← Show all

Beginner’s Guide To AI In Translation

When you think of machine translation, you probably think of dodgy sentences regurgitated by Google Translate: ‘Am I make self understood to you?’ No Google, you’re not. Given the fluid, evolving, and often artistic nature of human language, how could artificial intelligence ever possibly replicate the translation quality of a human?

Beginner’s Guide To AI In Translation

Technological developments and advances over the past two decades have meant that machine translation (MT) is more accurate now than ever before – and will continue to improve. Companies across the world are now taking their business global at high speed and low cost by making use of raw machine translation and post-editing (MTPE) by human linguists.

Forget what you think you know about AI in translation: here’s our no-nonsense guide to machine translation for business.

The Technical Bit

There are different models of machine translation, including rule-based machine translation and statistical machine translation, but it is neural machine translation that we will focus on in this article.

Neural machine translation is the application of artificial intelligence to convert the text from one natural language into another. Neural machine translation makes use of machine learning and neural networks to predict the most accurate translation outcome.

But… what are neural networks? Neural networks are a subset of machine learning and are at the heart of deep learning algorithms. Their name and structure are inspired by the human brain, mimicking the way that biological neurons signal to one another.

In simple terms, machines are learning and the deep learning algorithm used for the machine is structured as if it was a human brain.

In the case of translations, neural networks learn from human translators, so its output is more sensitive to context and nuances – meaning it is ultimately more accurate.

Doesn’t Machine Translation Mean Lower Quality?

If you’d tried to translate your company website using only Google Translate back in 2006 when it first launched then yes, it might have been nearly illegible, as Google was still using statistical machine translation back then. Raw machine translation (such as Google Translate) is now much more accurate, thanks to the shift towards neural machine translation which uses artificial intelligence.

There are no hard and fast statistics for MT quality or accuracy because so much depends on how technical, specialized, or colloquial the source text is, and on the language pair (i.e. you’re likely to receive a better translation for English into Spanish than English into a minority language such as Scots Gaelic). However, as more time passes, more data will be fed into these translation engines, and they will become increasingly more accurate.

The Benefits of Neural Machine Translation

Artificial intelligence in any field is attractive to businesses because it’s just like having an extra employee who can work 24/7 and never gets tired or asks for a break; the same applies to the AI used in neural machine translation.

NMT can also process more words than a person can: a professional human translator translates around 2,500 words per day on average, while DeepL can translate up to one million words per second.

In addition to increased turnaround times, artificial intelligence offers a much cheaper alternative to professional human translation services.

How Do Businesses Make Use of Neural Machine Translation?

Any business planning to operate internationally will inevitably need to make use of language services. While machine translation quality has improved leaps and bounds in recent years, we wouldn’t yet recommend using raw MT on every aspect of your communications.

Many companies are making good use of machine translation on user-generated content, e.g. reviews, comments and feedback, product descriptions, etc. The holiday rental app, AirBnB, is making use of MT for translating accommodation descriptions and reviews left by users.

For higher-quality translations at scale, many businesses follow the MTPE model: machine translation plus post-editing by a human linguist. Having a human translator review the content after machine translation has already been applied means the final translation is of higher quality, but the turnaround times are still kept to a minimum.

Getting Started with NMT

At Jonckers, we have developed our own AI-powered translation platform, WordsOnline, which is ISO18587 certified (‘Post-Editing of Machine Translation Output’). We combine machine translation with our pool of expertly trained linguists to provide the speed and agility of AI with the quality and authenticity of the human eye.

Try out your first AI-powered translation project today using our Translate Now instant ordering service. Just upload the file(s) you need translating, choose from 40 languages, select your service level – including raw machine translation at €0.005 per word – and pay online.

We also offer a translation subscription service through WordsOnline. You pay a fixed price monthly or yearly based on the volume of translations you require and receive translation ‘tokens’ to be redeemed throughout each month. Your tokens can be redeemed on machine translation post-editing (MTPE) services or desktop publishing (DTP) services, for highly designed content. You’ll also receive free, unlimited machine translation tokens.

Get in touch with Jonckers to find out more about how AI can power your business’ global expansion.

Change the way you translate!

Upload your documents to WordsOnline and pay online for a fast, simple service. 10% discount on your first order

We will help you transition to the new world of continuous localization!

This site is registered on wpml.org as a development site.