Measuring post-editing time and effort for different types of machine translation errors

Abstract

Post-editing (PE) of machine translation (MT) is becoming more and more common in the professional translation setting. However, many users refuse to employ MT due to bad quality of the output it provides and even reject post-editing job offers. This can change by improving MT quality from the point of view of the PE process. This article investigates different types of MT errors and the difficulties they pose for PE in terms of post-editing time and technical effort. For the experiment we used English to German translations performed by MT engines. The errors were previously annotated using the MQM scheme for error annotation. The sentences were post-edited by students in translation. The experiment allowed us to make observations about the relation between technical and temporal PE effort, as well as to discover the types of errors that are more challenging for PE.