Amazon Web Services has launched a new feature for its Active Custom Translation (ACT), which allows its users to customize their machine translations. The feature provides users more control over their translations by allowing them to include multiple translation examples in TMX, TSV or CSV formats to make parallel data (PD) sets along with batch translation job to customize the translation. Amazon Translate claims to provide high-quality language translations to its users at affordable prices.
According to the blog post describing the update, the ACT learns from various PD sets to provide customized outputs. In an example stated in the blog, if your PD states that ‘How are you?,’ ACT translates it to ‘¿Cómo está usted?’ instead of ‘¿Cómo estás?’ as it knows the customized translation for ‘¿Cómo está usted?’. Earlier, professional translation was done by providing a huge amount of customized translations to the model that had to be updated regularly in order to get accurate translations. The entire process often turned out to be extremely expensive and time-consuming. If the models were neglected and not updated regularly, the translations would not be accurate due to old models.
ACT eliminates the need to build custom translation models by being updated with the users’ parallel data. The major difference between the antiqued translation method and Amazon Translate ACT is that ACT can provide high-quality translation outputs even outside its PD sets domain whereas the custom translation models tend to perform worse than a generic model outside their domain of customized data, the blog states.
Amazon Translate charges the user based on the number of words translated that remains the same with the launch of this updated service. The PD sets can also be updated as per the users’ wishes at no extra cost.
Multiple translation groups and companies including Custom.MT, Welocalize, and One Hour Translation have also provided positive feedback on Amazon Translate ACT after using the service to serve their customers without incurring additional hosting or training charges.
In the same blog post, Marcello Federico, principal applied scientist at Amazon Machine Learning, says, “innovation is in our DNA. Our customers look to AWS to lead in customization of machine translation. Current custom translation technology is inefficient, cumbersome, and expensive. Active Custom Translation allows our customers to focus on the value of their latest data and forget about the lifecycle management of custom translation models. We innovated on behalf of the customer to make custom machine translation easy”.