Working in the translation industry, it is inevitable that you get asked questions about technology and the future. Many people seem to take it for granted that machines will one day replace our human translators, and I’m often surprised at how soon many people think this day will come.

For those of us who work with human and machine translation on a daily basis, computers are a great help, sure, but they are also a source of frustration at times, and by no means ready to boot us out onto the streets. What’s more, they are totally reliant on human input — and good quality human input at that.

In order to try and straighten things up and clarify a few things, I thought that I would use my first blog post on Pulse to take a closer look at this interesting yet thorny issue at the heart of our industry.

How does machine translation work, and how much better can it get? How do we use it today? And why will we always need human translation in the future?

Machine Translation Feeds on Human Input

What many people outside of the industry don’t realise is that machines don’t actually create their own translations. Instead, they draw from real-life human translations and connect the broken pieces together, like multi-coloured (and sometimes mismatched) blocks of Lego.

As tech-insider and American computer scientist Jaron Lanier has spoken and written about, this means feeding on page upon page of human content, effectively lifting and repurposing other texts in order to remain up to date and to learn a little along the way.

This practice is what has allowed machine translation to develop to the stage it is at today, where it can provide relatively comprehensible translations that give us the basic gist of a text. It is also deployed in the translation industry in the form of machine-aided human translation.

This is when machine translations are automatically generated and used as a springboard or a crutch for human translators.

This gives human translators a readymade base that they can fix up and build upon, sometimes speeding up the human process.

Tech Problems

But there are two key points to make here. The first is that in all cases, humans are still the ones doing the work. In his insightful writings, Lanier points out that it raises ethical concerns for machines to use human texts in order to craft their output, without asking for permission, without paying remuneration and without giving credit. This becomes especially problematic if professional translators begin to lose work as a result.

The second is that machines have pretty much come as far as they can using this kind of technology. Despite all the data being fed into these systems all the time, they still lack that human sense for language which gives us clear, crisp and comprehensible texts. In the translation industry, we see this day after day.

Quite often, it is a source of amusement when our machine translation tools churn out particularly ridiculous results. The internet is awash with no shortage of rib-tickling yet sobering examples.

The Human Touch

The problem is largely a question of both context and common sense. Machines are great at data-grabbing and dumping, but not so much at recognising or appreciating the subtleties of speech and text, which is precisely what makes language both so wonderful and also so complex.

Take synonyms, for example. Most words can be translated in a myriad of different ways, and only context can tell us which option to choose.

Computers can develop a certain sense for this, for example by recognising common structures which always use a certain term or by picking up on frequent relationalities, but this only gets them so far.

Outside of common boilerplate phrases, computers just have to stab in the dark, meaning that accuracy becomes a matter of statistical probability. And that won’t cut the mustard in a professional industry such as ours.

Speaking about stabbing in the dark and cutting the mustard, that brings me on to my next point. Any idea how machine translation tools handle these phrases? That’s right, quite often they are rendered literally. Cortar la mostaza. Hugge i mørket. Actually cut the mustard and stab in the dark.

And this is what happens with a great many idioms in multiple languages – they are translated word for word, even despite being common expressions that our machines could in theory learn to recognise.

When you think about just how often we use idioms in our everyday speech, the magnitude of the problem becomes clear and the penny starts to drop (not literally, of course, but you knew that).

What About True Artificial Intelligence?

The type of machine translation that we have today and which I have been discussing above is based on machine learning – a type of artificial intelligence which uses trial and error.

But right now, companies all around the world are working to develop what is often referred to as general AI or true artificial intelligence – computer systems that can actually think and reason just like us. Surely that will replace human translation?

The answer to that is actually yes – quite possibly.

But it won’t just be translators. If artificially intelligent machines can replace translators, then they will be able to translate factory workers, shop assistants, civil servants and even politicians.

In other words, what we’re talking about here is a world of science fiction very different to the world we live in today – one which may never come into existence and which certainly doesn’t lie just around the corner.

Which is all to say that for the foreseeable future, not only are human translators safe in their jobs, but they are absolutely necessary.

After all, language is in essence a very human thing – it’s not just zeros and ones that can be swapped out for identical counterparts, but a living, thinking and feeling phenomenon which warps and shifts with every passing day, coloured by nuance, affection and hidden meanings.

It is difficult to imagine how machines could ever take on the job of translating such complexities, and frankly it can be quite frightening to think that they may be entrusted with such a task.

Particularly as we enter an age of global challenge, when governments and companies need to collaborate more than ever across linguistic divides. So yes, by all means, let’s make use of the excellent technologies and the tools we have today, but always, always, always with a human in control.