Alan Melby is Professor Emeritus of general linguistics at Brigham Young University and president of LTAC Global, a small non-profit dedicated to promoting the development and use of standards in the translation industry. He began working on a machine translation project in 1970. He is an ATA-certified French-to-English translator.

In 2014 he was elected to the FIT Council, the governing body of the International Federation of Translators.

Alan Melby

will present…

QT21: a new era for translators and the computer

 

Abstract

The analytic evaluation of raw machine translation in QT21 will use a revised version of the MQM framework, which was developed during the QT-Launchpad project. The revision consists of harmonization with DQF from TAUS.

MQM is a framework for creating customized translation quality metrics from a standard hierarchy of error categories. Each metric is based on a set of structured translation specifications. The emphasis of MQM is on analytic metrics, but its error category hierarchy also supports holistic metrics. Multiple metrics can be combined in a commercial evaluation procedure. For example, holistic evaluations of accuracy and fluency and be used to decide whether to conduct an analytic evaluation of a given translation.

It is important to emphasize that MQM is not a metric. It is a framework for creating metrics. A fundamental assumption of MQM is that no one translation quality metric can possibly be suitable for all translation projects. This is because MQM has adopted broad definitions of translation and translation quality. Please see the debate about broad vs. narrow definitions in the trilogy of articles by Koby, Hague, Fields, Lommel, and Melby in the current issue of Tradumàtica (http://revistes.uab.cat/tradumatica/issue/archive see 2014).

Details about MQM are available at: www.qt21.eu/mqm-definition .

DQF is also a framework for translation quality evaluation. It was developed in parallel with MQM by TAUS. Details about DQF are available at: https://evaluate.taus.net/ .

The harmonization of MQM and DQF is a deliverable of the QT21 project, and it will be completed before the 2015 Translating and the Computer conference. The authors of this abstract have been involved in the harmonization effort.

The Asling 2015 presentation will describe the results of the harmonization and how MQF/DQF will be used during the QT21 project.

In 2006, when one of the authors debated the future of statistical machine translation at the AMTA conference in Boston, he interacted with several young researchers in the field of MT and discovered that he was the first human translator they had ever met.

QT21 will begin to break down the barriers that have arisen between human translators and machine translation research. Involving humans directly in the process of evaluating machine translation and using their feedback brings translators and computers into a new era.