Facebook’s fake news clean-up hits language barrier

The article by Nilesh Christopher was published in the Economic Times on April 13, 2018. Sunil Abraham was quoted.

More than a third of Indian users engage on the social media platform in local languages. Experts are sceptical about the extent to which AI tools could be effective in curbing fake news, given that Facebook’s AI engine is primarily trained to recognise English.

“If the whole country speaks the same language, it is not a problem. But India is a country with multiple languages and their dialects and the 20,000 global numbers at the face of it doesn’t sound enough,” says Sunil Abraham, executive director for the Centre for Internet and Society.

In July 2017, India became the largest user base for Facebook with over 241million users, with a majority of them accessing the social network on their smartphones. Facebook has not shared updated user numbers since then.

On the use of AI in weeding out fake news, Abraham says that such tools usually work only in languages where there is a history of natural language processing. “Languages like English have a huge corpora (large databases of digitised content from the language). In such cases, the AI analyses the language and will be more accurate,” said Abraham. “Whereas in Indic languages, there is no training data. How they would use AI is not clear. For many Indian languages, the basic infrastructure doesn’t  exist."

“I don’t know exactly what FB claimed. But understanding local languages, Indian languages, is still an unsolved problem — either in non-free software or free software,” says Anivar Aravind, executive director of Indic Project, a nonprofit initiative working on language engineering and digital rights of native-language users. Interestingly, the two Facebook-owned platforms: WhatsApp and FB have become the preferred social medium to spread false information and largely through regional languages.

Facebook declined to comment.

On Tuesday, Facebook founder Mark Zuckerberg in his testimony to the US congress promised measures such as building and deploying AI tools that take down fake news and increasing the content moderation team to around 20,000. He cited the forthcoming elections in India to point out that Facebook would verify every political advertiser and said, “to make sure that that kind of interference that the Russians were able to do in 2016 is going to be much harder for anyone to pull off in the future.”

Currently, Facebook has around 15,000 moderators who review content to identify fake news on the platform, Zuckerberg said last week. The social media giant has been accused of not protecting user privacy and allowing voter-profiling firm Cambridge Analytica to harvest personal information of 87 million Facebook users without explicit permissions.

Cambridge Analytica has been accused of voter manipulation in several countries, including India.

Content moderators in India are seeing business grow driven by internet platforms scrambling to curb fake news across the globe. “Zuckerberg’s mention to prevent misinformation by increasing scrutiny before elections is positive. They are the market leaders. This is likely to create more awareness and it is encouraging to hear Facebook take a proactive role,” said Suman Howladar, Founder of Foiwe Info Global Solutions, a content moderation company.