A majority of deaf 18-year-olds in the United States have an English reading level below that of a typical 10-year-old student, and so machine translation (MT) software that could translate English text into American Sign Language (ASL) animations could significantly improve these individuals’ access to information, communication, and services. Previous English-to-ASL MT projects have made limited progress by restricting their output to subsets of ASL phenomena – thus avoiding important linguistic and animation issues. None of these systems have shown how to generate classifier predicates (CPs), a phenomenon in which signers use special hand movements to indicate the location and movement of invisible objects (representing entities under discussion) in space around their bodies. CPs are frequent in ASL and are necessary for conveying many concepts.

This project has created an English-to-ASL MT design capable of producing classifier predicates. The classifier predicate generator inside this design has a planning-based architecture that uses a 3D “visualization” model of the arrangement of objects in a scene discussed by the English input text. This generator would be one pathway in a multi-path English-to-ASL MT design; a separate processing pathway would be used to generate classifier predicates, to generate other ASL sentences, and to generate animations of Signed English (if the system lacked lexical resources for some input).

Instead of representing the ASL animation as a string (of individual signs to perform), this system encodes the multimodal language signal as multiple channels that are hierarchically structured and coordinated over time. While this design feature and others have been prompted by the unique requirements of generating a sign language, these technologies have applications for the machine translation of written languages, the representation of other multimodal language signals, and the production of meaningful gestures by other animated virtual human characters.

To evaluate the functionality and scalability of the most novel portion of this English-to-ASL MT design, this project has implemented a prototype-version of the planning-based classifier predicate generator. The classifier predicate animations produced by the system have been shown to native ASL signers to evaluate the output.

Full Thesis:
Download Matt Huenerfauth’s Full Thesis

Thesis Advisor:
Mitchell P. Marcus and Martha Stone Palmer

Award Date:
June 1, 2006

University of Pennsylvania
Philadelphia, Pennsylvania, USA

Author Contact: