by Manuel Herranz

We live the Age of Automation, Artificial Intelligence and Cloud. The global reach of the internet and the juxtaposition of global businesses and online commerce with local languages, means that large projects in the millions of words can be processed quite easily, as they are required in one direction. But translation project which involves machine translating 2,000 words can become complex task when you need to translate it into 35 languages.

The first case is typical of evidence that needs to be translated for litigation, eDiscovery, dialogues or even translation of technical manuals. It is typical to process massive amounts of data from one language to another. But many SMEs do international business nowadays, and these can be travel companies, online bookings, e-shops. All these companies require thousands (sometimes even millions) of words translated into dozens of languages. They also need them fast and high-quality neural machine translation is the answer – it is good enough. They also need the translation service to manage their needs continuously, seamlessly, on a daily basis.

Big Data and complex formats translation workflow

3d rendering of realistic isolated devices showing magazine landing page on screen: tablet, phone, computer and laptop on white background.

Here is the challenge our clients ask us frequently:

  • High volume texts that need to be translated daily
  • Requests at different hours of the day or night with tight deadlines
  • Files containing complex formatting
  • Different languages and different expertise levels

 

Old-fashioned translation workflows

So, how was translation managed at the beginning of the internet? E-commerce is probably as old as the first internet web browsers, dating back to the early 90’s. And those 30 years sound like ancient history in a world where the amount of data doubles every two years for decades.

Over those 30 years, translation services tried to tackle the problem by employing a lot of multilingual staff and translators, all of whom were busy segmenting files and sending them to local translation agencies across time zones. Many translation companies de-localized and opened offices in different time zones. Everything was done via FTP, email, parcel delivery or even worst – snail mail.

However, as people became more used to the Internet, and dial-up gave way to faster cable connections, customers’ habits begun to change. Everything began to be available online: information, products, services… literally everything. Customers became used to getting answers fast. Shorter delivery times became the norm. Companies began to struggle to keep up with demand across the world. Automation was needed to keep track of all this. Translation became too cumbersome, too complicated, and too time-consuming to be extracted by hand, sent and uploaded manually.

Recent translation workflows

As it often happens, when there’s a need, there’s a way. The first Business Process Management systems were born. These were a systematic approach to automation by making an organization’s workflow more efficient and more capable of adapting to a changing environment – and the concept was adapted to translations. Some translation processes became automated and local servers, such as the old WorldServer, were used to speed the process up.  WorldServer became a standard until it was purchased by SDL looking to consolidate its position as the leader in content delivery (and dominating the whole process, from content extraction and filtering to packaging as a translation file to serving the translation environment). The workflow became more manageable and CAT tools more popular, but the speed at which translations were delivered still left a lot to be desired.

Modern translation workflows – and neural machine translation

Then came cloud technology and the era of the API.

Cloud technology was the leap forward many technologies were waiting for: it enables quick access to translation and editing platforms, and customer’s resources from anywhere. There is also no need to be tied to a physical device, like a PC at home or the office to extract and upload content or place an order. It also saves money not only by offering backup and storage but mainly because it allows translators to work simultaneously in parallel workstations from their home office. It allows sharing joint Translation Memory resources (see ActivaTM) and to communicate in real-time about the project with the other team members. Pangeanic’s Cor platform, for example, just allows for that.

And since it is easy to access core functionalities and data from different programs, why not have them talk to each other? API stands for Application Program Interface. That means a “pipe” through which a customer can send and receive source text and even large files and receive them back, fully translated. The objective? Content can be translated collaboratively, simultaneously and dynamically. Cloud editing and API combined are a “power couple” that speeds up the translation workflows.

The age of Pangea Machine Translation

If you have a large amount of material in a Content Management System (CMS) that needs to be translated in real time, PangeaMT’s translation API is designed to do just that. Scaling with the needs of the company, it integrates seamlessly into your CMS. All everything happens on the cloud.

When new material is posted on your website, Cor picks it up and it either machine translates it automatically, without any further investment or staff training necessary or requests human translation services. Either way, translated content is uploaded on your website. You can choose (and we always recommend) that the final translated files are reviewed by a proofreader before they are officially posted in the client’s CMS as multilingual content.

How is this done? 100% of our process is automated. This starts with an API request once content is published on the customer’s website. This triggers a process which involves job creation and sending the text to qualified translators, who work on the cloud on industry standard and format-agnostic XLIFFs files, in a collaborative way. Watch this video to see how content can be extracted from the outside, without the need to install any plugins.

Using Translation Memory is another way of shortening translation time and reducing costs.Translation Memory means storing every segment that is approved by a translator in a database. New requests (sentences) are checked at lightning speed versus all the existing translations and if a similar match, this is shown to the translator. For example, the sentences

"There are compelling reasons why we might wish to colonize an asteroid belt, but the predominant one is mining." (19 words)
"There are compelling reasons why we might wish to colonize Mars, but the predominant one is mining." (17 words)

share 89% of their content, as the translator only needs to substitute “an asteroid belt” for “Mars”. Translators can concentrate on spotting the differences and quickly translate the word or words but respecting the style. This is not machine translation, it is just re-call. Neural machine translation changed the game in 2017. Neural machine translation developments from 2017 meant that deep networks could produce near-human output which required light editing by humans to be comprehensible and publishable. Some clients have decided that neural output is good enough to be published or at least to be used as legal evidence.

Which challenges does Pangea Machine Translation API solve?

  1. High volume content delivery and publication are possible. PangeaMT translation API is designed for high-capacity clients. Aside from the most cutting-edge and most stable technological in the translation industry, there are over 4,000 certified translation experts standing on the other side of the user interface. For machine translation, there is no upper limit on the amount of text the API can process – and no lower limit, either. Human translation services are always an option for “final human touch”.
  2. PangeaBox can be used from your desktop. Just drop a docx, pptx, xlsx file into our app, choose the language pair and receive the file translated back in seconds, fully formatted!
  3. Cor is ultra fast – Once your CMS is integrated into our translation API, there is nothing else you need to do. You do not need to log into any platform, there is no need to send any emails, nor prepare any files for translation. Cor will take the content and apply artificial intelligence to select the most suitable translators for the job by analyzing the content, translator availability, time zone, ranking and experience in similar jobs, etc. Pangeanic’s certified translation professionals will get to work, and the translated material will be loaded back into your CMS.
  4. It’s easy and cost effective. Once the API is integrated into the client’s CMS, there’s nothing the client needs to do. Since the client’s workflow remains unchanged, there is no further investment in extra staff or extra resources. Everything proceeds as it always has. The translation work happens the way it’s supposed to: and the magic happens behind the scenes.