Machine Translation is not a completely automated process. Apart from actually building customised MT engines, localization teams have to put a considerable amount of energy into post-editing practices. KantanMT asked Chrome Data Solutions for some feedback on using KantanMT engines.
For those with little or no experience using Machine Translation, adapting internal systems for post-editing may seem like a challenge. We spoke to some of the engineers and post-editors at Chrome Data Solutions, who offered some insight into what to expect when Machine Translation is introduced into the localization mix.
Translators who took part in the interview had been involved with post-editing Machine Translation output from KantanMT engines since March 2013.
Firstly they were asked how they felt when taking on the role of post-editor, and without exception they all said that they had been slightly reluctant at first but then embraced the challenge.
After an initial learning period they said that they came to realise the real meaning behind the expression “garbage in, garbage out”, and that in order to produce a good quality translations, engines needed to be trained with good quality training data. This was a learning curve for the team at Chrome Data Solutions.
After a period of only five months the team had great success with integrating MT into their translation workflow and are happy that MT is increasing their producivity. They have “reduced translation time on average by 30%” (on projects where they use MT) and are continuously working on improving this.
Their advice to others who are considering MT:
1. Pay attention to details
2. Train your engine with good quality data from the start
Would you like to learn how Machine Translation can increase your business opportunities? Contact Kevin McCoy, KantanMT’s MT Success Coach: firstname.lastname@example.org
Are you already using KantanMT for your Machine Translation systems but would like to learn how to build better MT engines?
Sign up to KantanMT’s weekly webinar – Build, Measure, Learn.