Aleppian Moon

In late 2019, amidst the Lebanese uprising, several female journalists covering the news were subjected to toxic online abuse. Dr Hala Mulki, a data scientist, who uses sentiment analysis, a natural language processing research field focused on analysing people’s opinions, explored the Arab Twittersphere to find views about the revolution. She found extreme misogyny being perpetuated, yet it was not flagged by Twitter. “What I found sad is when it comes to low-resource languages, like Arabic, Twitter cannot always recognise the toxic content because it can’t understand colloquial Arabic, so misogyny is still there and misogynists can tweet against women without facing any consequences,” Mulki argues that if all the linguistic variations used to target women or minorities had been recognised by Twitter AI tools, for example, a policy that instantly recognises such abusive content would have been developed. “I can only talk about my domain.” She says, referring to Arabic speakers. “I expect that with English Twitter, some regulation can be successful.”

Mulki uses natural language processing and artificial intelligence to develop a machine learning-based model that understands Arabic dialects. She trains her model, feeding it thousands of linguistic inflexions and dialects from across the Arab world. Sometimes there are foreign words in the dialect; sometimes, Arabic words are written in Latin letters, a form of transliteration that has come to be known as Arabizi. 

Arabic sentiment analysis, the sifting, categorising and analysing of opinions research, is still in its infancy. Current language models that rely on official Arabic or Modern Standard Arabic don’t consider that people’s expressions in Arab social and political contexts are often colloquial and that most Arabic social media commentary is made using local dialects: “I am Syrian but cannot understand Moroccan Arabic, which has different dialects flavoured with Spanish and Tamazight, amongst others. One word which might mean something in the Syrian dialect is the opposite in the Moroccan dialect. So, designing a universal model that can understand the opinion embedded in those dialects would be very beneficial for social-political studies.” Originally from Aleppo, Mulki currently lives in Ankara, Turkey, with her elderly parents, working for ORSAM, a think tank with its finger on the pulse of the most challenging political events in the Middle East. 

Her work on dialects already contributes hugely to their analysis and reports. In addition, she can mine the Arab social media world for opinions about important issues such as elections and disinformation in different countries. Her work also has academic application in the political and social science field. “Computer engineering is without a soul”, she says, “but once you pair it with scientific and humanitarian research, it can result in novel ideas.” Her last research was a collaboration with Syrian colleagues supported by their Cara mentor, Dr Maria Kyriakidou, a Senior Lecturer at the Cardiff University School of Journalism, Media and Culture. Over eight months in 2021, they collected more than 100,000 tweets from Turkish Twitter, of which 30,000 subjective tweets were analysed to explore the contentious debates about Syrians. Their conclusions were insightful: discriminatory views, mistrustful attitudes, hurtful opinions, and sometimes rampant toxicity circulated and retweeted by bots from as far away as Asia. 

Mulki’s love affair with linguistics is rooted in her childhood in Syria. From age three, she started to learn English, but her mother’s library ignited her passion for Arabic in particular. Her mother taught Arabic, and her book collection inspired a love of the creative format of language in poetry. Mulki began to memorise and recite it: “The Arabic language is rich and fascinating,” she says. She studied computational linguistics as a subfield of computer engineering during her PhD, gradually becoming more interested in language processing and how it could shed light on the region’s cultures. 

Mulki graduated from the University of Aleppo and lectured in computer engineering for eight years before the demonstrations broke out: “The students were the main protesters and, one day, I found myself giving a lecture to just three students out of 300 because all of them were out there protesting against the regime. “By 2012, all my siblings had already left.” She says, “I left suddenly with my mum and dad. It was not a hard decision. We went to Alexandria in Egypt, where my brother had a lot of friends who could help us.” 

The family stayed in Egypt for a year; where Mulki tried to register to do her PhD at Alexandria University, but her application was denied for over nine months for unknown reasons. Finally, she won a scholarship to Selçuk University in Turkey to continue her doctoral work in computer engineering: “So we made another decision and came to Turkey in 2013. Here I am. I found a job. It is not academic, but I joined the ORSAM Center for Middle East Studies, where I discovered the importance of my research.” 

At ORSAM, she has worked on the Tunisian elections, analysed disinformation in conflicts between Qatar, Turkey and Saudi Arabia; conspiracy theories during the pandemic in Kuwait. She gives talks to undergraduate students on disinformation and hate speech in social media, but it is her research straddling the social sciences that will be revolutionary in the socio-politico domain.

She first encountered Cara in 2013 when she was in Egypt. “A friend of mine warned me, ‘Don’t join Cara because it is a fake organisation. You cannot rely on them.’ She laughs now. A colleague of hers at ORSAM told her again about Cara. “‘Cara is real? I asked, and he told me, it is!” She emailed Cara, convincing them that she was an academic, and met the research criteria despite working in a think tank rather than a university. She wrote a proposal explaining the implications of her work and how she needed to continue it. After a few months, she was accepted and became a Cara fellow. “After that, we wrote an academic journal paper, and now I am applying to international journals. It is my first time dealing with journals outside the computer science domain, and I am excited by this experience. I will start a remote Cara research incubation visit (RIV) with my mentor from Cardiff University, where I will work on misogyny during the 2022 World Cup, comparing traditional and social media. It is my first time dealing with traditional media, and this will result in novel research. Thankfully, due to the inability to leave my parents behind, Cara agreed that my RIV could be carried out remotely, during which I will be able to participate in online seminars and courses and learn how to write for the social sciences.”