Neural Machine Translation

Project Overview

This project focuses on a novel visual-interactive method for analyzing, understanding, and enhancing neural machine translation. The developed system aids users in automatically translating documents using neural machine translation while also identifying and correcting potential translation errors. User corrections can be leveraged to refine the neural machine translation model, resulting in automatic improvements across the entire document. This initiative was conducted in collaboration with the Institute for Natural Language Processing (IMS) at the University of Stuttgart and originated as a master's thesis: A visual analytics approach for explainability of deep neural networks. The student developed the base implementation for this approach.

My Contribution

I coordinated the project during the writing phase, added features and visualizations to the system, enhanced its usability, and authored most of the paper. This included creating all images for the paper and producing demonstration videos. I also prepared and delivered an online presentation at the Graphics Interface conference.

Publications

Results of this project are published in the paper Visual-Interactive Neural Machine Translation. This paper was recognized as one of the top submissions at Graphics Interface 2021. We were subsequently invited to submit a revised and extended version of the article to the journal Computers & Graphics: Visualization-Based Improvement of Neural Machine Translation. Additionally, content is presented in the poster Visual-Explainable AI: The Use Case of Language Models.