Visual Analysis of Translation Data

Abstract

This thesis is the result of a collaboration with the College of Arts and Humanities at Swansea University. The goal of this collaboration is to design novel visualization techniques to enable digital humanities scholars to explore and analyze parallel translations. To this end, chapter 2 introduces the first survey of surveys on text visualization which reviews all of the surveys and state-of-the-art reports on text visualization techniques, classifies them, provides recommendations, and discusses reported challenges. Following this, we present three visual interactive designs that support the typical digital humanities scholars workflow. In Chapter 3, we present VNLP, a visual, interactive design that enables users to explicitly observe the NLP pipeline processes and update the parameters at each processing stage. Chapter 4 presents AlignVis, a visual tool that provides a semi-automatic alignment framework to build a correspondence between multiple translations. It presents the results of using text similarity measurements and enables the user to create, verify, and edit alignments using a novel visual interface. Chapter 5 introduce TransVis, a novel visual design that supports comparison of multiple parallel translations. It incorporates customized mechanisms for rapid and interactive filtering and selection of a large number of German translations of Shakespeare’s Othello. All of the visual designs are evaluated using examples, detailed observations, case studies, and/or domain expert feedback from a specialist in modern and contemporary German literature and culture. Chapter 6 reports our collaborative experience and proposes a methodological workflow to guide such interdisciplinary research projects. This chapter also includes a summary of outcomes and lessons learned from our collaboration with the domain expert. Finally, Chapter 7 presents a summary of the thesis and future work directions.