What is Machine Translation, and do you need it?

No doubt technology is quickly reshaping the way we perceive the world and go about our daily lives. We’re all having to adapt to an ever-changing world, otherwise we risk being left behind, just like technology and my parents – sorry Mum.

There have been so many advances in our industry as well. As a seasoned translator I remember the old days behind my typewriter armed with bottles of white-out and tonnes of paper, and today I’m astonished to see how far the industry has come, what’s new out there and the different offerings industry players are bringing to the market.

Machine Translation (MT) is one of those offerings that are grabbing lots of attention these days. It’s not new by any means, but over the last ten years, free services like Google Translate have helped highlight a clear need resulting from the different ways we communicate in a globalised world. Documents, texts, websites, books, WhatsApp messages, bots, scanned docs, virtually anything containing words can be translated. Other advances such as internet and processor speeds, cloud storage, speech-to-text recognition, and many others, means we can capture a lot of data, millions and millions of words, move it from one application to another, reutilise, clean, recycle it, rinse and repeat. Everyone has caught on to this: linguists and service providers, and businesses, too.

There is fierce market competition out there: what is the best engine, which service provider has the best offering, etc. And while a lot of progress has been made in the field of machine translation, the industry is not quite there in terms of fully automated translations. We are getting close, though.

But there are some unknowns as well. Some people don’t know exactly what it is, and some businesses think they don’t need it, others think they do but are not sure why, they’ve heard about it. If these questions sound familiar, my advice would be to work closely with your language partner and ask for insight that will help with your specific needs and objectives.

Let’s touch briefly on the four types of MT: Hybrid Machine Translation, Neural Machine Translation, Rule-based Machine Translation (RBMT) and Statistical Machine Translation (SMT).

Hybrid MT (HMT)

The HMT approach involves using different MT systems in parallel, most commonly RBMT and SMT systems. The resulting translations can be very effective in terms of quality. However, it requires a lot of editing by human translators.

Neural Machine Translation (NMT)

Neural machine translation is a corpus-based machine translation. The tool is trained using large corpus of language pair segments and their respective translations, usually containing hundreds of thousands or even millions of translation units. The approach is based on neural networks.

Rule-Based MT (RBMT)

The approach is based on grammatical rules, by grammatically analysing the source and the target languages to produce the translated text. However, it takes time to achieve a level of efficiency, mainly because it depends on lexicons, requiring a lot of editing.

Statistical MT (SMT)

SMT analyses enormous volumes of bilingual content with the purpose of finding correspondence between source and target words. Google Translate is a good example of SMT. SMT is great for straight forward translation, the problem is that it doesn’t take ‘in-context’ into account, often resulting in mistranslations.

Today Statistical MT is the preferred approach mainly because of the rules of language change, and that has an impact on the Rule-Based MT approach. Statistical MT doesn’t depend on rules, and a lot less time is required to build engines compared to Rule-Based MT systems.

Depending on the MT type some businesses and industries may be better served than others, automotive, retail, pharma, come to mind. What the clever language providers would have come to realise is that right type of MT can be part of tailored solutions that meet specific business needs.

It is key to work out the intricacies of a suitable workflow before you get started. What needs to be achieved, and how do we go about it? There are many elements that need to be taken into account: machine translation engine, translation memory, CAT tool, glossaries, etc., a lot of processing power and a translation project management system to run it all. And don’t forget to throw an experienced Production Team into the mix.

Despite all the technology and refined workflows, the resulting translations will be only as good as your assets are. The linguistic aspects need to be as robust as the rest of your workflow. How good is the MT engine? Does it integrate well with other language software? How good is your Corpus? Is there enough training data to build sound translation engines? Are your translation memories and glossary language and industry specific? Have they been validated by the end client? Are suitable linguists in place?

Finding out which approach works best for your business doesn’t have to be a minefield. Work with your language provider and find the best approach for your business.

Don’t have a language provider yet? Get in Touch.