MQM tags can be allocated to any partial sections of segments. Values can be, for example: "Wrong terminology", "bad phrasing", "wrong translation" etc. So, an MQM tag marks the respective paragraph to a certain, qualitative extent. In addition, an MQM tag-specific comment and an evaluation (of grave, mean, or light mistakes) can be assigned to each MQM tag.
Inserting MQM tags in the source
MQM tags can also be inserted into the source language and separately evaluated accordingly. This way, even translation errors, which already originated in quality problems of the source language, can be gathered in a useful way. Whether this is possible or not can be set when creating the project.
Determining MQM settings
It can be determined in translate5 which MQM types are available by default for a task. In addition, when creating a task, each task can be given its own set of MQM types only valid for this task.
Please have a look at the structure of translate5's zip import package format to see, how this works.
Background on MQM
MQM quality assessment on subsegment base was inspired by and implemented for QTLaunchpad. In short, the multi-dimensional quality metrics (MQM) developed by QTLaunchpad includes a concept and a procedure to define and measure translation quality according to uniform standards - regardless of whether the translation was done by a human or a machine. Before QTLaunchpad, no approach for reliable comparability existed. QTLaunchpad is a project supported by the EU by DFKI (Deutsches Forschungsinstitut für künstliche Intelligenz), Dublin City University, University of Sheffield and Athena Institute for Language and Speech Processing. Therefore, translate5 presents an ideal setting for measuring translation quality with the criteria of QTLaunchpad.