Neural machine translation (NMT) is the new black. In less than three years, it has already accomplished what its predecessor statistical MT managed in 10 to 15 years, namely becoming the de facto standard for automated translation.
Tech giants such as Google, Microsoft, Amazon, and even Salesforce lead the way with their implementation of NMT, and key verticals in high-tech industries will also soon benefit immensely through the carefully managed incorporation of NMT technology into their localization framework.
It is also clear that language service providers (LSPs) will need to embed the technology within their key service offerings in the very near future.
Managed MT: Bringing the Benefits of NMT to Enterprise Translation Buyers
Take Milengo, for example, who are preparing to launch a new service category called Managed MT.
“We have always been early adopters of MT technology throughout its various incarnations, whether it be rule-based, statistical, and now neural MT,” explained Milengo’s CEO, Roman Kotzsch. “That’s because we have always been very quick to recognize the value that MT provides to our enterprise customer base when managed expertly.”
Indeed, MT has been providing value to Milengo’s customer base for years now. This has been borne out by the example of telecommunications software company Interactive Intelligence (ININ). Their business objective back in 2009 was to find a low cost but quality-oriented localization solution for their large volume help content.
The large translation volume meant traditional human translation was far too expensive, and therefore it was considered low priority in terms of localization. Through the custom development and management of SMT-based localization workflows for six languages, Milengo was able to achieve overall cost savings ranging between 30% to 50% when compared to ININ’s previous pricing model.
But now it seems that even more is possible with NMT.
Nearly 80% Faster Time to Market at Almost 80% Less Cost
“We saw that statistically-based post-edited MT (PEMT) achieved good but not outstanding cost and time-to-market improvements,” Kotzsch explained. “Standard implementations typically delivered a productivity of 400-550 words per hour compared to traditional human translation, which is at 300-350 words, i.e. an increase of 46% on average.”
“Now our first tests with our new neural-based Managed MT engines have shown a significant increase in output of automated translation quality, which has increased linguist productivity to around 2,000 words per hour. As a result, time-to-market for localized content can be reduced by almost 4 times when compared against the former statistically-based MT processes,” Kotzsch said.
Milengo’s early tests indicated that traditional full post-editing (PEMT) models may well soon become unnecessary and not cost-effective enough for enterprise translation buyers, when considering the sheer improvement in the quality of raw MT output. As a result, Milengo are now in the process of redesigning their review processes to allow for faster identification and correction of NMT output errors.
The ultimate aim for Milengo is to pass on the resultant cost benefits to its translation buyers through the increased efficiency of their Managed MT solution. As time-to-market is reduced by almost 80% when compared to the traditional translation model, customers can expect to pay only 20 – 25% with Managed MT compared to traditional translation costs.
“With Managed MT, we want to offer our customers the best possible price to quality ratio by capitalizing on the increase of translation productivity through NMT technology. We will also offer a major further advantage that is very new to the technical translation market, such as a flat fee subscription model similar to new consumer offerings from Spotify, Netflix, and so on,” Kotzsch said. “This will make localization budget planning much easier and more convenient. Translation buyers have every right to manage their localization needs on their terms, and we are confident that our subscription-based pricing model gives them this opportunity.”
Key Application Areas For Managed MT
NMT still has a way to go before being able to tackle highly creative content and low-resource languages. However, many existing MT engines are already proving useful in verticals where mainly technical accuracy is paramount, and where the high volume of content makes human translation or even PEMT very expensive and slow for enterprise translation buyers.
Some examples of such content types are technical documentation, product descriptions, and online help and other high-volume product support content.
Milengo is currently offering translation buyers from high-tech industries the opportunity to participate in free pilot projects to verify whether Managed MT—with its ultra-high productivity and low costs—could work for them for the above-listed content types. At this point in time, Milengo is limiting the scope of the pilot projects to English and German, but will look to add French, Italian, Spanish and Chinese very soon.
If you are interested in participating in Milengo’s Managed MT free pilot project to evaluate this high-performance solution for your company and content, you can request a consultation here.