MEDLINE as a Parallel Corpus: a Survey to Gain Insight on French-, Spanish- and Portuguese-speaking Authors’ Abstract Writing Practice

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 3676–3682 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 3676 MEDLINE as a Parallel Corpus: a Survey to Gain Insight on French-, Spanish- and Portuguese-Speaking Authors’ Abstract Writing Practice Aurélie Névéol†, Antonio Jimeno Yepes∗, Mariana Neves⋆ † Université Paris Saclay, CNRS, LIMSI, France ∗ IBM Research Australia ⋆ German Federal Institute for Risk Assessment (BfR), Germany aurelie.neveol@limsi.fr, antonio.jimeno@au1.ibm.com mariana.lara-neves@bfr.bund.de Abstract Background: Parallel corpora are used to train and evaluate machine translation systems. To alleviate the cost of producing parallel resources for evaluation campaigns, existing corpora are leveraged. However, little information may be available about the methods used for producing the corpus, including translation direction. Objective: To gain insight on MEDLINE parallel corpus used in the biomedical task at the Workshop on Machine Translation in 2019 (WMT 2019). Material and Methods: Contact information for the authors of MEDLINE articles included in the English/Spanish (EN/ES), English/French (EN/FR), and English/Portuguese (EN/PT) WMT 2019 test sets was obtained from PubMed and publisher websites. The authors were asked about their abstract writing practices in a survey. Results: The response rate was above 20%. Authors reported that they are mainly native speakers of languages other than English. Although manual translation, sometimes via professional translation services, was commonly used for abstract translation, authors of articles in the EN/ES and EN/PT sets also relied on post-edited machine translation. Discussion: This study provides a characterization of MEDLINE authors’ language skills and abstract writing practices. Conclusion: The information collected in this study will be used to inform test set design for the next WMT biomedical task. Keywords: Parallel corpus, biomedical NLP, translation quality 1. Introduction In the biomedical domain, information is typically pro- vided in English through scientific publications and pa- tient oriented fact sheets prepared by governmental insti- tutions. Due to the large number of publications that is becoming available, machine translation methods support making this information available in languages other than English. Translation from a variety of languages into En- glish is also important in biomedicine. For instance, it can support researchers working on clinical reports, which are usually available in the national or regional language of the territory where the patient is treated. Given the lack of natu- ral language processing (NLP) tools available for languages other than English, translation can be envisaged as a pre- processing step (Campos et al., 2017). However, machine translation methods rely on the avail- ability of large parallel corpus for training, tuning and eval- uation (Koehn, 2009). Characteristics of language pairs, such as morphological complexity of the target language and relatedness of the two languages, are major indica- tors of success for translation systems (Birch et al., 2008). However, training corpus size is a bottleneck, especially for modern neural translation systems (Koehn and Knowles, 2017). The use of triangulation has been explored to com- pensate for the lack of parallel data in a specific language pair (Gispert and Mariño, 2006). Crowdsourcing has also …

Content truncated to 500 words