As with all business decisions, choosing a translation method is all about balance. While just fifty-odd years ago, human translation – at an average rate of 2,000 words daily – was the only option, today’s LSPs (Language Service Providers) are translating up to 10,000 words a day.
The secret? AI translation-based technologies such as natural language processing (NLP), neural machine translation (NMT), machine learning (ML), and all that lies between.
Leading global brands no longer rely on human translation alone to scale their operations, as the process is costly and subject to manpower limitations. Instead, global leaders such as Netflix, Airbnb, Booking.com, Uber, Amazon, and more rely on the perfect combination of human translation and AI-based technology to achieve high localization quality at a lower cost, while meeting tight project deadlines.
MT and NMT: Where’s the Other 6%?
Machine translation (MT) has existed for two decades, but more recently the localization industry made the leap from MT’s statistical-based translation to NMT’s (Neural Machine Translation) neural network-based translation. This technological progress improved the quality of machine translations so much that they became a viable solution for many industries. Cost-effective, quicker, and more accurate than ever, NMT began the localization revolution, enabling global companies to translate content at an unprecedented scale. When done well, NMT can reach 94% of the quality of human translation. But that still leaves 6% room for improvement. Enter NMTPE…
Machine Translation Post-Editing
Today, when most people say they use MT, they’re really referring to NMTPE – Neural Machine Translation Post-Editing. Although the old, simpler name is the one that stuck, NMTPE uses machine translation, which can handle large volumes of text, and then follows it up with a review by professional, human translators. It’s faster and cheaper than having a person do all the work, but still has that human touch that makes the final result sound like a native speaker wrote it.
Scaling up with NMTPE
The speed and quality of NMPTE today make it a terrific option for all types of translation, including website translation, manuals, and even customer correspondence. Excellent translations, human review, and individualized brand glossaries to ensure consistency make NMTPE practically indistinguishable from human translation.
It also makes scaling a business significantly less costly. Editing a high-quality machine translation is significantly faster than translating from scratch. That means that the human translators working on NMTPE get through more content in the same amount of time.
With advanced translation engines, the total cost of translation is easily 30% lower than when using human-only translation.
For some industries, NMTPE is quickly becoming the standard translation method of choice. For example, eCommerce and tourism both rely heavily on NMTPE over human-only translation for their content. Companies like eBay and Booking.com have invested impressive resources into maximizing this tool. With the user-generated content that both industries use, volumes are so large as to make it nearly impossible for human translators to keep up.
Companies in other industries, such as gaming, can only incorporate NMTPE as a small percentage of their overall translation strategy. Especially for more sophisticated games, dialogue and other in-game texts are nuanced and tone of voice is important. Machines still can’t get it quite right.
BLEND’s NMTPE Process
1. Choosing the Translation Engine
There are a lot of translation engines on the market, and different ones are more adept with specific languages or industries. Therefore, the ideal choice of translation engine changes for every industry, topic, style, and language pair. Nonetheless, many organizations use only one translation engine. For example, the EU, which publishes content in 24 official languages, still uses only one translation engine for all language pairs and types of content.
To guarantee top-quality translations, BLEND tests 2-3 translation engines for every new client or project type before having human reviewers – native speakers of the target language – choose which one produced the best results. Then, the same engine is used for that client and content type going forward.
2. Translation Memory & Machine Learning
Hosted on the BLEND self-service platform, the translation memory is vital to brand consistency. With a customizable glossary based on past translations, it guarantees that CTAs, H1 headers, taglines, and legal terms are exactly how you want them, every time. For large-scale clients with massive amounts of data, machine learning is also used at this stage. This takes advantage of dedicated translation engines that are customized over time.
3. Human Translators
The final step is putting together a team of dedicated, hand-picked translation editors. These professional translators are prescreened, encouraged to communicate directly with the client, and given regular training about the client’s goals, services, and business model. By using the same post-editors for each project, the translators learn the company’s preferences and refine the translation memory. This, in turn, means less work for the translators, which leads to even faster results and lower costs.
The Right Balance to Support Global Expansion
The advantage of NMTPE is the balance it offers by juggling speed, quality, and price. With most major players already primarily using NMTPE, companies that hesitate will quickly lose their ability to compete in the race for new markets.
BLEND’s professional services team can help you find the combination of translation methods that is most cost-effective for your particular needs.