Tech

Meta claims breakthrough in AI translation ‘superpower’ as it translates over 200 languages ​​| Science & Technology News



Automated universal translation has long been a sci-fi dream. A new paper from artificial intelligence researchers at Meta, Facebook’s parent company, claims to be taking a step towards that goal.

The paper shows that machine learning, the technology behind AI, can translate 204 different languages, twice as much as has been tried before, with higher quality than has been achieved before.

That includes more than a hundred languages ​​that are rarely spoken, such as those of the Acehnese in Indonesia and the Chokwe in Central and South Africa, which have always made it difficult for computers to translate because they have very little difference. online presence.

Facebook and CEO Meta Mark Zuckerberg praised the feat, calling AI translation a “superpower,” and the researchers themselves were only slightly less enthusiastic.

It’s the latest development in artificial intelligence, a controversial area of ​​science that has recently been in the spotlight after a Google engineer was fired after being fired. claims a chatbot that can express thoughts and feelings.

Professor Philipp Koehn of Johns Hopkins University, one of 38 Meta scholars and researchers who collaborated on this work, said: “The paper presents impressive work to improve translation quality at the production level. Export to 200 languages.

“There will also be a lot of resources released that will allow people to take this model and retrain it themselves, promoting research in that area.”

More about Artificial Intelligence

“That’s right, I think it’s a pretty big deal.”

However, although the paper claims that it has “laid an important foundation towards the realization of a universal translation system”, the computer scientists not involved in the project insist that it is is a small step on a long and winding road, with no definite end. .

‘An impressive engineering feat’

Dr Alexandra Birch-Mayne, Reader in Natural Language Processing at the University of Edinburgh, says the paper’s central machine learning technique, a model known by the baroque term Sparsely Gated Mixture of Experts, is not is new.

Its biggest contribution, she said, is to regroup, clean and present new data in languages ​​that have not yet appeared widely on the internet, the main data source for machine translation.

“It’s an impressive engineering feat. It’s not necessarily a breakthrough in terms of basic science,” Dr. Birch-Mayne told Sky News.

In addition to translating languages ​​with fewer speakers, the article also claims to set a new standard for translation quality.

Read more: Facebook boss Mark Zuckerberg personally sued Cambridge Analytica’s massive data breach scandal

Data and algorithms are publicly available

Measuring progress in machine learning is a challenging task, but by means of a metric known as BLEU, the Meta paper has improved translation quality compared to previous cutting-edge technology by 44%.

“BLEU is an imperfect metric,” said Dr Diptesh Kanojia, Lecturer in Artificial Intelligence for Natural Language Processing at the University of Surrey. “However, BLEU score citation is a standard practice in natural language processing research.”

“If we look at this improvement purely statistically, the 44% improvement is quite significant.”

While the work will be used to improve Facebook’s software, the language data and algorithms used to translate it will be made publicly available, meaning for the first time there will be an authoritative dataset. rights to languages ​​such as Eastern Yiddish, Northern Kurdish and Cape Verdean Creole for use by other researchers.

Crucially, Meta researchers find native speakers to check their translations, a time-consuming task that helps preserve both the quality of the algorithm and the underlying language data.

“What’s admirable is the interaction with the community. They’re not necessarily initiating this trend, but they’re following good practice,” said Dr. Birch-Mayne, co-author Dr. Birch-Mayne. time notes the limitations of efforts involving native speakers in Europe and the US. rather than in the host country of the language.

Some researchers have criticized the fact that the paper was released without peer review, accusing Meta of practicing “media peer review”.

Professor Koehn defended the approach, saying it was “common practice in the field … for better or for worse” and helped improve the speed at which research results were communicated.

Read more: Amazon’s Alexa will soon be able to read you stories in the voice of a loved one – even if they’re dead

Advances in machine learning

The paper is one of several recent advances in machine learning, which is improving at a much faster rate than the researchers expected. A model released by Google last week solved a third of MIT’s undergraduate math problems with 50% accuracy, a dramatic increase in performance.

But although each new breakthrough brings speculation about new forms of consciousness, most people in the field believe that AI systems are neither sentient nor intelligent, saying they can’t. more than mimicking the data they are provided with. A robot uprising is not in the card.

The greater danger of AI systems is that they will lead to disaster by giving people false confidence in their very limited capabilities – a scenario that is very real given the sensitivity of the tasks that can be done. translation capabilities at Facebook, which have been criticized in the past for failing to have native-speaking moderators detect calls to violence on their platform.

Mr. Zuckerberg promised that “advancements here will enable more than 25 billion translations per day across our apps,” which Facebook says could include detecting harmful content, securing elections nominate and restrict online sexual exploitation.

Dr Birch-Mayne, who has just completed a three-year project on 17 languages ​​in Africa and India, working alongside the BBC, warns against using machine translation for anything with real precision. importance.

“You can’t rely on these systems,” she said. “It may be true, but it may not.”



Source link

news7f

News7F: Update the world's latest breaking news online of the day, breaking news, politics, society today, international mainstream news .Updated news 24/7: Entertainment, Sports...at the World everyday world. Hot news, images, video clips that are updated quickly and reliably

Related Articles

Back to top button