17 janvier 2013
info:eu-repo/semantics/OpenAccess
Victoria Arranz et al., « The A to Z of Manually Translated Parallel Corpora for the Evaluation of Machine Translation: Establishing Protocols through Experience. Session 5 - Assessing Quality in MT », HAL-SHS : linguistique, ID : 10670/1.4286ic
The evaluation of Machine Translation (MT) systems often requires the use of reference translations in order to compare the MT system output against those translations. This comparison allows measuring the quality of the automatic translations. Such reference translations are generally produced by professional translators who follow a series of well-defined protocols and guidelines defined for the specific evaluation data (or corpora) to be produced. The current paper aims at describing in detail how the entire evaluation-data production procedure is carried out. This will imply going through the different stages and needs of the project, the constitution of guidelines, the setting up of protocols and teams, etc. All of it is done bearing in mind that one such project is a dynamic entity and that it evolves as requirements change, feedback from teams is received and as we learn to optimise it through experience.