Machines can translate copy quickly and with increasing accuracy. An efficient and accurate way to localize materials is to use machine translation (MT) with human editing (machine translation post-editing or MTPE). However, as with any technology, successful implementation is a process. Here are seven steps toward effectively implementing MT.
1. Choose the Content for MT
Not all content is equally suitable for MT. Straightforward, structured, professionally written content works best; for example, frequently asked questions, social media posts, or after-sales customer care. Creative and literary content, such as advertising and marketing materials, is best localized completely through the work of human linguists. The first step, then, in implementing MT is to choose content that is well-suited to MT.
2. Take Care With Personal Data Rules
Machine translation engines vary in their level of security with personal data. Not all are compliant with standards such as HIPAA or GDPR. For example, when you use online translation engines such as Google Translate or Amazon Web Services, you give them the right to use your content however they choose, which could mean data ends up on the Internet. If regulations or policies require that you protect customer's personal data, check to be sure the technology you choose does so.
3. Train the MT
All MT engines require training, which requires human input. If possible, train the MT engine yourself using the words and phrases that your company uses most often. A stock MT will guess at the translation of unfamiliar phrases, but a custom-trained MT will translate them accurately.
The methods you'll use to train the MT vary somewhat depending upon the type of translation engine. Accurate translations require at least 100,000 segments. You can build your own corpora, access public ones, or buy corpora to train your MT. Be sure the corpora you buy or access includes data relevant to your industry.
4. Choose Your Post-Editors
After MT, you'll need either a light or full post-editing to ensure translation accuracy. Light post-editing changes only what is illegible, inaccurate, or offensive, while full post-editing fixes any errors, including cultural ones. Post-editors will review the raw MT output and change as necessary to prevent misunderstandings and cultural blunders.
Post-editors have linguistic training as well as technical training on how each type of machine engine translates and its common errors. They are flexible and willing to work with MT and can decide quickly whether to discard or edit an MT output.
In choosing the post-editors for your project, you'll consider their experience with the language pairs and machine engine type you're using.
5. Sample First
You are using MT to save time and money; however, if the output you receive from the translation engine is poor, you'll end up paying more in the long run. You can test first to see if the quality will be good enough to send to a post-editor. Generally, the larger the sample, the better. You can compare the MT output with that of a human linguist to see how far apart they are. If the translations are very far apart, you might want to choose a different kind of engine, do further training, or abandon using MT for this project. If the translations are close, then you're ready for MT. Note that MT output is better for some language pairs than others; for example, MT output for Spanish and Italian is often acceptable, while it typically isn't for Finnish and Japanese.
6. Agree on Pricing
Just as with any project, gaining agreement from all stakeholders on price is important before you begin. When determining a price, take into account the language pairs involved, the results of the sample tests, and the technology you'll be using. To obtain a rough idea of the budget, you can estimate using historical figures. Using technologies that can generate machine translation quality estimation or MTQE will provide a more accurate estimate.
7. Go for It!
Now, you're ready to deploy. Results may not meet your expectations immediately. However, as you continue to tune and train the technology, the MT output will improve. After each project, evaluate the results of your post-editing machine translation or PEMT to determine what needs improvement. Over time, you'll achieve the efficiencies you want.
Andovar can provide turnkey translation solutions, including MT and post-editing. MTQE scores let us know how accurate MT output will be before we start, meaning we can accurately estimate the cost. Our technology also automatically forwards passages to the most appropriate translation engines or human translators, as appropriate. Contact us by entering your email, and a project expert will answer all your questions. We can help you go global.