Faster, cheaper, and with knowledge of infinitely more languages than human translators: machine translation (MT) sounds too good to be true. So why isn’t everyone using it? As a team of translators with over a decade of experience, we’re here to explain that there’s more to the story of machine translation than software developers would have you believe.
Keep reading to learn what machine translation is and more detail about the different kinds available. We also share some of the use cases of MT and its perks and drawbacks compared to human translators.
Ready to delve into the emerging world of machine translations?
What is machine translation?
Machine translation is where computer software uses automated processes to translate text from one language to another. Typically, it uses algorithms to understand grammar, syntax and vocabulary to translate the text.
Anyone who’s played around with ChatGPT will know any software that uses machine learning can also be funny, charming, and anything else you need it to be. But only to a limited extent, as we’ll cover later.
It’s important to understand that machine translation is not one specific technique and instead refers to a category of translation techniques:
What are the different types of machine translation?
There are four main types of machine translation:
Rule-based machine translation
Rule-based machine translation (RBMT) uses a set of predefined rules and dictionaries to translate text from one language to another.
Statistical machine translation
Statistical machine translation (SMT) analyses large amounts of bilingual texts to identify patterns and rules in the text. These rules create a statistical algorithm to translate future texts.
Neural machine translation
Neural machine translation (NMT) also uses a method based on analysing large volumes of bilingual texts. However, instead of creating statistics-based rules, it applies deep learning to translate between the two languages from scratch without relying on set rules and patterns.
Hybrid machine translation
Hybrid machine translation (HMT) uses the above methods to harness their strengths and overcome weaknesses. For example, RBMT can better understand grammar, while SMT understands an organisation’s preferred synonyms and phrasing. NMT combines these translations to craft a fluent and natural-sounding text-based output.
The technical side of machine translation can be very complicated, but if you’re in charge of outsourcing a translation project, we’d recommend focusing more on the translation agency’s experience than their specific methods.
How accurate is machine translation?
Different types of machine translation vary in their accuracy. Weglot rated machine translations as 85% accurate, but independent studies are far and few between.
Generally, machine translations struggle in two main areas:
- Often machine translations fail to make the appropriate word choice around vocabulary level, verb choice, and terminology. For example, in English, we say, “Take a photo”. While in German, they say “Mach’ ein foto”, which translates to “make” or “do” a photo. Both make sense, but only one verb sounds correct to the native language speaker.
- The hardest inaccuracy to correct in machine translation is in a text’s creativity and implied meaning. If you’ve written text to persuade a specific audience or communicate a sense of something, a direct translation from an algorithm will often fall short.
Do translators use machine translation?
Yes, more translators are working alongside machine translation than ever before, but it is still far from the norm. However, most translators do use computer-assisted translation (CAT) software.
CAT software such as Phrase, Trados, and MemoQ increases the quality and consistency of translations. This type of software uses a translation “memory” created from previous human translations, so they cannot be technically classed as machine translations.
Translators who use machine translation do so in one of two ways: as a technician rather than a translator or using a method known as Machine Translation Post Editing (MTPE). MTPE is where the agency generates a machine translation and then has a human translator edit and correct any mistakes.
Is MTPE the answer to fast and affordable translations?
Post-editing can be a way to access cheaper translations and results in more accurate translations than pure machine translations. But, our founder Lucy explains why we choose not to use MTPE at LEaF:
Here at LEaF, we don’t offer MTPE for two reasons:
One, the quality of MTPE translations is much lower than that of pure human translations. This is because machine translations stick more closely to the source text than a translator would and translators tend to lose their creative flair when editing a machine translation.
Secondly, we’ve found that producing high-quality MTPE that is equivalent to a human translation often takes the same amount of time as translating the text in the first place. Therefore asking editors to take a lower fee for MTPE (as its main draw is cost-reduction) isn’t something that sits right with us.
How machine translation is affecting the translation process
Machine translation has made the translation process more accessible for a plethora of industries by making translations cheaper and faster.
However, machine translation’s inaccuracies have shifted some human translators’ roles from producing translations to reviewing, editing, and improving machine-generated translations. For this reason, you can still find many agencies that stick to computer-assisted translation software rather than utilise MT.
Can machine translation replace human translation?
It would be shortsighted of us to say machine translation could never fully replace humans in the future. However, for the foreseeable future, machine translation is just another tool for human translators to use to speed up the translation of large volumes of text.
If you’re wondering when can unsupervised machine translation work? For now, human-free translations are only appropriate when the implications of translation mistakes are low and there is no budget for a more reliable service.
Why can humans be better translators than machines?
At this point in time, humans are better translators than machines for a range of reasons:
- Well-trained professionals write better-quality content in terms of creativity and readability.
- Linguists and translators consistently craft more accurately translated content, especially when translating colloquialisms, metaphors, and idioms.
- Humans can gain peer-to-peer input to shift the tone of a piece to match the social norms of the local audience.
- Human translators can often be just as affordable as machine translations, or at times cheaper than complex hybrid machine translation.
- Finally, there’s less need to include additional proofreading stages as the text is accurate in the first place.
Our CEO and senior translator, Lucy, on why humans outmatch machine translations and MTPE:
“Good professional translators use quite a lot of creativity in the translation process – we don’t just swap one word for another. Instead, we are constantly playing with the words and phrasing to create a text that works in the target language and doesn’t sound translated.
“Using a machine translation engine to translate a text and then adapting it zaps your ability to think and translate freely – it is much harder to remove the clunkiness of a machine translation than it is to craft a fluent sounding original translation.”
Should I use machine translation?
So, with the facts presented on machine translation, the question still stands; should you use machine translation?
If the nuances of your text aren’t necessary and your budget is tight, then machine translation is a serious option to consider. Any agency offering prices of less than £0.12 per word is likely not providing a standard translation service, but rather using MTPE. But it’s worth bearing in mind that MTPE can also often cost more than this.
However, you should avoid pure machine translations if mistakes and misunderstandings in your text can:
- Lead to harm, such as mistakes in medical or assembly instructions.
- Waste consumers’ or employees’ time or money, for example, when detailing transport requirements.
- Cause a loss of revenue by wrongly translating tone of voice and USPs in marketing text.
- Lead to legal implications from unclear terms and conditions or disclaimers.
Choosing your translation method: What to do next
There’s also a world of different types of translations and methods that don’t involve machine translations.
At LEaF, we don’t use machine translation. Our approach focuses on creative translations and localisation, which require a deep understanding of the nuances of both languages involved. We understand that machine translations can be tempting due to their cost-effectiveness and speed, but the risks of errors, inaccuracies, and cultural mistakes can outweigh the benefits.
If you’re uncertain about which direction to take, or you want to work with human translatiors but you’re worried about the cost, let’s talk. We can help you build a business case for your team and provide you with a cost-effective solution that meets your needs. Browse our website for more information on our human translations, or email us at hello@leaftranslations.com.