QE Viewer: an Open-Source Tool for Visualization of Machine Translation Quality Estimation Results

QE Viewer: an Open-Source Tool for Visualization of Machine Translation Quality Estimation Results Felipe Soares University of Sheffield Broomhall Sheffield S10 2TG United Kingdom fs@felipesoares.net Anna Zaretskaya TransPerfect Passeig de Gr`acia, 11 Esc B 5-2 08007 Barcelona, Spain azaretskaya@translations.com Diego Bartolom´e TransPerfect Passeig de Gr`acia, 11 Esc B 5-2 08007 Barcelona, Spain dbartolome@translations.com Abstract QE Viewer is a web-based tool for visual- izing results of a machine translation qual- ity estimation (QE) system. It allows users to see information on the predicted post- editing distance (PED) for a given file or sentence, and highlighted words that were predicted to contain MT errors. The tool can be used in a variety of academic, edu- cational and commercial scenarios. 1 Introduction This paper presents a web-based tool for visual- ization of machine translation quality estimation (QE) (Specia et al, 2018) results. The tool allows users to submit one or several bilingual files with machine translation (MT) output and see informa- tion about its estimated quality, namely the pre- dicted post-editing distance (PED), the distribution of segments with different PED scores in the file, and general information about the file in terms of number of segments and word count. Most impor- tantly, the tools also allows to see all segments in the file and highlights words that potentially con- tain MT errors. The source code is available at: https:// github.com/soares-f/qe-viewer . 2 Functionality of the Tool 2.1 Features QE Viewer is an easy-to-use web interface that al- lows users to upload a bilingual file in TMXL1 for- mat (or multiple TMXL files in a .zip archive) that c⃝ 2020 The authors. This article is licensed under a Creative Commons 3.0 licence, no derivative works, attribution, CC- BY-ND. 1TXML format is an XML-based format of bilingual files used at TransPerfect. contains source segments and MT segments. The user also has the option to introduce a unique iden- tifier for that submission, allowing possible inte- gration with other MT workflows. Once the file is submitted it is processed with the QE system and the user sees the results page that contains a sensitivity bar allowing the user to adjust the system’s prediction; file-level informa- tion on total number of segments, total number of words, predicted average PED; a histogram rep- resenting the distribution of segments in the file with different level of predicted PED (Figure 1); segment-level information showing the predicted PED for each segment in the file, while also high- lighting the words that, according to the QE sys- tem’s prediction, contain MT errors (Figure 2). 2.2 Integration with Different QE Frameworks The tool is system-agnostic which means that it can be integrated with any QE framework, given that a pre-defined API request can be met. For this purpose, we implemented a wrapper for the OpenKiwi framework (Kepler et al., 2019) that handles the requests. 3 Technical Implementation QE Viewer is implemented in Flask,2 which is a Python micro-framework for web development that allows quick development and at the same time it can scale …

Content truncated to 500 words