Machine Translation — Does it Really Work?

Dina Indelicato
3 min readAug 2, 2017

--

A lot of us have used Google translate. When we see something in a foreign language online, and we want to read it, it’s pretty easy to do. And we can get a pretty good idea of what it says and means, even though the grammar and word usage is not always good.

There are translations “machines” that are much better than Google Translate, of course. Basically, they are pieces of software, and their translations are as the programs make them.

Two Types

Machine translation is either rules-based or statistical. Rules-based translators takes standard rules of grammar of each language, along with vocabulary, and perform the translations. They are pretty good, so long as the content is not highly specialized or complex. Medical research and other highly technical content is tough for a rules-based machine.

The other type is statistical. It uses translations tools and, over time, it learns more and more about languages and becomes better. This is known as artificial intelligence (AI), and it is usually preferable to earlier types of rules-based software. For example, if there is a lot of medical terminology that is re-used many times, the machine will come to remember it and translate it accurately from that point forward. It’s ability to continue to learn is usually unlimited.

Can Machines Replace Humans?

The short answer is “no.” And the reasons is this. No matter well a machine is learning, there will always be syntax and linguistic details that are not yet known, and nuances, like irony, sarcasm, and puns will not translate correctly. It takes a human who is intimately familiar with both the original and target language to get these and translate them correctly. And if an exactly correct translation is needed, then a machine cannot do it completely. At the very least, you will need a human to take that machine translation and edit it.

Improvement in Machine Translation

Yes, machine translation is getting better, especially because AI technology is improving. Especially, if the same topical area is used over and over again, the learning will get better. And, you can always add in things like a glossary of terms.

Especially for long amounts of content which are repetitive and not complex, machine translations do pretty well. Still, a human translator should perform a post-translation edit just to make certain.

Can Machine Translation Ever Stand Alone?

This, of course, depends on the purpose for the translation. Some personal translations can tolerate errors — letters, some articles that are needed for study purposes, etc.

Businesses cannot usually afford translation mistakes — websites, marketing material, brochures, reports, handbooks and manuals, etc. All of these things must be of highest quality and will require the human touch. The same applies to any legal document, any items for publication, or research documents for international sharing. High quality is demanded, and a machine translation, by itself, is just too risky.

The verdict is still out on some areas of translation. For example, e-commerce retailers, with large catalogues to be translated into several languages, could spend a fortune using human translators for everything. They are willing to put up with some syntactical and grammar errors in order to get the volume they need, and get it quickly. In these instances, the meaning of product descriptions is usually not lost in translation.

It’s Always a Personal Choice

In the end, you must decide on the quality level of the translation you need or want. There are plenty of review sites online for translation agencies, such as Pick Writers. Your best bet, in most instances, is to find a translation service that has a good reputation and that allows you to choose only human translators or a combination of machine and human methodologies. You can then determine the quality you need for each piece of content.

--

--