Upgrade to ChromeUpgrade to FirefoxUpgrade to Internet ExplorerUpgrade to Safari

Ratchadapisek, Bangkok, Thailand

Will AI be responsible for the decline of the human translator?

Posted 08 July 2019

Request a free quote Request a call back

We hope not, though there are so many changes afoot at the moment. Who’d have thought that the advent of the Internet age would signal the gradual decline in bricks and mortar stores, or that Smartphones would make photographers out of all of us, without us having to invest in high quality cameras?

Then there’s the possibility of autonomous vehicles which require no drivers to deliver goods. The impact of these changes on jobs and daily living are endless.
The speed of progress must be a huge challenge for even the savviest of companies, so what impact will AI have on the human translator?

Will they be replaced? Some think so, and companies like ours are being kept on our toes as we feel the huge steam roller steadily heading in our direction.

But we’re not there yet!

Humans versus machines?

Machine translation (MT) is not new technology: the first public demonstration took place in 1954 when a handful of Russian sentences were machine translated into English, live at IBM in New York.

Since then, MT has come on in leaps and bounds, both saving valuable time and money and offering other benefits for organisations across various sectors.

Nonetheless, MT systems are still not suited to every business or document type.

Types of MT

• Statistical (SMT) – relies on huge databases of existing translations, and algorithms match up the most likely translation from part phrases known as n-grams (words are broken into short segments and work in a similar way to sat-nav technology drawing from a database of voiced n-grams – hence the occasional mispronunciation of place names). Words are grouped together in 3s for example, so a 6-word sentence would be made up from two sets of accurate 3-gram segments, not necessarily combining to make a fluent whole.

• Neural MT (NMT) – also known as ‘deep learning MT’ which provides a more sophisticated and natural output. NMT analyses whole phrases and recognises word similarities. This is ideal for languages where sentences are often reversed, such as German. However, this isn’t necessarily the holy grail as it can only translate individual sentences that don’t necessarily work well together as a whole and it takes a lot more ‘engine’ training to produce good results.

• Hybrid engines – mixing two different methods to refine translations further.

When considering Machine Translation, ask yourself two questions

1. What are you translating?

MT systems handle technical documents and instruction manuals well, for example, but aren’t suitable for more creative and marketing style documents.

2. Who is going to read the translations?

Machine translation that hasn’t been post-edited is unlikely to be good enough for your clients or end users. They need clear, accessible information, and MT is still not yet at this stage in its development.
But if you’re simply looking to share a few documents with some colleagues internally, MT may be the perfect solution – you might need only the gist of a report for example.

When only the human touch will do

A good use for machine translation output is to combine it with human post-editing, so you gain the benefits of technology combined with the accuracy of human translation.

However, there are content types where only a human will do, these include:

• Creative content – marketing collateral and web pages
• Specialist technical or scientific subjects
• In less commonly spoken languages with fewer corpora for effective MT Engine building.

And then there’s the cultural side – end users will come from all sorts of different cultural backgrounds and while MT can provide a literal translation, it’s a long way away from understanding language nuances and cultural differences. Only a human translator is capable of tailoring culture-specific content to suit a particular target audience.

Do you think MT is for you?

Could your company benefit from MT?

ALM translations uses MT as part of its workflow where appropriate. If you have a large volume of previously translated material (in bilingual versions), ALM Translations may be able to create a bespoke Machine Translation engine for your company. ALM also offers a post-editing service, specifically for Machine Translated text.

Take a look at our technology pages for further information.


Request a Call Back