We’ve all seen hilarious mistranslations caused by machine translation (MT).
Remember the out of office road sign? Or the excessive order of eggs at the PyeongChang games? That’s not to mention the age-old giggle-inducing menu mistranslations you find while eating out on holiday (remember when we could do that - good times!).
But joking aside, despite its obvious shortcomings and inaccuracies, can machine translation ever be used? Should it ever be trusted? Or is translation a task that should only ever be carried out by professional (human) translators? And if machine translation is automated translation, then what is the difference between that and computer aided translation? Let’s take a look at all these questions.
But before we go on, I’ll start with a warning - free machine translation can be helpful for getting a rough understanding, but professional human translation is essential for critical business needs. Don’t worry, we’ll explain why.
Professional human translation and machine translation — the differences
Although some may be obvious, it’s important that we first have a look at the differences between human translation and machine translation. Roughly speaking:
Professional human translation
The obvious one - translated from a source language to a target language by a real person.
Translated by a machine from one natural language to another, with no human involvement.
Translators are subject area specialists with expertise gained through years of translation work, working in industry, translation-related qualification, etc.
Some make use of AI and can be customised for specific industries.
Can be members of professional associations, showing commitment to the industry and professionalism.
Machine translation can be used in its raw (unedited) form or it can be post-edited (by a human translator) before use.
Machine translation was first pioneered in the 1950s and has undergone many changes over the years. Although Google Translate is probably the most well-known, there are lots of other providers, such as Microsoft Translator, Systran and DeepL.
As well as different providers, the actual engines themselves vary and have evolved over the years, with the 4 current main types being:
- Rule-based Machine Translation (RBMT)
- Statistical Machine Translation (SMT)
- Hybrid Machine Translation, and
- Neural Machine Translation (NMT)
That’s an awful lot of abbreviations; what do they mean?
Rule-based Machine Translation uses built-in linguistic rules and bilingual glossaries and dictionaries. The grammatical structure of the source language is transferred, using these rules, into the target language by analysing the input source sentences.
Statistical Machine Translation has no knowledge of language or grammatical rules, and instead uses computer algorithms to analyse huge amounts of monolingual and bilingual content (known as a ‘corpus’ or ‘corpora’ in plural form) for each language pair to create statistical translation models - basically using statistical likelihood (‘probabilistic mathematical theory’) to convert source language sentences into the target language.
Hybrid Machine Translation, as its name suggests, combines the use of multiple machine translation approaches in one system.
Neural Machine Translation involves the use of neural network models (multiple processing devices modelled on the brain) and a deep-learning based approach to predict the likelihood of a set of words in sequence.
Different types of MT = different translations (and errors)
Why is it important to know about the different types of MT, you might ask? Well, if you do decide to go down this route, it’s very important to understand how each works, because each has its own peculiarities, which produce differing results and are characterised by different sorts of errors.
The higher level of fluency in an NMT translation, for example, can mask inaccuracies; because it ‘sounds’ fluent, the false appearance of accuracy is given. Meanwhile translations produced using SMT can contain errors due to it not factoring in context and this type is also very susceptible to errors of negation (a missing ‘not’ can make all the difference).
It is possible to train machine translation engines using data and glossaries, and if done with only texts in certain domains, the engine becomes more subject-oriented (for example engineering) or even client-oriented. The results still won’t be anywhere near perfect, but should be better than running it straight through a free online MT engine.
Human element is crucial for many projects
The main thing that must be clarified is that there are many times and industries where only human translations must be used. If your document contains confidential information, then it goes without saying that it must be a human translator. Similarly for legal documentation, where absolute accuracy is needed, a human translator is the only way to guarantee such high quality.
The key benefit of a professional translator is the high quality. This comes from the linguist having the correct subject area knowledge (engineering, legal, patent, etc) and document type (marketing, e-learning, scholarly article, etc) for the source documentation. And linguists are generally natural perfectionists, not stopping until the exact terminology is found and the source conveyed in the most appropriate tone. Whereas with machine translation, speed and cost are the real gains. Once the source document is fed into the engine, the translated result will be churned out instantly. You can also talk to a translator to discuss tricky terminology.
And, the obvious one, although inaccuracies can occur both with machine translation and with human translators – we’re all ‘human’ after all – I think it’s fair to say that machine translation will be much more afflicted by errors.
Where is professional human translation best used?
If you need a high-quality translation, where accuracy is a key concern, or want to make a good impression for a new target market, then human translation must be used. Translators are highly-skilled professionals and will be aware of and respectful to cultural nuances and sensitivities, something which is not possible by a machine (as this recent article demonstrates). Professional translators can also take on board any brand and tone of voice guidelines.
Having deep knowledge of the particular industry in both the source and target languages and cultures means that a translator can also highlight any potential issues with the source content. All these things are just not possible with machine translation. Whether it’s your website, marketing materials, terms and conditions (the list goes on), professional human translation is the way forward.
In addition to our core professional human translation services, we now offer a Post-Edited Machine Translation (PEMT) service for certain projects. If you would like more information about our PEMT service or would like us to assess whether your content would be suitable for this process, please contact us via email email@example.com or use our quick quote form.
Where is machine translation best used?
So, when is machine translation (such as Google Translate) useful? It can be helpful for content that is high volume and needed fast, but where accuracy is of low importance. Comments, social media posts, reviews, instant chat facilities and website help sections are all good candidates for machine translation.
Machine translation is also useful for so-called ‘gist translation’. You might have received an email in another language and you need to understand a rough gist first. Machine translation can come in handy in this case, allowing you to then assess whether a professional human translation is needed for better accuracy.
What about a mixture of human and machine translation?
Human translation can be combined with machine translation. This is known as post-edited machine translation (PEMT, also known as MTPE). A file is translated first using a machine translation system to create the raw output. This output is then reviewed and edited by a professional human translator.
It is quite a different skill to professional translation, so you may find that not every translator is able, or even willing, to post-edit. Clear guidelines are necessary for post-editing to work successfully, since under/over-editing can be one of the main issues.
Machine translation — not to be confused with computer aided translation (CAT)
Although machine translation sounds like a similar technology to computer aided translation (CAT) tools, CAT tools are in fact a type of translation environment technology used by translation companies and translators.
Like a graphic designer uses specialist software to create banners and logos or a photographer uses specialist editing software on photographs, professional translators use CAT tools to help them translate. Although, machine translation can be plugged into these tools, if the translator so wishes.
The source document will be opened in the tool and a Translation Memory (or TM) attached to it. A translation memory is essentially a bilingual database containing translation units (source segments alongside their translations) for a particular client or subject area. These units can be sentences, phrases or even single words.
As the new document is translated by the user, the translation memory is searched automatically to see if any of the sentences in the source file have a ‘match’ in the translation memory. Matches may be exactly the same or so-called fuzzy matches may be found – where part of the sentence has been previously translated. This allows the user to draw upon and leverage their previous work, meaning more consistent translations and a higher level of efficiency.
To put it simply, the CAT tool remembers and suggests previous translations – that may well be stored in the translator’s brain – but retrieved much faster and easier!
In summary, while MT does have some uses when the aim is speed and/or cost rather than accuracy, a professional, human translation is a guarantee of quality and a necessity when accuracy is crucial.
If you would like more information about the interesting topic, feel free to get in touch, or head over to find me on Twitter or LinkedIn to share your thoughts.
Categories: Industry News