AI Has a “Champagne Problem”

AI has a champagne problem. Professionals are overly concerned about the “real” thing, but the average buyer just wants something to do the job.

We’ve all been there. We are at a celebration for something and out come the bottles:

“Want some champagne?”

Then, someone takes a peek at the bottle and says “This isn’t actually champagne because it isn’t from the champagne region of France. It is sparkling wine.”

It is an interesting piece of trivia, but the fact is that most people at the party just want a little bit of bubbly. They don’t really care what it’s called, and they are going to keep calling it champagne whether it is "correct" or not.

The same thing has been happening lately with the term “translation”. Machine Translation or AI takes a piece of text and morphs it into a different language, and certain people take offense when people refer to this as translation. They say it isn’t “translation” because it doesn’t translate concepts, understand semantic meaning, or take context into account. Basically, it isn’t “translation” because it doesn’t come from a translator.

The problem with this perspective is the same as with the champagne comment: It is an interesting point at best, and a snarky distraction at worst. At the party the comment distracts from the celebration, in translation it distracts from the hard conversations we need to be having. And when we aren’t looking, everyone will go back to calling AI and MT output “translation” anyway.

So what are the hard conversations that we need to be having?

Essentially, NMT and generative AI have made it possible to take text or speech in one language and convert it into a second language. Instantly, inexpensively, and relatively accurately. That means our industry is changing and we need to change too.

Do AI and MT translate perfectly? No.

Is this “translation”? I don’t know. That's a philosophical conversation to have over a beer or coffee, not one to have when we are talking about the future of people's employment. It is interesting, not useful.

What I do know is that buyers are interested in the “sparkling wine” version of translation that they are getting. Good enough seems good enough, and arguing with them about what to call it won’t make them change their tastes or buy the more expensive version.

They will only opt to keep translators in the loop if we can close the gap between value and cost, and if we can explain the specific advantages of the premium product.

In fact, I imagine that the language industry will develop a similar shape to the sparkling wine industry. A quick Google search suggested that the global champagne industry is valued at around USD 7 billion, and the broader sparkling wine industry is valued around USD 36 billion. In other words, the “fancy stuff” is about 1/5 of the total industry, and the rest of the industry varies in quality and price.

So, if premium, "human" translation will occupy a smaller chunk of the total industry, what should we do? Are we being pushed out?

I don’t think so. Things are shifting, and this will inevitably lead to lost jobs and different ways of working. That is serious and unfortunate.

But, as language professionals we still have a massive opportunity to earn a living off of the work we love, and it starts with changing the story we tell ourselves about translation.

Let’s think, does everyone in the sparkling wine industry wish they were working with champagne? I doubt it. There are people who enjoy the process of making and delivering their product, and they are making a living doing it. The same will be true of the language industry. We can enjoy working with AI and MT, and we can make a living doing it.

There are many opportunities to work with new language tech and to use it to produce amazing translation work, even if it doesn’t feel like the old gold standard based on some definitions of translation.

We can also protect ourselves as industry professionals by doing research. We can find out where "champagne" style translations will be needed, and we can also find out how to help make "sparkling wine". The language we use about this makes it feel like a step down at first, but let's remember that most people like sparkling wine just as much as champagne. So why does it feel less rewarding to produce?

Because of the value that we assign to it- acting like it is inherently worse. Like it is less "ours". We can take control of the narrative and take pride in the work we do with machines, even if it is new and different. This is especially true now as generative AI allows us to influence the output of the machine, not just post-edit.

Working with generative AI is also where we will get to maintain a sense of creativity in our work, a major value among language professionals. While working with generative AI you have the chance to manipulate and train the model based on your creative choices. It is creatively different than translating manually, but it is creative nonetheless.

The point is that we have the chance to do satisfying work while making money as long as we lean in, but if we insist on “champagne or bust” we will find that it gets harder and harder.

Finally, we can make a huge difference in our own futures by specializing, upskilling, and using the new tech. This is hardly a new idea, but it has all of a sudden become controversial in reaction to new tools that are highly polarizing and poorly understood.

As translators, we need to do what any entrepreneur in any industry needs to do: Find the space where we can maintain a competitive advantage. This space is never static, so we will always need to adapt ourselves to keep our skills in demand and successfully trade them for income.

Right now this looks like using tech to translate faster and more accurately so that we can compete with the machine. Human quality is better, but it isn't always worth the wait or the cost for the end buyer. The right tech can help us bring both of those down while still earning well. We can close the gap so that buyers continue to think the bump in quality is worth the cost.

When we are justifying the value of translators to buyers, or even to other translators, we must explain the advantages concretely instead of philosophically. Telling them our translations are better because they come from translators is not helpful. "Understanding" and "context" are not inherently helpful concepts for the end buyer, we need to explain if and why our involvement will create different outcomes.

Then, based on the difference we create, the buyer will see the value of keeping humans in the mix with the machine.

It all comes down to outcomes; philosophical arguments are not a business strategy and pride is not a superpower. Our explanations need to make sense for the interests of the client because translation is just like champagne: If they can’t taste the difference, they aren't going to pay for it.

Let's make sure we are talking about how we can bring the most value to every situation that needs language services, not just the ones that require "champagne" translations. When we work with the tech to make our services more accessible, with "sparkling wine" as well as champagne, our services will be accessible to a lot more people. And our careers will be better for it.