Where human accuracy meets the speed of neural machine translation.
With machine translation post-editing (MTPE), the language industry is finally overcoming the eternal struggle of man versus machine. It’s welding a partnership. It’s giving us “robot” and “human.”
Today, machine translation engines can process huge bulks of text and produce a decent translation output in a matter of minutes.
But decent is often not good enough. Pure machine translation can still feel clanky, unnatural. That’s why it gets coupled with a human process: post-editing.
Machine translation, sometimes referred to as MT, is a sub-field of computational linguistics that uses software to translate text or speech from language A to language B.
Although MT may seem quite a recent phenomenon, its origins can be traced back to the ninth century, when Al-Kindi, an Arabic cryptographer, developed techniques for systemic language translation. These included cryptanalysis, frequency analysis, probability and statistics, which are still used in modern machine translation.
Over the centuries, many scientists toiled with the idea of developing a tool making instant translation possible, but it was in the 1950s that MT research programs really started to pick up the pace.
However, it is only over the past 15-20 years that MT has ceased to be an academic topic and really impacted our lives. April 2006 market the launch of Google Translate, and since then the developments have been mind-blowing.
There are basically 3 types of machine translation:
Rule-based machine translation (RBMT): relies on linguistic and grammar rules developed by language experts and customized to specific niches. These rules dictate how a certain word or sentence should be translated in the target language.
Statistical machine translation (SMT): doesn’t rely on linguistic rules and words. Translations are produced by analyzing large volumes of existing human translation and the sentences are automatically mapped from one language into another.
Hybrid machine translation (HMT): uses multiple MT methods within a single machine translation system. This approach was developed once it became clear that the other MT methods could only deliver poor translation accuracy.
Neural machine translation (NMT): currently one of the most popular MT systems. NMT uses a large neural network to learn from every translation work and constantly self-improve.
At LingPerfect, we have developed a translation process in which raw machine translation is edited by professional human translators, correcting spelling mistakes as well as grammatical errors, improving the style and checking that the target texts are consistent with any existing terminology (machine translation software can be integrated with a CAT tool that uses client-specific translation memories).
This enables us to provide machine translation post-editing services make it possible to manage high translation volumes, enhancing quality and reducing time-to-market.
LPE makes the text coherent and error-free, with minimal human intervention. It's perfect for rush jobs and non-client-facing documents.
FPE refines the text further. It makes the translation stylistically appropriate, as well: opting for more powerful synonyms, better syntax, rhythm. Appropriate for client-facing content.
The benefits of Machine Translation Post Editing cannot be ignored. In some cases, it can save you up to 50% on costs and provide a four times faster turnaround time.
With a thorough post editing process, MT Post Editing services can bring significant savings in a variety of content. But there are a few caveats.
Fact 1: MT Engines are not equally efficient for all language pairs. While it works perfectly with some language pairs, we’ve seen that machine translation is still glitchy when it comes to other language pairs. It means that the time- and cost-effectiveness varies greatly from language to language.
Fact 2: Creative content is still best processed by a human translator. When your message contains specific cultural notions, puns, and wordplay, human linguists beat the machine by a long shot. In other words, what you get out of machine translation requires so much human editing that it simply isn’t worth the while. So keep your marketing collaterals, website copy, and social media posts in the hands of someone with a pair of eyes and whose name isn’t spelled in digits. In other words, if you want high quality translations, you still need “a little of that human touch”.