What is machine – or, as it’s most often called these days  – AI translation?

Having your product or content translated is kind of like ordering a delicious meal for someone else.

You’ve already enjoyed it (your home market content), and you know it is wonderful. You want your new friend (your new market) to try it, so they order that particular meal.

Instead of the mouth-watering dish made from scratch with freshly harvested ingredients that you raved about, they get a pre-packaged frozen meal sloppily tossed onto a plate (machine-translated content).

This is the difference between a great professional translation and a machine translation.

I recently experienced this first-hand. I was looking into Customer Relationship Management (CRM) systems and logged into a popular CRM. I was redirected to their Italian site, and discovered that I couldn’t manage a site that was supposed to be in my native language. The company had used machine and AI translation on their entire site.

We ended up having to call support, because the Italian translation was so bad, we couldn’t figure out how to put the site back into English.

Machine translation is a hot topic these days. After this experience, and seeing so many SaaS companies scaring their customers away and leaving lots of money on the table, we put together this article to help you understand how machine translation works and some of the problems you will experience with it.

What is machine or AI translation?

First things first. Machine translation has become a hot topic in language services and marketing circles. Translators everywhere grow increasingly frustrated as more people point to machine translation as a replacement for their skills. News flash! It isn’t true.

According to Wikipedia, machine translation is a sub-field of computational linguistics that investigates the use of software to translate text or speech from one language to another.” Basically, using computer power to translate speech or text to another language. Researchers continue to work on making translation a service accessible to everyone, and they have come so far. To really have the discussion about machine translation, we need to dig into the history a little bit, so read on.

Where it started, and where it is now

The idea of machine translation first popped up in the 1940’s, and researchers started actively working on it in the 1950’s. Governments all over the world funneled a ton of funds into machine translation research, and there was some modest success. Funding dried up in the 1960’s when countries began to realize what was promised couldn’t be delivered. For example, in the U.S. a ten-year report was published showing the woeful lack of progress, and machine translation wouldn’t get decent funding again for many years.

Researchers struggled so much because there are very complex mental processes taking place during translation. As a translator, my knowledge needs to encompass culture, idioms, context, intent, and grammar for both the source and target language. That is where machines struggle. Scientists have been trying to teach computers these mental processes.

How can computers be taught to think like a person, when we don’t completely understand the human mind ourselves?

How do computers translate?

There are a few different approaches scientists have taken throughout the years. The main ones are: rule-based, statistical, example-based, and neural machine translation.

●  Rule-based: This was the first approach. It is based on dictionaries and semantics. A rule needs to be created for every possible situation, which is what makes this method less efficient than the others. It may work well for something like a weather report.

●  Statistical: This approach works best when the computer has millions of human-translated documents to work from. Google switched to statistical machine translation in 2007 and used documents from the UN to train their computers. It doesn’t work very well for ambiguous situations.

●  Example-based: A Japanese scientist proposed this method in 1984. Using example texts previously translated by humans, a computer compares the new text with phrases in its database and reconstructs a translated text based on the knowledge it has.

●  Neural: This approach is the latest and greatest. It uses an artificial neural network, basically artificial intelligence, in order to make the best educated guess about the sequence of words in a sentence. Google and many other translation programs use this approach now.

What’s wrong with machine-translated websites?

Before we dig into this further, we need to make one fundamental consideration:

Our goal as translation expert for our clients is for their customers not to realize the content they’re reading is translated. Machine translation hasn’t reached that point, and we may never see the day that it does. Even if a site visitor doesn’t realize it’s machine translated, they’re going to assume that a poor translator was hired to work on the site, and they’ll most likely leave.

You need to consider what goes into creating your site and marketing content in the first place.

All that effort you put into your marketing strategy? POOF!

Creativity and ambiguity

Machines struggle to accurately translate witty titles and clever taglines. Creativity is the Achilles heel of machine translation, and this is particularly vital to the sort of translation I find myself doing most often: marketing copy translation.

