Designing a Learner Translator Corpus for Training Purposes

This paper presents the development of the MeLLANGE corpus, a multilingual, aligned and annotated learner translator corpus (LTC). Unlike other learner corpora, MeLLANGE focuses on translation-related rather than language acquisition issues. The corpus thus contains students' performance into their mother tongues. Moreover, the intention is not to deliver a repository of students' errors, but rather give both trainers and trainees the opportunity to identify possible translation problems, as well as possible solutions, in a multilingual, corpus-based environment. The information attached to the corpus is divided into metadata regarding the source text, the translator and the translation situation, linguistic annotation, and error-annotation according to an error typology developed within the project. We show examples of how the MeLLANGE corpus can be exploited in teaching translation from a data-driven point of view, and from a data-based point of view.