Machine translation (MT) has evolved significantly since its beginnings and its previous rules-based machine translation (RBMT) systems. RBMT simply wasn’t reliable and unfortunately it’s probably what you still associate with MT – but good news, no one uses it anymore!
Since MT started using Deep Learning technology in 2015 it has become increasingly more accurate. Essentially, MT is becoming more intelligent, it now uses algorithms to ‘teach itself’ the most natural word and phrase combinations for each language pair that it’s fed.
So why are we mentioning this? Well, if used correctly, it can be a fantastic efficiency ally when it comes to managing your website content, provided that you use a blend of machine and human.
Need some guidance on best practices? Read up!
The Basics: The evolution of machine translation
To really understand the development of MT and how it can actually help you in your website translation project we want to give you a quick history lesson. So, let’s take a step back in time and actually look at how it evolved.
The beginnings: Rule-based translation (the 50s-80s)
Where did it all begin? Interestingly, machine translators started being pioneered in the 50s during the Cold War for the simple reason that Americans needed a way to understand Russians. And computer systems helped them with this task.
Initially, the way MT worked was like a dictionary. You know the typical guide you buy to order a beer when you travel to Germany or ask for the toilet in Italy?
Well, this first version of MT worked similarly with exact equivalents in the other language. It was difficult though to get it to process whole paragraphs or complex phrases because it couldn’t understand the context.
The development: Statistical Machine Translation (80s-2015)
In the 1980s things started to look up. Scientists started establishing common statistical correspondences with exact equivalents and were able to get word by word or small paragraphs translated by breaking down input sentences into words, phrases, or syntactical arrangements.
The machine essentially crawled or used data from giant libraries of translated texts (a corpus), to find all the examples of words, phrases, grammatical structures, etc that the library texts contained, and then delivered the translation.
But the machine needed literally millions of sentences in two languages to understand the context and structure of a translation. To get this, abstracts were taken of the European Parliament and the United Nations Security Council meetings because these were available in all languages of the member countries.
This method quickly reached a limit though, as it was difficult to get it to take context into account and translate bigger texts and the more rules scientists created, the more complex it became.
The achievement: Neural Machine Translation
The next stage and where we are now is neural machine translation developed with DeepLearning – it has basically taken statistical machine translation to a new level.
The technology is based on neural networks: instead of telling the machine how a word or sentence should be translated, scientists gave it many different examples in machine language of how it could be translated and they let the machine give its own output of a sentence.
As we mentioned in the introduction, it basically teaches itself. Neural machine translation was built to replicate a human brain so it’s programming itself to constantly correct and improve itself.
Best practices when using machine translation
So now we’ve seen how far machine translation has come since the 50s you can start to grasp just how useful it can be when it comes to large translation projects such as website translation.
Within website translation, there are different ways to translate:
- Raw machine translation
- Machine translation post-editing
- Professional translation (from source language)
In order to use MT to its best abilities, there are a number of best practices to follow.
Do segment your website by content types
Do define sensitive content
Prioritize content that is very important to your audience vs. content that might not be essential to your business.
For example, your homepage is a priority and gets a lot of traffic, but your blog or your terms and conditions might not be, so make sure you draft a hierarchy of content to define which one needs to be professionally translated.
Or for example, some of your services are key to your business whereas others are not.
Do define a budget to translate your priority content
Your website should convey your message clearly in all the languages you support, especially when it comes to your priority content.
Make sure that for these areas you have defined a specific budget to get it professionally translated, or that you have internal native reviewers that know your brand well and can be invited to the project to modify the contents,
Do use machine translation wisely
Some areas in your website such as support pages, old blog articles, or product descriptions are very word heavy and would be very expensive to translate. These can be labeled as non-priority content so they would work well with machine translation and can be left untouched or you can just run a quick QA on them.
Don’t consider all your website content is equal
Your home page is where most of your traffic lands so it needs to be in your priority list, as well as your most visited pages and company description section.
Don’t use machine translation for all your website
Machine translation can automate most of your workflow and be very accurate for certain types of content, but it might not get all areas right, especially if the text is very creative or has jokes or references in it.
Best practice is to do some manual editing or order professional translations for priority content.
Don’t leave important content untranslated
Leaving some content untranslated can be a no-brainer if translating with a professional is too expensive and you have doubts about the quality of machine translation.
But if you have determined that this content is a priority to your business, you could be losing out on traffic and new customers if you leave it untranslated. Therefore we recommend you don’t translate other content that might not be so crucial so you can allocate a budget for the content that really matters.
So we’ve taken a quick look at the history of machine translation and how it has developed over the years.
And, I’ve shown you exactly where to use machine translation to your benefit on your website localization project. Machine translation isn’t the enemy – in fact, it’s a great efficient tool that speeds up the translation process and can for many cases only require some small edits.
Knowing how to use it to your advantage means your website translation workflow will be greatly improved.
Learn more about Weglot with a 10-day free trial – compatible with any CMS.