Ambiguous words or phrases are another issue that often trips up machine translation programs. Take these two sentences for example:

  • The pen is in the box.
  • The box is in the pen.

The Economist credits machine translation pioneer, Yehoshua Bar-Hillel, with this realization: In English, the word “pen” is the ambiguous word in these two sentences. The second sentence alludes to a box in an animal pen perhaps, since there is no writing pen for a box to fit into. Many machine translation programs still take the statistical approach to resolving ambiguous words, but in this case, they would be wrong.

Search Engine Optimization

Ahh, yes, people often forget about this aspect of translation as it relates to global marketing. What’s the point of having your website translated if it’s just going to be buried in search engine results? Machine translating programs don’t take SEO into account. So all of that careful keyword research? It won’t be conveyed onto your translated pages.

When I translate website and marketing content, I have to do keyword research, just like the original writer of the material. Marketing teams put a lot into researching and testing the best tags, calls-to-action, and links, all of which machine translation botches with exact translation.

Google doesn’t like Google Translate

Yes, it’s true. The world’s biggest search engine penalizes sites that machine translate their web pages. Google doesn’t like any machine translated content that doesn’t at least get human review, and they fervently guard their users’ experience.

Another thing to consider is that machine translated content might be considered duplicate content by the search engine’s web crawlers, penalizing your site. Google wants each translation to be hand-vetted in order not to be considered spam.

Confidentiality and data leaking

If you haven’t been put off by the huge potential of losing customers, delivering a bad user experience, messing up you CTAs, SEO keywords, brand message and Google penalizing machine translated content and your website, I got one more for you: Confidentiality breach and data leaking.

I will just copy and paste an extract from the Google Terms of Service:

“Your content in our services”

Some of our Services allow you to upload, submit, store, send or receive content. You retain ownership of any intellectual property rights that you hold in that content. In short, what belongs to you stays yours.

When you upload, submit, store, send or receive content to or through our Services, you give Google (and those we work with) a worldwide license to use, host, store, reproduce, modify, create derivative works (such as those resulting from translations, adaptations or other changes we make so that your content works better with our Services), communicate, publish, publicly perform, publicly display and distribute such content.”

Now, you might want to think twice before you input your sensitive content in an online translation tool.

There are other issues, such as the fact that these engines prompt users to improve translations. And the problem lies within the fact that users don’t need to have any translation or language / industry-specific skills. So, let’s think again: How sure can you be that the content is accurate?

When can I use machine translation?

Machine translation is not completely useless. You can still do good things with it. A few ideas:

Gisting content: You can use machine translation when you need to get a general idea of the content. Honestly, this only works well with particular language pairs. In most cases, your content needs to be translated through several languages that may be more popular, and I’ll let you figure out the results.

●  Testing potential languages: You might use a plugin or program to TEMPORARILY translate your site to gain some insight into the demand for a particular language. Google Analytics can show you which countries your visitors are coming from. Finding content in their language (even if it is not translated well) will attract new visitors.

●  Informal message translation: Maybe you want to communicate with family or get some simple message across. Don’t use idioms, and keep your sentences simple. As long as you don’t have one of the harder-to-convert language pairs, you should be okay.

If your goal is to establish rapport with your new target market, the stilted slightly-off language of machine translation isn’t the way to do it.

Solution? Determine Your Needed Languages and Work With a Pro!

Even researchers don’t see machines replacing the human translator completely. They see machine translation’s true effectiveness being in mass triage of documents or audio. That isn’t your marketing or website content though, is it? You, or perhaps your marketing team, carefully created your content, and you need a human expert to make sure the core concepts, nuance, and intent are apparent in the translated language.

You already understand the value of well-written content; Your team carefully researched and tested various CTA’s, headlines, landing pages, and taglines, and looked into and developed ideal client personas to better target your audience and provide content they wanted to see.

Working with a professional translator will preserve all of your efforts and really connect you with your new audience. Don’t throw away all of that work by not giving your translated content the attention it deserves as you step into a new global market.

And, most importantly, don’t be like this restaurant.