Sources de biais dans la traduction humaine et la traduction automatique post-éditée de résumés médicaux Cochrane

Abstract : The aim of this study is to compare human translation (HT) and post-edited machine translation (MT) in the medical field in terms of translational bias. Cochrane systematic reviews aim to provide up-to-date, impartial information on the effectiveness of interventions, in order to support evidence-based medical decision-making. Translation of Cochrane review abstracts plays an important role in the multilingual diffusion of medical knowledge. Cochrane France has been translating Cochrane abstracts into French since 2010, first using specialized translation providers and since 2013, integrating statistical MT and partly volunteer post-editing into the process. Presentation of research results should be exempt of errors and as objective as possible. We define sources of translational bias, then, as elements in these translated texts that are susceptible to interfere with that communicative objective, and have the potential to impact readers’ interpretation of trial characteristics (interventions, participants, outcomes…), their results, or the authors’ confidence in them. We present here some first corpus observations from this study in an attempt to characterize the potential differences between translations produced by the two processes. Our preliminary research suggests that HT and post-edited MT differ in terms of errors and biased translations. While terminology and syntax errors are observed in both sub-corpora, such errors appear on a more elementary level in the post-edited texts, most likely due to the influence of MT. The HT sub-corpus, on the other hand, contains more biased translations of modal markers, especially hedges, which we attribute to the greater degree of interpretation involved in traditional translation.
Document type :
Conference papers
Complete list of metadatas

https://hal-univ-diderot.archives-ouvertes.fr/hal-01496863
Contributor : Hanna Martikainen <>
Submitted on : Tuesday, March 28, 2017 - 12:43:47 AM
Last modification on : Friday, January 4, 2019 - 5:33:30 PM

Identifiers

  • HAL Id : hal-01496863, version 1

Collections

Citation

Hanna Martikainen. Sources de biais dans la traduction humaine et la traduction automatique post-éditée de résumés médicaux Cochrane. Congrès Mondial de Traductologie, Apr 2017, Paris, France. ⟨hal-01496863⟩

Share

Metrics

Record views

75