Adedapo Adesanya
By Adedapo Adesanya
Facebook has unveiled a software based on machine learning which the company said is the first to be able to translate from any of 100 languages without relying on English.
The open-source artificial intelligence software was created to help the massive social network deliver content better in 160 languages to its more than two billion users around the world.
In a blog post, Facebook Research Assistant, Ms Angela Fan, noted that, “This milestone is a culmination of years of Facebook AI’s foundational work in machine translation.”
She said the new model is more accurate than other systems because it does not rely on English as an intermediary translation step.
It’s called M2M-100 and was trained with a data set of 7.5 billion sentence pairs from 100 different languages and the team built the universal model with 15 billion parameters.
“When translating, say, Chinese to French, most English-centric multilingual models train on Chinese to English and English to French, because English training data is the most widely available.
“Our model directly trains on Chinese to French data to better preserve meaning. It outperforms English-centric systems by 10 points on the widely used BLEU (bilingual evaluation understudy) metric for evaluating machine translations,” Ms Fan said.
Facebook said it already handles an average of 20 billion translations every day on its news feed, and that it hopes the new system will deliver better results.
“Breaking language barriers through machine translation is one of the most important ways to bring people together, provide authoritative information on Covid-19, and keep them safe from harmful content,” she said.
Facebook researchers used specific criteria for language selection, which includes those from different families with geographic diversity and those that are widely used.
However, they avoided directions that were statistically rare, including Icelandic-Nepali or Sinhala-Javanese, it noted.
Then the team organized languages into 14 groups based on linguistic classification, geography and cultural similarities.
“People living in countries with languages of the same family tend to communicate more often and would benefit from high-quality translations. For instance, one group would include languages spoken in India, like Bengali, Hindi, Marathi, Nepali, Tamil, and Urdu,“ she said.
Although Facebook provides an option to “Translate” on posts, the firm says the new AI model is 90 per cent more accurate, as it does not need to first translate text to English and then to the desired language.
