Skip to main content

Can we tell humans and machines apart?

I spoke to Unbabel’s VP of Language Technologies about the “imitation game” and the human behind the machine

If you’ve been online today, there’s a really good chance you have been asked to prove you’re not a robot. You may have thought “oh, not that again!” while trying for the second time to tag all the images containing a traffic light. Next time you see one of those CAPTCHA messages, you’ll have a name to ‘blame’: Alan Turing. CAPTCHA is the acronym for Completely Automated Public Turing test to tell Computers and Humans Apart, developed with the principles created by English mathematician, computer scientist, logician, cryptanalyst and philosopher Alan Turing.

The purpose of the Turing test, described in the 1950s paper Computing Machinery and Intelligence, was to answer the question “ Can machines think?” by analyzing if a human interrogator could tell, through conversation, the difference between a machine from a human being. The Turing test became better known to the general public after the film The Imitation Game (2014), showing the creation of the decoding machine that helped the UK decrypt German messages during World War II.

“Instead of trying to produce a programme to simulate the adult mind, why not rather try to produce one which simulates the child’s? If this were then subjected to an appropriate course of education one would obtain the adult brain.”
A. M. Turing

Would machine translation engines pass the Turing test?

I asked this question to Alon Lavie, VP of Language Technologies at Unbabel, Consulting Professor at Carnegie Mellon University (Pittsburgh — Pennsylvania USA) and the 2021 winner of the Makoto Nagao IAMT Award of Honor for his contributions to the field of machine translation. “In conversational settings, today’s chatbots are getting closer to passing the Turing test in terms of their language generation capabilities… [these days] it largely depends on the language, type of content and the context. In some situations, it’s becoming increasingly difficult to tell. But in most enterprise use cases, we still need humans to ensure consistent high-quality, brand-compliant translations.”

Ex Machina

Ever since the introduction of Machine Translation into the workflow of translation processes, one of the biggest concerns of the translator/editor community has been that machines will eventually replace human expertise. Alon shared with me the two views he has witnessed stemming from this fear coming from translators:

“Machine translation is so bad it isn’t worth my time”

When translators are asked to post-edit machine-translated text, they often find that the target text is so bad that it would take more time and effort to correct it than it would to translate the text from scratch. Alon tells me that while this was true a few years ago with earlier MT technology, with today’s neural MT, the time an editor spends on a machine translated job has drastically improved for most of the major languages.

“The new neural models are very good most of the time, and they typically generate mostly fluent and grammatically correct language. But they still make critical mistakes. The good news is that AI technology for detecting translation errors is also evolving rapidly. So it will become easier for translators and editors to detect those mistakes, but we will continue to need humans to fix these errors”.

“Machine translation will do everything and I won’t be needed at all”

Machine-led efficiencies are being developed in many fields in this day and age, not just linguistics and translation, and many professionals from all areas are wondering how their roles are going to change because of it and how best to keep up with the swift changes in their markets. If we think back to when the Industrial Revolution began, this is not a new concern. Almost two hundred years later, we perform work in all fields using tools and machines that we can’t imagine working without, although at the time of their invention, we saw them as a threat.

Everything that these tools are meant to do is help us by doing the work that can be performed by a machine, so that humans can step in where their input is irreplaceable.

Alon had this to say:

“I honestly believe that the translator community will never be out of a job, but their work will look different than it has in the past. For the most part, translators will become the expert editors who ensure the accuracy of the target text, the correct semantics of the subject matter, and the brand language of enterprises.”

Pocketful of resources 🎁

The Association for Machine Translation in the Americas (AMTA) for academic papers, presentations, conferences and tutorials has a lot on the latest in AI translation; Alon recommends it to anyone who wants to know more on the topic.

Another great resource to check out is the European Association for Machine Translation (EAMT). We’re going to take a moment to mention that we’ve just had seven papers accepted for their Annual Conference in Belgium! This is our biggest ever participation, so way to go, Unbabelers!

Let’s keep this conversation going
Do you think the machine translation engines we have today would pass the Turing test? Share your thoughts with us!

Can we tell humans and machines apart? was originally published in Unbabel Community on Medium, where people are continuing the conversation by highlighting and responding to this story.