Output details
11 - Computer Science and Informatics
University of Edinburgh
462 Machine Translation Systems for Europe
<22> Originality: Largest comparative study on machine translation performance. Machine translation systems for 462 language pairs were built, evaluated. A regression analysis revealed that language distance, amount of reordering, and target side morphology are key factors in determining translation quality. The paper also tackles multi-source translation and pivot translation.
Significance: Study provided important insights into what language pair characteristics determine translation quality with the current state-of-the-art machine translation models. The paper also made other novel contributions on combining and pivoting machine translation systems.
Rigour: Experimental results on the most language pairs ever published.