MEINRAD's Blog

Which languages are suitable for machine translation?

Written by Meinrad Reiterer | 22 November 2022

Things are moving fast in the world of machine translation – the results produced by MT tools are getter better and better, and the range of languages offered is growing all the time. But be careful: just because a text can be machine translated from one particular language to another, that doesn’t necessarily mean the results will be good. Especially for professional translation services.

Translating a document instantly into as many languages as you like at the touch of a button used to be a mere pipe dream – but now, thanks to machine translation, it’s an everyday reality for many businesses around the world. Google Translate offers translations into over 100 languages, and MT services are constantly being enhanced in terms of both their functionality and the range of languages they cover. So why wouldn’t you use MT to produce translations in these languages in future? Unfortunately, what sounds tempting in theory can often turn out to be not such a good idea in practice. The quality of the MT output isn’t (yet) good enough in some languages, so it’s only suitable for internal use at most. The main languages where you can be confident of very good results are the major European languages – provided you use professional MT services in combination with full post-editing, you can reduce what you pay for translations of technical documentation (for example) and still get the high-quality texts you need. So machine translation shouldn’t be used “blind”, and only with the appropriate specialist expertise. That’s where experienced language service providers can help.

The more widely spoken, the better

One reason why some languages don’t produce good results with machine translation is simply that it’s harder for the MT engine to “learn” them. Essentially, the more widely spoken a language is, the better the various MT tools will “speak” it as there’s more reference material which can be used to train the engine. The results also vary depending on the subject matter of the translation: it’s not uncommon for an engine to produce outstanding results in one specialist area but to be less useful in another. And the source language is another factor, as many MT tools work – at least in the background – on the basis of English, meaning that texts in another language are first translated into English and then into the target language. So there are various factors involved, which makes it difficult to give a simple answer as to whether certain engines are suitable for certain languages.

Post-editing to the rescue?

Of course there’s also post-editing, the process of editing the machine translation output by a human translator. This ensures that the quality of a machine translation is comparable with a human translation and that it can be used for external purposes. But if the time required for post-editing is about the same as for a human translation, or maybe even longer (from having to read and delete/rewrite the MT output), you lose one of the key commercial benefits of machine translation. Just because a machine “spits out” a translation doesn’t yet mean the “Machine translation and full post-editing” workflow is automatically viable, let alone beneficial.

So how can you benefit from machine translation?

It isn’t a good idea to pre-translate a document yourself with a machine translation engine and then send it to a translation agency for post-editing. Why? For one thing, most businesses will use a free MT engine to do this, which often won’t give them the privacy they need for their data – we strongly recommend using fee-based Pro versions of machine translation engines instead. For another, a good translation agency will keep up to date with the rapid developments in the machine translation sector and will test the various MT engines, so they’ll know which engine is most suitable for your documents and how it can be optimized for a particular specialist area. The agency can use their experience to give you useful advice, and they’ll be honest and open about whether your texts are suitable for MT in the first place, and if so, which languages can be machine translated. So even if none of the major MT engines are a workable option, the agency can help you find a solution for the specific situation.

Using trainable engines

One potential solution for languages which are generally less well-suited to machine translation is to use a trainable MT engine. But be aware that you’ll need huge amounts of previous translations and reference documents, as a (small) translation memory isn’t enough to train an engine to produce good results. For small translation projects, it usually won’t be worth the effort. But if you need to get large amounts of text translated, this initial investment can really pay off. And for many specialist areas there are also MT engines designed to produce high-quality results in these specific fields.

So our advice is to talk to your translation agency. They can guide you through the world of machine translation and help you find a way to benefit from it without reducing the quality of your translations.

 

 

Main image: © Adobe Stock