Machine Translation: Friend or Foe?

Concept of artificial intelligence in use as a translator,french to german to english. 3d illustration.

Dialogue’s latest webinar: Can We Trust Machine Translation?

We kicked off December with a webinar addressing one of the biggest questions on the tips of so many multilingual tongues: can we trust machine translation? The simple answer: yes, to an extent, as long as you choose the content carefully and use it in an informed way. We were joined by our AI translation partner Memsource to help our clients make the right choices to get the most out of machine translation.

A big thanks to all of you who joined us to learn more and to challenge us with your questions and concerns.

Why not use Google Translate for AI translation?

Well, you absolutely can, as well as other popular generic machine translation engines like Microsoft and DeepL for content including some product descriptions and bulky content you need to understand the gist of quickly. Advancements in neural technology have vastly improved machine translation flow and accuracy in recent years. The big BUT is: avoid creative content as word play, ad slogans and locally sensitive content will be lost in AI – and mistakes could be costly. For anything in between (e.g. web FAQs and training materials) we recommend you combine raw machine translation with human post-editing.

What are the best machine translation engines for my content?

Our advice: choose your machine translation engine carefully. Output quality will vary hugely depending on language combination and subject matter. Memsource research reveals that Google Translate fares well for Arabic and Turkish, but you may want to think twice about using it for Chinese! DeepL pulled in great results for medical translation but fell down in the entertainment sector. Selection isn’t a straightforward process but there are tools (e.g. Memsource Autoselect) that can analyse language combination and subject matter to choose the strongest engine for it.

Can I use raw machine translation, or does it need editing? Can I trust edited machine translation?

It really depends on 4 factors: engine selected, language combination, subject matter, and intended use. If you need just to get the gist of a text or if it’s a Spanish technical translation, raw machine translation can work. However, since the quality of raw machine translation output varies hugely it’s normally advisable to get it edited by a professional translator. Ensure that your machine translation is checked carefully against the source text, your instructions and glossaries for accuracy, consistency and, most importantly, coherence. The amount of editing required will depend on the machine translation quality but the human touch will definitely ensure you can sleep at night!

Source: Memsource

Can I customise machine translation engines to my own preferences and terminology?

Absolutely. Engines such as New TranX, Nuitrans and Language Weaver can be trained to your preferences. You’ll need a decently-sized, good-quality translation memory (min 50k segments or ca. 600k words) along with a termbase to get the engine started. It makes sense to do this if you have large amounts of ongoing translation annually. There’s an initial set-up cost per language combination and you’ll need to set aside time to clean, update and train the data on an ongoing basis. But this option can certainly be worth the effort in the long term.

How Dialogue can help with your machine translation needs

Interested to find out more? If you’d like advice on whether or how to use machine translation for your content, contact our Translation Manager, Dan Thomas: [email protected] or call + 44 (0) 1793 513 001.

Share this post

Share on facebook
Share on google
Share on twitter
Share on linkedin
Share on pinterest
Share on print
Share on email