Neural machine translation (neural MT or NMT) is a game-changer for Asian languages such as Chinese, Japanese, and Korean—even more so than for Western European languages. We started to see it almost three years ago, with the initial steps for NMT in production environments, not only for understandability (gisting) but also for MT post-editing. This is one of the most exciting developments I’ve seen in my 13 years working on machine translation. And as we saw in the recent TransPerfect Global Content Forum in Hong Kong, our clients are starting to see the advantages of inserting MT into the the localization ecosystem.
Neural machine translation allows any world language to be machine translatable with acceptable quality. One of our first successes was in the media industry for Indonesian—a language that can be challenging. We managed to train a machine translation solution and deliver the project in the client’s required turnaround time. In order to make sure that the project’s success was not only due to the content type—in this case subtitling—we decided to carry out pilots for other accounts and projects.
In the travel and hospitality industry, you can find many types of content—such as property content—that are well-suited for machine translation because of their terminology and style. We trained an engine, ran a few improvement rounds prior to production, and managed to get the edit distance to the right level so that we can combine trained machine translation, light, and full post-editing services in our client's website for Chinese, Japanese, Korean, and Indonesian. This tiered approach to content is key for large websites, not only in travel but also in retail, and it can help our clients achieve the optimum return on their localization investment.
But we didn't stop there. We wanted to try with our largest vertical (life sciences) for high-risk clinical content. Thanks to many of our clients piloting AI-based workflows, we have trained engines in a wide variety of Asian languages, and the results have exceeded our expectations. We are able to reduce costs and turnaround times, while not decreasing quality. Quality is essential for our clients and therefore we keep the same number of human steps (two or three as required) so that client reviewers cannot see a decrease in quality in our blind tests. In this vertical, getting the engines to their optimum level is essential before going live.
One of the main benefits of the machine translation engines we train is that they can be used not only for human workflows but for additional applications as well. The most successful to date is a private, secure machine translation portal for companies that want to replace the use of publicly available, online translation tools. TransPerfect's AI portal enables organizations to translate in real time, not only short texts but also documents in any format using the company's trained engines. Besides the portal, the engines are accessible via APIs: MT engines can be integrated into any application such as customer support, chatbots, or e-discovery, among others.
For more information on how to start integrating artificial intelligence into your translation workflows, email email@example.com.