Chen says that although a few reading comments from Facebook, Twitter, and others have managed to produce some of the unintelligible English words, the system often lacks content in other languages. This work should instead be carried out by volunteers such as his team, who look for similarities and are trained to remove it and reduce its spread. “The methods that are needed to get words and things do not mean that they will not happen if they are in another language,” he says.
Translation services and Google expertise such as Translatotron and real-time sensors use artificial intelligence to switch between languages. But Xiong finds these weapons incomplete in Hmong, a very complex language while the subject is the most important. “I think we’ve become dissatisfied and rely on high-end machines like Google,” he says. They say ‘they have a simple language,’ and then I read it and say it very differently. ”
(A Google spokesman acknowledged that small languages ”have a very difficult task to translate” but said the company “has conducted research studies that are more conducive to translation into smaller languages,” using machine learning tools and public feedback.)
All the way to the floor
The language barrier on the internet extends beyond the US — and down, literally, to the code in which it is located. Yudhanjaya Wijeratne is a researcher and scientist of the Sri Lankan think tank LIRNEasia. In 2018, they began following bot networks whose social media activities encouraged violence against Muslims: in February and March of the same year, several Sinhalese Buddhist riots targeted Muslims and mosques in the cities of Ampara and Kandy. His team Notes The “hunting ideas” of the bots, he wrote a list of hundreds of thousands of Sinhalese movies, and shared them on Twitter and Facebook. “They say good things and every good thing – canned words,” he says. (In a statement, Twitter claims that it uses social media and autonomous systems to “apply our policies impartially to all working people, regardless of their background, ideology, or political affiliation.”)
Linked by MIT Technology Review, a Facebook spokesman said the company had granted independence to the investigation into the platform’s response to violence in Sri Lanka, published May 2020, and modified some of these demonstrations, including the recruitment of several Sinhala and Tamil officials. “We have used the expertise of hate speech experts in Sinhala to help us quickly and efficiently identify potential threats to the content,” he said.
As the bot’s behavior continued, Wijeratne began to question this. He decided to look into the libraries and software tools used by companies, and found that hate speech in many non-English languages did not exist.
“A lot of research, multilingualism like ours has not been done,” says Wijeratne. “What I can do with the three lines of rules in Python in English took me two years to look at 28 million Sinhala words to build great demonstrations, to build basic weapons, and then to bring things to a level where I can and possibly do such a thing in word analysis.”
Following the suicide bombings of Colombo, the capital of Sri Lankan, in April 2019, Wijeratne developed a tool to analyze hate speech and lies in Sinhala and Tamil. The system, called Supervisor, is a free software that integrates content and integrates warnings into fake news. The warnings come from volunteers who have been trained to monitor the situation.
Wijeratne emphasizes that this work is not just about translation.
“Many of the things we take lightly that are often mentioned in research, especially in language development, show good results in English,” he says. “And the same algorithms, even those used in slightly different languages - whether it’s West Germany or the Romance language – can come with different results.”
Natural language change is the basis of self-improvement systems. Wijeratne published the paper in 2019 which analyzed the differences between their accuracy in different languages. He also said that statistical calculations available in the language, such as data and web pages, are possible in which algorithms can work. Languages from poor countries or regions are not allowed.
“If you are building, say, Empire State Building for English, you have its plans. You have weapons, ”he said. “You have everything you have and all you have to do is put this together. In any other language, there is no blueprint.
“You do not know where the concrete comes from. You don’t have the iron and you don’t have the staff, either. So you may be hitting one brick at a time and relying on either your grandchild or grandchild to complete the task. ”
The planning team is known for its language justice, and it is not new. American Bar Association it describes the justice of the language as a “framework” that preserves human rights “by communicating, understanding, and understanding in the language of their choice and by sounding better and more powerful.”