Will AI translation make learning a new language pointless? Experts weight in as Google, Samsung, ChatGPT spur ‘giant strides’ in technology
- Speech translation apps are improving, and while there are still limits to the AI-powered technology, experts say big advances are on the horizon
- The goal for apps is real-time speech-to-speech translation. Linguists weigh in on the impact this would have on language learning were it to be achieved
When the protagonist in the British cult sci-fi novel The Hitchhiker’s Guide to the Galaxy sticks a creature called a Babel fish in his ear, it translates all alien speech for him. Can artificial intelligence (AI) give us something similar?
Thanks to advances in machine learning, says Anatol Stefanowitsch, a linguist and professor at the Free University of Berlin, in Germany, we already have simultaneous speech translation systems that could make the arduous work of learning a foreign language obsolete.
Computer-based speech recognition and translation have become good enough for many everyday purposes, he points out, and modern smartphones often have a speech translator app installed on them.
The goal is machine translation of speech in real time, that is, with as little delay as possible.
The technological prerequisites for this now exist, says the Goethe-Institut. Experts at the German cultural association, which promotes cultural exchange and study of the German language abroad, distinguish between translating one language into another and foreign language proficiency, the latter going “far beyond simple simultaneous translation”.
Just a few years ago, it seemed like magic when your smartphone displayed the translated text of a menu you photographed with its camera. Today there are real-time speech-to-text apps that can help you break down language barriers – albeit with some time lag in the dialogues.
If, say, you’re in a foreign country, don’t know the language and want to make a restaurant reservation, the Live translate feature will not only display a translation of your call but also make it audible using a computer-generated voice. Here, too, there’s some time lag between the spoken and translated word.
Other limitations can include difficulty extracting the desired speaker’s speech from background speech or noise. And while translation accuracy has improved over the years, complex sentences, language nuances, idiomatic expressions and the like often present problems.
It’s better when people do the translations, not the machines
Verbatim translations of “a piece of cake” for something easy or being “under the weather” for feeling unwell, for example, will make no sense to a foreign interlocutor.
The more that speech translation apps are asked to do, “the more likely they still are to fail”, says Stefanowitsch, but adds: “Speech recognition and translation are areas where further giant strides by machine learning can be expected.”
The Goethe-Institut sees foreign language learning on the cusp of profound change spurred by rapid development of AI models such as ChatGPT, which generates humanlike text and whose applications include language translation. They will “transform the learning process as well as the way we communicate”, it says.
Will it make any sense at all to learn a foreign language in this brave new world of AI? Yes, according to Stefanowitsch. “I definitely think it’s valuable,” he says, adding that communication isn’t merely about the exchange of information, but also human interaction.
“In the future, too, you’re not going to want to maintain a friendship, let alone a romantic relationship, with an app constantly acting as an intermediary.”
What’s more, Stefanowitsch says, you can’t become fully immersed in another culture when “every utterance has to be translated by a computer”. Every language embodies a particular perspective of the world, he says, and you can only come to know it by learning the language.
This sentiment is shared by the Goethe-Institut, which points to the growing number of healthcare workers in Germany whose mother tongue isn’t German. Natural speech engenders empathy, it says, asking: “Do we want to live in a world where healthcare workers communicate with patients via simultaneous translation?”
With the availability of digital translation tools, traditional foreign language teaching in schools seems old-fashioned. Nevertheless, Stefanowitsch says that schoolchildren, “for the most part”, are learning foreign languages the right way, arguing that language-learning apps can complement school lessons, but not replace them.
The Goethe-Institut expects the role of foreign language teachers and instruction to change “from simply imparting knowledge to actively supporting learners”.
When the learners are at home, they could then have AI tools do their homework for them, whether it involves a foreign language or another school subject. Consequently, “in future we’ll have to steer clear of all the kinds of homework that can be handled by so-called AI applications”, says Stefanowitsch.
AI has its flaws though. Machine-generated texts are very much aimed at producing a kind of “average” language, which is why “they always sound extremely hackneyed and not very original”, he says.
The Goethe-Institut warns that a proliferation of standard speech through widespread use of simultaneous speech translation systems could reduce people’s vocabularies as well as endanger dialects and vernaculars, thereby impoverishing the language.
Stefanowitsch’s verdict: “It’s better when people do the translations, not the machines.”
