The increased adoption of machine translation requires better methods for MT output quality estimation. And yet the widely accepted pricing model based on fuzzy matches has limitations, for example when applied to new segments. In the research community there are several well-established confidence estimation methods that are designed to predict post-editing effort and/or different machine translation quality measures applied to the translation unit level. This presentation will show a unique and more precise model for estimating compensation at the document level. Alexandru will show how fuzzy matches and machine translation confidence estimation fit into this model, and he will show how the domain post-editing effort is correlated with the post-editing learning curve.
Alexandru Ceausu is a computational linguist with a focus on machine translation. Alexandru completed his PhD with a thesis on statistical machine translation for morphologically-rich languages. At euroscript, Alexandru is part of the team that designs and develops customized machine translation engines deployed in different translation workflows. Before euroscript, he worked as a post-doctoral researcher at the Dublin City University in a project for machine translation domain adaptation. Before that, he worked as a researcher at the Research Institute for Artificial Intelligence in Bucharest, where he was involved in a multitude of European research projects.