Friday, December 09, 2022

Tech firms take it slow to build the ultimate AI translator – Nation World News

But researchers warn that the dream of a real-time conversation translator is still far away

A South African man speaks Sepedi to a Peruvian woman who only knows Quechua, yet they can understand each other.

The universal translator is a staple of science fiction, but Google, Meta and others are locked in a battle to achieve as many languages ​​as possible with their AI models.

Meta chief Mark Zuckerberg announced Wednesday that his firm now has a block of 200 languages ​​that can be translated into each other, a number doubling in just two years.

P14 220708 Pic14S1

Photo: Screen Grab

Meta’s innovation, trumpeting in 2020, was to break the link with English – a drain language for a long time due to the vast availability of sources.

Instead, Meta’s models go straight from Chinese to French without going through English.

In May, Google announced its own great leap, adding 24 languages ​​to Google Translate after pioneering technologies to reduce noise in samples of lesser-used languages.

Cepedi and Quechua were, of course, among them – so Peruvians and South Africans could now communicate, but so far only with text.

Researchers warn that the dream of a real-time conversation translator is still far away.

quality vs quantity

Google and Meta both have business motivations for their research, not least because the more people using their devices, the better at feeding data back into the AI ​​loop. They’re also in competition with the likes of Microsoft, which has a paid-for translator, and DeepL, a popular Web-based tool that focuses on fewer languages ​​than its rivals.

Researcher François Yvonne said the challenge of automatic translation is “particularly important” for Facebook to filter out hate speech and inappropriate content.

This tool will help moderators who speak English, for example, in identifying such content in many other languages.

Meta’s promotional videos, however, focus on the freer aspects of the technology – amateur chefs showcasing recipes from far and wide right at their fingertips.

But both companies are also at the forefront of AI research, and both accompany their announcements with academic papers that highlight their ambitions.

The Google paper, titled “Building Machine Translation Systems for the Next Thousand Languages,” clarifies that the firm is not satisfied with the 133 languages ​​that are already available on Google Translate.

However, as the cliché goes, quantity does not always mean quality.

European dominance

“We shouldn’t imagine that 200×200 language pairs will be at the same level of quality,” said Yvonne of Facebook’s model. European languages, for example, will probably always be an advantage because there are more reliable sources.

As regular users of tools such as Google Translate and other automated programs will attest, the text produced can be robotic and mistakes are not uncommon.

While this may not be a problem for day-to-day use of restaurant menus, it limits the usefulness of those tools.

“When you’re working on translating an assembly manual for a fighter jet, you can’t afford a single mistake,” said Vincent Goddard, who runs the French tech firm Cistran.

And the final nut to crack is inventing a device that can translate the spoken word seamlessly.

“We’re not there yet, but we’re working on it,” said Antoine Bourdes, who runs Faire, Meta’s AI research lab.

He said that META’s speech translation project currently works in very few languages.

“But the interest will be in linking the two projects, so that one day we can speak 200 languages, while retaining the intonation, the emotion, the accent,” he said.

Nation World News is the fastest emerging news website covering all the latest news, world’s top stories, science news entertainment sports cricket’s latest discoveries, new technology gadgets, politics news, and more.

Latest News

Related Stories