Skip to content

Machine translation and the future for translators

The rise of the machine translators!

Machine translation is a term used to describe automated software that translates source content (speech or text) into a target language. Often machine translation (MT) software is operated by human translators who “supervise” the machines. However, MT software is capable of operating without human support.

MT software can translate vast amounts of information far quicker than traditional (human) translators. This enables companies to cut costs and speed up their processes.

However, MT systems do require “training” (and occasionally, adult supervision!) in order to correctly learn the intricacies of a language pair. MT software also needs to “learn” vocabulary and sometimes, detailed information regarding the subject of the text it’s translating.

Machine translation is often used in conjunction with traditional translators. A human translator or editor can post-edit copy that has been first translated by a machine. This involves checking the translation for accuracy, as well as phrasing to ensure the language sounds natural.

Translation companies are increasingly using MT software alongside their existing resources. In fact, in 2016, SDL (one of the largest LSP’s in the world) revealed that they translate 20 times more content using MT than with human translators.

 

Types of machine translation software

 

Generic MT

Generic MT refers to the big platforms that many of us use daily – Google Translate, Bing, etc. These platforms provide instant, ad hoc translations. It’s possible to pay these platforms to connect directly via API, enabling batch pre-translation.

However, we’ve all had a giggle at Google Translate’s expense at least once! Generic MT is the first generation of machine translation software, and although it’s improving, it still makes many mistakes.

 

Customisable MT

Customisable MT can be educated in terminology for specific fields. It learns. For example, eBay’s MT software has learnt to understand and correctly translate all the abbreviations commonly used in electronic commerce. This type of MT is often used in areas like medical or legal translation, where terms must be precise and the content is often repetitive.

 

Adaptive MT

Adaptive MT works with human translators by providing suggestions, in real-time. In this way, it’s similar to predictive text. However, adaptive MT also simultaneously “learns” from the translation input of its human partner. Therefore, it continuously improves it’s own performance!

Adaptive MT has been shown to dramatically increase productivity, and may well come to define the translation industry of the future.

 

How does the software work?

Fundamentally, there are three different kinds of machine translation software: rules-based, statistical, and neural.

 

Rules-based systems

Rules-based MT software makes use of numerous algorithms — which recognise grammar, syntax and language patterns (phraseology) — together with a dictionary for regular words. A rules-based system usually delivers coherent translations, but which may lack subtlety or nuance.

 

Statistical systems

Statistical MT systems, on the other hand, don’t have detailed knowledge of language rules and workings.  Rather, they “learn” how to translate by processing huge amounts of data. As the processed data increases they are able pattern-match with existing copy.

It’s possible to develop Statistical MT for a specific field, by feeding it related data. Usually, the statistical systems deliver highly fluent, but sometimes, less consistent translations.

 

Neural Machine Translation (NMT)

Neural Machine Translation is the latest innovation! NMT uses machine learning technology which enables the MT to learn how to produce the best results. These platforms require enormous processing power, as machine learning is basically attempting to copy or replicate the processes of a brain.

Many MT enthusiasts, researchers, and developers are switching to Neural MT because of its exceptional performance, compared to the more rigid rule-based and statistical systems.

 

Should we welcome machine translation?

  • Machine translation is faster

When it comes to speed, machine translation is vastly superior to human translators. Computers can translate faster than even the most seasoned human!

  • Machine translation can save money

For many organisations, this is the main attraction in using machine translation. It is far cheaper to buy MT tools, than it is to pay a human.

  • Machine translation may improve security

MT help improve data security by reducing the interaction humans have with secured content. Although for some, this is reminiscent of Cyberdyne Systems, and quite scary!

 

In conclusion

Currently all machine translation software requires for post-editing by a human. That said, it seems machine translation is here to stay.

However, it is unlikely that it will ever be able to replace human translators in every case.

It is possible that MT will learn how to accurately translate prescriptive copy — such as legal contracts or insurance forms — without human assistance. However, when is comes to text that needs to be able to “reach out” and fully engage with an audience – it is unlikely that machines will ever replicate the subtlety and excellence of a professional, human translator.

Related Posts