Skip to content

Is machine translation the future of legal translations?

It’s no secret that translation is now a very different activity from what it was even five years ago. Artificial Intelligence, which has led to a vast improvement in machine translation, has transformed the field. Whereas up to about three years ago hilarious mistranslations were legion on various travel sites, for example, we now see far more convincing and even idiomatic language use.

One inevitable result of this impressive development is that, instead of hiring experienced human translators, more and more professionals are now relying heavily on machine translation in their dealings with clients and customers. However, we need to ask ourselves to what extent we can rely on these tools to communicate professionally and effectively.

Is a machine translation enough for legal translations?

There is absolutely no doubt, of course, that machine translation saves time. It takes care of a lot of the basic problems that no longer need to be addressed. The tools available produce reasonable to good translations of many standard phrases and conventional sentences.

However, as you work with such tools for longer, a few things start to stand out. One is the fact that it does not always penetrate to the essence of the target language. Nor, of course, does it possess consciousness. This is of course the greatest problem faced by AI: a lack of consciousness. Language is a living thing and it is different in the hands of every user or speaker.

Pinning it down in a complex web of rules is still a distant prospect. And that is what gives machine translation the feel of ill-fitting clothes: what it produces is never tailor-made for the task at hand, rather it generates a ‘one size fits all’ result. So the question is then to what extent we can rely on it to provide solid, unambiguous and convincing texts?

Can you use machine translation for long sentences?

One of the things machine translation still hasn’t tackled properly yet is long sentences. Below is an example of a Dutch sentence that came up recently. Adapted slightly to ensure anonymity, it reads:

De maatschap van elk lid wordt jaarlijks, uiterlijk drie maanden na het begin van het lopende boekjaar, ingeschaald door het bestuur voor wat betreft het vaste winstaandeel en voor wat betreft de positie van haar variabele winstaandeel, binnen de grenzen van het remuneratiebeleid.

The translation produced by the tool (in this case DeepL):

Each member’s partnership is scaled annually, no later than three months after the beginning of the current financial year, by the board in terms of its fixed profit share and in terms of the position of its variable profit share, within the limits of the remuneration policy.

For a native Dutch speaker this might well be an acceptable legal translation. After all, it’s an accurate reproduction of the original. But it would baffle an English speaker, who would need to read it a number of times before having even the slightest notion of what it is trying to say. That of course defeats the purpose of translation.

So what is wrong with it?

To start with, the machine clearly has very little idea about how English is used in practice. It simply translates the elements of the sentence, one by one, in the order in which they appear. But one of the most important aspects of English is word order. This is because it is probably the least inflected European language, meaning that the words themselves reveal little about their grammatical function. It is therefore particularly dependent on how sentences are put together.

A few problems with the machine translation in question:

  • use of the passive voice. English uses the passive for special emphasis. It reads more awkwardly than the active voice and there’s no reason for it here
  • expression of time: there are two separate such expressions, both for the same event! Even worse, they appear in the middle of the sentence. This makes a jarring impression on the reader
  • by the time we get to “within the limits” at the end of the sentence, we no longer know what is being done within them. This because the action that applies to them – “each … partnership is scaled annually” – appears as far away as possible, i.e. right at the beginning of the sentence.

A good legal translation communicates its message clearly and intelligibly. The machine translation given above does not. It requires intervention by a human being who instinctively, and from experience, knows how to communicate in English.

What would be a good translation?

Here is an attempt which addresses the problems outlined above:

No later than three months after the start of every financial year, the board allocates each member’s partnership to a pay scale as regards its fixed profit share and variable profit share position within the parameters of the remuneration policy.

Although this sentence contains a fair amount of information, it is now clear who performs which action and in relation to whom, what the applicable rule is (within the parameters…) and when the action occurs. The sentence is ordered in a way that is immediately intelligible to a non-Dutch speaker. This is now lucid communication.

Wondering what legal translation agency Fiducia can do for you? Then contact us to explore the possibilities together.

Dave van den Akker Fiducia

How can we help?