When Should You Use Machine Translation?
With the rising ubiquity of artificial intelligence, localization managers and marketing and communications professionals may be wondering when and how they can make the best use of machine translation, and when it’s best to stick to a human translator, whether for post-editing or from scratch. In this article, we’ll explain what machine translation is and the different ways it can be used. Keep reading to learn more.
How Good Is Machine Translation?
Machine translation (MT) refers to the use of artificial intelligence to automatically translate a text. The most popular machine translation program is Google Translate, but there are many others, including versions customized for specific domains and client accounts.
MT software has been making incredible (if slightly unnerving!) progress in recent years thanks to advances in machine learning and natural language processing. Gone are the days of Google translating word-by-word with zero regard for context! With the help of artificial intelligence, neural MT is now able to produce a relatively accurate and understandable translation of a wide variety of text types and languages.
For example, to show you what we mean by accurate and understandable, here’s a side-by-side comparison of an excerpt from the first chapter of Albert Camus’s L’étranger, translated by DeepL vs by literary translator Matthew Ward:
|Source text||Unedited MT||Human translator|
|J’ai pris l’autobus à deux heures. Il faisait très chaud. J’ai mangé au restaurant, chez Céleste, comme d’habitude. Ils avaient tous beaucoup de peine pour moi et Céleste m’a dit : « On n’a qu’une mère. » Quand je suis parti, ils m’ont accompagné à la porte. J’étais un peu étourdi parce qu’il a fallu que je monte chez Emmanuel pour lui emprunter une cravate noire et un brassard. Il a perdu son oncle, il y a quelques mois.||I took the bus at two o’clock. It was very hot. I ate at Celeste’s restaurant, as usual. They all felt very sorry for me and Celeste said, “We only have one mother.” When I left, they walked me to the door. I was a little dizzy because I had to go up to Emmanuel’s house to borrow a black tie and an armband. He lost his uncle a few months ago.||I caught the two o’clock bus. It was very hot. I ate at the restaurant, at Celeste’s, as usual. Everybody felt very sorry for me, and Celeste said, “You only have one mother.” When I left, they walked me to the door. I
was a little distracted because I still had to go up to
Emmanuel’s place to borrow a black tie and an arm band.
He lost his uncle a few months back.
As you can see, Matthew Ward’s version sounds more natural and has a more pleasing style, but the machine translation is good enough if you just want to understand what the source text says. Plus, it would only take a few small edits to make it sound more natural.
Does that mean human translators like those on our team are about to be replaced? The short answer is no. The long answer is that, for many types of content, it is no longer necessary to have a professional translate the whole text from scratch. Instead, the language professional will have other important roles to play in the coming years, including:
- Post-editor, making the necessary changes to ensure the machine translated text is accurate and fluid
- Advisor, helping clients decide where to use MT and which MT software is the best for their needs, and to set up their workflow
- Quality analyst, assessing the quality of MT output
- Transcreator and writer, adapting creative or wordplay-heavy texts that are beyond MT’s ability
- Language data specialist, leveraging their linguistic knowledge to help train MT software
The Different Types of Machine Translation Service
As you can see, there are many different ways to use machine translation and different roles that a human can play in that process. In the table below, we’ve summarized the main types of machine translation service that OXO offers, as well as the pros and cons of each, to help you choose the best one for your use case.
|Human translation||One of our professional translators who is a native speaker of the target language and who has specialized knowledge of your industry translates your text from scratch (while referencing previous translations and other resources). Next one of our editors further improves the text and ensures its fluidity.||A good translator will create a text that sounds fluid and engaging, and that you would never guess was a translation. Rather than translate literally, they will adapt the tone and content to the target audience.||Slow and costly. Quality can vary widely depending on the skill and experience of the translator and editor. Productivity, however, tends to range between 1,500 to 2,500 words/day||– Creative marketing copy|
|Full MTPE (machine translation with heavy post editing)||After running your text through a machine translation engine, an editor carefully edits the output line by line to ensure accuracy, terminology and style consistency.||Essentially delivers a human-quality translation in less time than a translator who has to type everything from scratch (and therefore lower cost).||Not helpful for texts that are very creative, use a lot of slang, or are badly written. You rely on a single human, as opposed to the previous choice.||– Blog articles
– Press releases
(machine translation with light post editing)
|An editor runs your text through a machine translation engine and then quickly edits the output to check for mistakes or inaccuracies.||Good when you need to save time/money and the translation is for informational purposes (i.e. needs to be readable but not necessarily eloquent).||The text will make sense but may sound like a translation, and the editor will not follow your glossary or style guide. Not appropriate in situations where an error could negatively impact health & safety or your brand image, or have legal repercussions.||– Internal documentation and communications that are not a priority|
|Raw MT||We run your content through MT with no human interventions. See below to learn about the main types of MT engine.||Almost instantaneous. Very inexpensive.||Same risks as light MTPE but even more so.||– Customer reviews and discussion forums
– Monitoring what customers around the world are saying about your product/brand in the media
Different Types of Machine Translation Engine
On top of choosing the right level of human intervention for your needs, you also need to choose the right machine translation engine, as they are not all built the same or designed for the same purpose. Speaking very broadly, there are three main categories of MT engine: custom-trained engines, subscription-based engines, and free engines. Here’s a brief description of each.
|Type of MT engine||Description||Pros||Cons|
|Custom engine||Machine translation engine custom trained for your company’s needs. A language consultant trains Google AutoML or a similar service with data specific to your domain or from your company’s previously translated content to create an MT model that is fully customized to you. The output is most likely to have better results.||After the up-front investment in engine training and data acquisition or collection, it is much cheaper and faster than MTPE, potentially delivering the same level of quality, or better.
|Requires a large volume of previous high-quality translations to train the AI. For smaller or newer businesses, it is impossible to collect such data from their own content, but investing in domain-specific corpora is an alternative.|
|Paid subscription of out-of-the-box solution||Either you buy your own subscription or you pay an LSP a small fee to use theirs. Your text is run through the MT to get an instant translation.
|Almost instantaneous. Very cheap. Includes data confidentiality. Perfect when you just want a basic understanding of what the source text says.
|Likely includes some mistakes and will sound like a translation.
Some engines are better than others.
|Free online version||You run your text through one of the many free machine translation solutions available on the internet.
|Almost instantaneous. Free. Perfect when you just want a basic understanding of what the source text says.
|Not suitable for confidential material (no data security).
Likely includes some mistakes and will sound like a translation.
Some engines are better than others.
In short, there are many possible ways to use machine translation to save on costs and time while still producing a translation that is of an appropriate quality for your purposes. In almost all of these scenarios, seasoned language professionals like those on OXO’s team still have an important role to play when it comes to selecting and implementing an effective MT solution. Thinking of using MT to boost your efficiency? Talk to us today to find the right strategy for you.