Chris Hokamp - Transfer Talk - 21st August 2015
Title: Text Prediction and Quality Estimation in Translation Interfaces
Supervisor: Prof. Qun Liu
This transfer report presents new approaches to Text Prediction in the context of mixed-initiative Computer-Aided Translation (CAT) interfaces. We present a novel formulation of the ranking problem for text prediction, and develop a deep learning architecture for a new task, called “sub-segment quality estimation”, which can be used both to reject prediction candidates, and to support text prediction for post-editing scenarios. Prototype implementations are described using concepts from Component Oriented Design, which is our abstract type system for the Interactive Elements and Data Services that compose a translation interface. We design a series of experiments to test the effectiveness of our new components, and suggest both automatic and human evaluation methods. We focus upon three interrelated research topics:
- Developing state of the art text prediction algorithms in the context of Computer Aided Translation
- Integrating text prediction into manual translation and post-editing workﬂows by using quality estimation
- Designing a type system for Translation Interface components
The preliminary work conducted thus far, its relationship with existing work, and our proposals for contributing to the state of the art in each of these areas will be described in this report. We also give an overview our relevant publications to date, and present a detailed plan for the remainder of the Ph.D. work.