The pitfalls of machine translation for large-scale projects

Whether translators like it or not, machine translation is here to stay, and it is becoming a popular solution for companies with large, repetitive translation projects that they will keep adding documents to.

A machine translation engine automatically translates sentences into another language, while a translator or editor may post-edit the text to make sure that it is translated correctly and makes sense. Some engines “learn” from previous translations that the translator has made, and can incorporate those solutions into similar sentences that it finds in future documents.

In theory, this speeds up the translation process (the translator’s role is more editing than actual translating, since they are working with the sentence that the MT has made) and generally reduces the company’s translation costs.

But sometimes, this is all too good to be true. Although this method can speed up large translation projects, if too many corners are cut then it can lead to subpar work and some embarrassing mistakes once the translation is published.

To illustrate this, I have taken some examples from a large project that I recently worked on, where the end client was an online clothing retailer wishing to create an English version of their website and product listings. Because of the scale of the project, there were a team of several translators all working on different sections at the same time – another move that saves time but runs the risk of having “too many cooks”.

Consistent terminology

The translation agency acting on behalf of the end client asked the translators to help build a glossary so that we could be consistent in translating some of the terms. This is a good idea when you know that the same terms are going to arise several times in your project – especially when the company was looking for a specific “luxury” tone to be used in their texts. However, there were a couple of factors and words that the end client did not think about.

The first is that one word in French can sometimes be translated several different ways in English. For example, in a clothing context, the word froncé in French could be translated as pleated, gathered, ruched or shirred in English – and you may not know which one to use unless you have a picture of the garment to help guide your choice. If you only give the machine translation engine one possible translation, then it will use it every time.

Pleated, ruched or gathered?

Secondly, other words were also being translated in several different ways by the other translators. For example, a semelle cranté on a shoe could be translated as notched sole or lug sole, and a poche passepoilée could be a welt pocket or a piped pocket, depending on where the translator finds their terminology.

These two examples also highlight the risk of using translators who are not specialised in the field – they will not be sure if there is a difference between all these terms, and may risk feeding inaccurate information into the machine translation.

It was up to me to flag these to the client to ensure that the correct terminology is being used consistently across the board – but by then, they had already started publishing content on their website.

Missed opportunities

The machine translation will translate similar sentences the same way each time, which the translator may mark as correct because…technically, it’s not wrong. But when one word in French can be translated several ways in English, you are missing out on a chance to make the text more colourful, unique and engaging by not giving your translators the time to transcreate it from scratch. As a result, the texts across the website end up sounding very samey – which is a no-no from a marketing standpoint.

This was something that I raised to the translation agency too, but they felt limited by the fact that the client did not want to pay extra or give us extra time to have these texts rewritten – so there wasn’t a lot we could do about it. The translators were forced to stay quite close to the machine-translated text due to the short deadlines.

Lack of context

This particular company also asked the translators to translate a series of single words and phrases that they knew would come up throughout the website, such as taille (size, waist) and gilet (jacket, cardigan, waistcoat, gilet). As you can see, once again, these words are very hard to translate in isolation, and may change meaning depending on the context, so giving a translator a list of recurring words is not always helpful to either party.

As a result, certain parts of the website – such as photo captions – ended up containing out-of-place words, and looking as messy as if the company had not had any humans involved at all.

Conclusion

Due to these issues and despite our warnings, when the company’s website went live, it did…not look great. There were many translation errors that could have been avoided if the end client had been clearer about the purpose of each text, if they had not isolated so many words and sentences to be translated on their own without context, and possibly if fewer translators had been involved.

While machine translation and post-editing is a time- and money-saver for companies with large-scale projects, they also need to be prepared to provide the translator with a lot of context rather than assuming a one-size-fits-all approach. With screenshots, notes, and just a general willingness to collaborate, a lot of these mistakes could have been caught before the website went live. Instead of saving money, the client had to pay extra to have parts redone and also risked reputational damage to a brand that they wanted to position as high-end.

If you can find a trusted human translator with the subject knowledge you need for your project, it could prove to be an even better investment!

About me | My services | Contact

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

This site uses Akismet to reduce spam. Learn how your comment data is processed.