Machine Translation; Good or Bad?

Translation is a topic which can stir emotions! Especially for retailers selling internationally.  Historically the only way to get good translations was with a human who speaks both languages and has experience, ideally qualifications, in translating from one language to another.  With the advent of online translation engines many people jumped at using these platforms only to find that the quality of translations ranged from poor to embarrassing. This reinforced a belief that only humans can do a good job at translations, a belief that is still widespread today.

However, even human translation can be an issue if you look for problems. On a number of occasions I’ve sat with international users of systems and listened to them argue among themselves about whether this or that translation is good or bad. One thinks its’ bad and the other thinks its good.  Even though it’s been written by a professional translator with all the experience and necessary qualifications.  

The truth is, translation can be subjective, especially where there are not a direct equivalents of words. It’s important to understand this as looking to deliver a perfect translation is an un-achievable goal even with humans. But only if you know it’s been translated! Perfect translation isn’t what it’s all about. What is most important for retail translation is that it is acceptable in the context and technically correct so that the consumer understands what they are buying. 

With advances in technology the quality of machine translation, driven by artificial intelligence techniques like machine learning, has advanced to be good enough for most situations. Especially if you find the appropriate tool and use it correctly. 

There are now translation tools and companies that specialise in retail machine translation. Full disclosure of course, at Webinterpret, the company where I work we provide machine translation as one of our services.

Translation engines should be built to take into account the context of the translation. By context I mean the engine should understand that it is dealing with products to be sold online. Ecommerce products are always made up of the same set of parameters. They have a title, a description, price and attributes, among other things.  

Translation engines that are focused on retail ecommerce can also understand the context of a product based on it’s category.  This gives specialised translation engines an extra way to understand how to translate For example, a tablet in a computer hardware category should be translated as tableta in Spanish, but a tablet in a pharmaceutical category should be translated as píldora.  A generic engine can not understand this which is why a specific retail translation engine can deliver better results than, say, Google Translate even though Google have such massive resources available to them. 

A good retail translation engine will break the translation down to focus on these areas in order of importance.  In eCommerce the order of importance for product translation is usually as follows: 

  1. Title: this should be a nearly perfect translation and built for clarity and SEO. An experienced engine will have had most titles through before and ideally built up a human checked translation memory to ensure quality. 
  2. Product Attributes: the factual information about a product such as what its made of, what colour it is, what it contains, size, etc… This is key in ensuring that a customer knows exactly what they are buying. Again a good engine will have seen the vast majority of these attributes before. 
  3. Short Description: buyers will often see the short description under the title and will read this to get a quick idea of what the product is if it’s not obvious.  It’s harder to machine translate longer sentences so be aware of these translations.
  4. Long Description: Usually contains marketing material and brand information alongside product information.   This is useful but not hugely important in ensuring the buyer knows what they are getting and frankly very difficult to translate automatically.  

What is machine translation good and bad at?

Good

Luckily machine translation can be good at the product segments that are most important to get right. The fewer words needed to translate and the clearer the context the better a machine can be. 

Titles are usually repeated many times over time if the translation engine is experienced enough, i.e. if it has translated a lot of products, and they are short and concise. This allows an engine to build up experience on short, specific and descriptive sentences that are often checked by a human translator to ensure quality. 

Product attributes are also easy to put into a context and usually are limited to one or a few words, so they are good for engines. 

As you start to get into longer and longer descriptions and sentences the worse an engine will perform, so short descriptions can be OK, but long descriptions are usually poor. Fortunately with the surrounding good translations these can be forgiven and it’s always a good idea to inform your users that they are reading machine translation.

Bad

Many ecommerce sites will also contain pages of information that are sales focused and use flowery sentences, such as branding pages and Sales pages.  While these can help your domestic sales if they are not localised correctly they will hurt your image not help it. Translation engines are not yet good at these types of translations. In fact many human translations will not have the knowledge or skill to manage this well given that there are some specific words or phrases that have subtly different meanings in some languages. 

Automatic translation of your seasonal marketing pages isn’t a great idea given that in many countries the holiday seasons are quite different.  

This of course gives you both a challenge having to keep up with other seasons but also an opportunity that by focusing on holiday’s cross border it’s usually high season for your products somewhere!  For this sort of translation, if you deem it necessary, many translation companies will provide a service called Transcreation. This is where experience and qualified translators take your local content and convert it into something relevant within the country for which you’re translating. 

Transcreation can be expensive so I only advise this if you are focused on marketing directly in another country you’ve already tested and know you have a good market. I certainly wouldn’t recommend it for all languages or without checking the potential. 

However, once you decide to push into a market it is worth investing on at least the basics of your brand. Some large companies have been stung by not understanding the local lingo. Mitsubishi for example entered the Spanish market with a car called the Pajero and only started looking at the meaning when they sold absolutely no cars. They soon understood why when they realised Pajero means “tosser” in Spanish.

Other companies like McDonalds looked at the sensitivity before entering the market and for example changed their tag line from “I’m Loving It” to “I just like it” in China because they found that some Chinese find using the word “love” in marketing offensive. 

Finally, machine translation will not work well for your legal content. In today’s environment it’s more important than ever to get your T&Cs and Privacy Policies correct and just translating them with by machine into multiple languages isn’t going to help you here.  Again a standard translator isn’t going to be your best bet either and there are specialist legal translators you can use. I advise covering all countries in one set of pages so you only have to translate on page instead of making variations for each language. 

About Mark Ellis

Mark is the VP for Growth and Partnerships at Webinterpret, a leading Cross Border Technology company. Before that Mark was leading the operations for eBay's European cross border program. He has over 20 years of eCommerce industry experience, guiding and delivering strategic change in retailers, working with companies such as Dyson, Regatta, Boots and Arcadia Group, leading multi-million dollar programs for industry giants like Dunnhumby and working with small businesses delivering innovative retail solutions.

Check Also

How to Analyse your Cross Border eCommerce Performance

If you’ve read our guide from start to finish you will know by now that …

Leave a Reply

Your email address will not be published. Required fields are marked *