LeConTra: A Learner Corpus of English-to-Dutch News Translation

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 1807–1816 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 1807 LeConTra: A Learner Corpus of English-to-Dutch News Translation Bram Vanroy, Lieve Macken LT3, Language and Translation Technology Team, Ghent University Groot-Brittanni¨elaan 45, 9000 Ghent, Belgium bram.vanroy@ugent.be, lieve.macken@ugent.be Abstract We present LeConTra, a learner corpus consisting of English-to-Dutch news translations enriched with translation process data. Three students of a Master’s programme in Translation were asked to translate 50 different English journalistic texts of approximately 250 tokens each. Because we also collected translation process data in the form of keystroke logging, our dataset can be used as part of different research strands such as translation process research, learner corpus research, and corpus-based translation studies. Reference translations, without process data, are also included. The data has been manually segmented and tokenized, and manually aligned at both segment and word level, leading to a high-quality corpus with token-level process data. The data is freely accessible via the Translation Process Research DataBase and GitHub, which emphasises our commitment of distributing our dataset. The tool that was built for manual sentence segmentation and tokenization, Mantis, is also available as an open-source aid for data processing. Keywords: translation corpus, learner corpus, empirical translation studies, keystroke logging, translation process 1. Introduction All empirical research starts with a need for data, and depending on the research topic, this data can be hard to find. For our own research, we were specifically concerned with using behavioral process data (as mea- sured by for instance keystroke logging and eye track- ing) for translation difficulty to answer questions such as: how does the translation behavior provide insights about difficulties that a translator has encountered? Al- though such datasets exist, they are scarce for English- to-Dutch translation, the exception being the limited datasets of Daems (2016) and Vanroy (2021). Further- more, to get plenty of behavioral data per subject, we were interested in a dataset with many translations per translator to cover a wide range of linguistic phenom- ena, rather than plenty translations per source text, so we opted to have a few translators work on many texts rather than many translators on few texts. As a result, we collected a dataset for English-to- Dutch translation with behavioral data in the form of keystroke logging information, and we make it freely available. We are especially contributing to two fields: learner corpus research (LCR) and translation process research (TPR). On the one hand the dataset that we publish contains translations of students of Translation as well as reference translations (Sec. 3.2.1), and on the other we include translation process data in the form of keystroke logging information. In what follows we first provide some non-exhaustive background about translation process research and translation difficulty, the field and topic in whose light this dataset was originally created, as well as exist- ing endeavours in learner translation corpora (LTC). Then we discuss the dataset itself, with a focus on …

Content truncated to 500 words