Researchers conducted a ‘first-of-its-kind experiment’ that could help AI crack down on cyberbullying in different languages.
Researchers in Ireland have looked to make AI moderation tools for detecting cyberbullying more effective using state-of-the-art machine translation technology.
A new study was conducted by researchers at the Dublin City University (DCU) Anti-Bullying Centre and Adapt, the Science Foundation Ireland research centre for AI-driven digital content. It leveraged machine translation for cross-lingual cyberbullying classification among pre-adolescents.
For an AI moderation tool to efficiently identify whether a social media post or comment qualifies as cyberbullying, it needs to be trained with large amounts of fine-grained annotated data, which can be expensive and ethically challenging to produce.
Where such data does exist, it may be unavailable in the target language. Since manual translation is costly, the new research proposed a workaround using machine translation – which involves the use of software to translate text from one language to another.
The researchers looked at using machine translation to automatically translate a pre-adolescent cyberbullying dataset. This translated data could then be used to train an AI moderator to detect cyberbullying.
The study was enabled by funding from phase two of Meta’s Content Policy Award Grant, led by Dr Tijana Milosevic at the DCU Anti-Bullying Centre and Dr Brian Davis of Adapt.
Published by Cambridge University Press earlier this month, research for this article also involved TransPerfect. It was authored by Kanishk Verma and co-authored by Davis, Milosevic, Prof Maja Popović, Dr Angela Mazzone, Alexandros Poulis, Yelena Cherkasova and Cathal Ó hÓbáin.
“This study presents a first-of-its-kind experiment in leveraging machine translation to automatically translate a unique pre-adolescent cyberbullying gold standard dataset in Italian with fine-grained annotations into English for training and testing a native binary classifier for pre-adolescent cyberbullying,” the authors noted.
“In addition to contributing high-quality English reference translation of the source gold standard, our experiments indicate that the performance of our target binary classifier when trained on machine-translated English output is on par with the source (Italian) classifier.”
There are a number of ways in which pre-adolescent children are bullied online, including flaming, harassment, denigration masquerading and cyberstalking.
The researchers said that new forms of cyberbullying are being reported every day and pre-adolescent children “are not equipped with ways to deal with such negative experiences growing up”.
AI moderation tools can help mitigate this, but acquiring datasets to train AI models comes with its own set of challenges such as consent and assent, maintaining privacy, vulnerability and confidentiality.
“In the given scenario, research such as this can be ground-breaking in making a real difference to both the technological advancement in this area, and the lives of children being affected by such harmful online experiences every day,” the researchers added.
Updated, 9.30am, 24 October 2022: This article was updated to include details of the research funding and clarify the researchers involved in the study.
10 things you need to know direct to your inbox every weekday. Sign up for the Daily Brief, Silicon Republic’s digest of essential sci-tech news.