跳至主要内容

Article 4 min read

Machine translation—is it good enough for customer support?

Maria Almeida

最後更新 March 15, 2024

In 1954, at the first public demonstration of a machine translation (MT) system, researchers from IBM declared that translation would be fully handled by machines within the next few years.

Of course, we now know how far from the truth that statement was and continues to be. Even earlier, in 1947, American scientist and MT pioneer Warren Weaver said:

“One naturally wonders if the problem of translation could conceivably be treated as a problem in cryptography. When I look at an article in Russian, I say: ‘This is really written in English, but it has been coded in some strange symbols. I will now proceed to decode.'”

A few years later, Weaver followed up with this: “No reasonable person thinks that a machine translation can ever achieve elegance and style.”

Translation requires more than a decoder ring

If you’ve ever tried to translate poetry or literature with an MT service, the output might look a lot like you’re decoding secret messages. It raises the question: Why is translation so difficult?

André Martins, head of research at Unbabel and one of the leading researchers in the field, explained over lunch, “Humans have a huge knowledge about the world and that is crucial for translation. As a human, I can understand the context, the culture of a specific language, and that is key in translation. Machines don’t have yet that kind of knowledge.”

As a human, I can understand the context, the culture of a specific language, and that is key in translation. -André Martins

MT systems are trained to read parallel sentences, which, as Martins says, “is really like teaching a parrot to talk—instead of making the parrot truly understand the context and meaning of what’s being said.”

That said, the latest advances in the field involve “Neural Machine Translation” (NMT), which has rapidly become the new state-of-the-art technology. Essentially, the computer system acts more like a brain by imitating biological neural networks, progressively learning and improving with more data.

But even these new NMT systems aren’t necessarily up to the job of delivering a consistent tone of voice with minimal errors, both of which are requirements for use by a modern enterprise business.

Machine + human translation in customer support

For most people, Google Translate is the first thing that comes to mind at the mention of machine translation. But would you trust Google to translate everything you need to tell your customers?

Probably not—for this or that reason (and many others). The thing is, a human quality translation without grammar mistakes, typos, and mistranslations requires a human. And the tech giant is not alone in facing these challenges when it comes to NMT—nor is it the only company trying to address these challenges.

“It’s very easy to beat out of the box Google,” according to Andy Way, one of the fathers of MT and a professor at the ADAPT Center at Dublin City University. Why? Because for neural machine translation to work we need human translations to feed into the systems and train the machines to learn. Once the system receives all the data, it starts to learn patterns and to produce better translations. And today you can’t do that with Google.

Neural machine translation relies on human translations that feed into the systems and trains them.

So, does this mean you can reliably translate customer support emails or even real-time chats using machine translation? Has the technology reached its full potential?

Pure machine translation systems lack the ability to adapt to different content types (FAQs vs. customer tickets, for instance), and they are unable to customize to the requirements of specific clients, and likewise are unable to choose the right tone (formal vs. informal).

In short, pure machine translation systems lack the “human touch” required for understanding cultural references and contextual differences. Today, however, NMT combined with advanced, automated quality assurance and post-editing by humans, ensures translations that are sound, and sound good—and often delivered within 20 minutes. Machine translation may not be at the end of its road, but it’s come a long way toward meeting critical business needs—allowing companies to support a global customer base and to provide a consistent customer experience, no matter the language a customer speaks.

By itself, machine translation will always sound like a “better parrot” but aided by data and people, it can offer an elegant solution to businesses and support teams struggling to respond to all their customers, the world over.

Maria Almeida is currently living in a world powered by the randomness of her own thoughts but she’s never giving up on her personal plot to change the world by telling stories. A couple of years ago she mistakenly brought a bear disguised as dog home and named him Fausto (she has been trying to domesticate him ever since). She’s the Almighty Duchess of Content at Unbabel and also co-founder and co-host of É Apenas Fumaça, an independent media project from Portugal.

相關案例

Article
6 min read

Science-based targets are the key to sustainable business

To help combat climate change, many companies are setting science-based emissions reduction targets. Learn more about these efforts and the impact they can have on the planet.

Article
5 min read

Here's how customer service teams are actually using AI

From bots to automated workflows under the hood, Generative AI tools for customer service driving higher productivity, happier agents, and satisfied customers

Article
5 min read

That’s a wrap: A look back at Zendesk Relate 2022

We loved seeing you at Zendesk Relate 2022. Here are a few highlights—and a peek into what you can still explore online.

Article
6 min read

We’re placing some bets on the future of customer experience

Join us at Relate to hear our five big bets on what the customer experience will look like by 2030.