literacy. American Sign Language (ASL) is not simply a method of representing English sentences. It is possible for an individual to be fluent in ASL, while having limited fluency in English. To overcome this barrier, we aim to make it easier to generate ASL animations for websites, through the use of motion-capture data recorded from human signers to build different predictive models for ASL animations; our goal is to automate this aspect of animation synthesis to create realistic animations. This dissertation consists of several parts: Part I, defines key terminology for timing and speed parameters, and surveys literature on prior linguistic and computational research on ASL. Next, the motion-capture data that our lab recorded from human signers is discussed, and details are provided about how we enhanced this corpus to make it useful for speed and timing research. Finally, we present the process of adding layers of linguistic annotation and processing this data for speed and timing research. Part II presents our research on data-driven predictive models for various speed and timing parameters of ASL animations. The focus is on predicting the (1) existence of pauses after each ASL sign, (2) predicting the time duration of these pauses, and (3) predicting the change of speed for each ASL sign within a sentence. We measure the quality of the proposed models by comparing our models with state-of-the-art rule-based models. Furthermore, using these models, we synthesized ASL animation stimuli and conducted a user-based evaluation with DHH individuals to measure the usability of the resulting animation. Finally, Part III presents research on whether the timing parameters individuals prefer for animation may differ from those in recordings of human signers. Furthermore, it also includes research to investigate the distribution of acceleration curves in recordings of human signers and whether utilizing a similar set of curves in ASL animations leads to measurable improvements in DHH users' perception of animation quality.

Publication Date

12-2021

Document Type

Dissertation

Student Type

Graduate

Degree Name

Computing and Information Sciences (Ph.D.)

Department, Program, or Center

Computer Science (GCCIS)

Advisor

Matt Huenerfauth

Advisor/Committee Member

Cecilia Alm

Advisor/Committee Member

Kristen Shinohara

Campus

RIT – Main Campus

Share

COinS