AFCN and Full Fact Unite for AI-Powered Fact-Checking in Arabic 

April 2, 2024

Written by: ARIJ’s AFCN team

Since ARIJ (Arab Reporters For Investigative Journalism), started its Artificial Intelligence (AI) journey in 2021, its focus has always been, and will continue to be, supporting its community of investigative journalists and fact-checkers, facilitating their work, and encouraging them on using AI ethically, to ensure that this use aligns with our principles and does not impact negatively our work in promoting transparent, objective and high quality journalism in the Arab world.

One of the main aspects that AI has shown an increased achievement in, is fact-checking. At the same time, AI is a main reason for creating deepfakes and spreading mis/disinformation which makes the fact-checking process more challenging, and encourages us to develop new techniques to facilitate our work.

After several years of working, supporting and enhancing the work of fact-checking organisations, initiatives and freelancers in the Arab world, the Arab Fact-Checkers Network (AFCN) from ARIJ, has concluded that technology is no longer a luxury when it comes to fact-checking. Understanding the role of AI and its importance for fact-checking, how it can support the process, and simultaneously harm it, has become one of the network’s essential focuses. 

One of our main conclusions too, was that our work as fact-checkers should not remain 100% manual. Arabic-language fact-checking tools have become a critical need to accelerate our work. After localising several tools like INVID and developing new ones like AFCN Chatbot and AFCN marketplace, the network continues to facilitate and expedite the fact-checking process in Arabic by using AI, through collaborating with leading partners and tech companies.

This is why, ARIJ’s AFCN has been collaborating with “Full Fact” starting late 2023, in developing their “AI Fact-Checking tools” in Arabic language. This project, supported by Google News Initiative, is one of the main AFCN’s projects to integrate AI in the work of Arab fact-checking organisations in the aim of facilitating and expediting their work. 

Today, on the occasion of the “International Fact-Checking Day”, ARIJ’s AFCN and Full Fact are launching the tools in Arabic language, after months of testing, annotating data in Arabic, translating data to Arabic, technical development, and discovering how these tools can play a crucial role in developing the work of its fact-checking community. 

The Arabic data annotation concentrated on identifying key terms, classifying claims, determining the types of claims, identifying claims that deserve fact-checking, predicting quotes, and identifying claimants, among other aspects. This process presented challenges because of certain characteristics of Arabic language, where the same spelling can correspond to words with different pronunciations and meanings, leading sometimes to incorrect matches. Additionally, accurately determining the relevant topic proved difficult when dealing with parts of words or elements of compound nouns. The annotation process is set to continue over the upcoming months, aiming to enhance and expand the features of the tools in Arabic language.

Full Fact AI tool contains the following features:

• Monitoring news to know the most important thing to be fact-checked daily.
  
  
• Easy advanced-search for the claims in media outlets/social media pages that the fact-checking organisations monitor.
  
  
• Knowing when someone repeats something they already know to be false and setting alerts for this.
  
  
• Checking things in as close to real-time as possible.

Full Fact has been developing tools to support fact-checkers since 2016. These tools were initially developed in English and are in active use in several English-speaking countries. The plan had always been to extend them to work in other languages, and with recent support from Google, Full Fact has produced versions working in both French and Arabic. Several core components are based around “large language models” published by Google and other organisations. 

These are large statistical models of language, capturing information about how words are combined to form sentences and how sentences are combined to form documents. Several of these models are already multilingual, so Full Fact’s tools needed only small changes to make use of those capabilities. Working closely with AFCN, Full Fact were able to evaluate the performance of the tools in Arabic and where necessary, perform some fine-tuning of the models. 

“Full Fact AI tools” in Arabic, starting today, will be available for 20+ fact-checking organisations/initiatives (Akhbar Meter, Annir, Beam reports, BN Check, Chayyek, Etreek 24, Fact Yemen, Kashif, Fact-O-Meter, Matsada2sh, Misbar, Saheeh Masr, Saheeh Yemen, Sawab, She Checks, Sidq, Tafnied, Tahaqaq Palestine, Tahaqaq 360, Tayqan, The Checker and Tuni Fact) from 9 Arab countries (Egypt, Iraq, Jordan, Lebanon, Libya, Palestine, Sudan, Tunisia and Yemen) with a plan to make it accessible to more organisations from more Arab countries in the next few years. 

“In today’s challenging media environment, AI is a crucial partner for fact-checkers dedicated to upholding accuracy in any language. Making Full Fact AI’s tools work in Arabic was the first part of the process. This was followed by onboarding and training of the organisations’ managers, editors and fact-checkers followed by regular support sessions. This ensures that the organisations not only know how to use the tools but are also equipped with the skills to integrate AI into their organisation successfully,” said Kate Wilkinson, Full Fact’s Senior Product Manager.

“We recognize the vital role of collaboration in elevating the effectiveness of the fact-checking community. Partnering with Full Fact to provide their AI tools in Arabic with the support of GNI, will mark a significant milestone for AFCN’s fact-checking community. This collaboration will help in addressing numerous challenges providing a transformative opportunity to overcome obstacles and enhance fact-checking impact”, said Rawan Damen; ARIJ Director General. 

ARIJ’s AFCN collaboration with Full Fact in providing AI tools that support the fact-checking process in Arabic language, has significantly reduced numerous challenges. These challenges include overcoming linguistic obstacles, dealing with data complexities, and technological barriers. This collaboration has simplified what could have been a more complicated and time-intensive process, streamlining the fact-checking procedure and making it more efficient and accessible، with increasing fact-checking organisations productivity and ability to deliver accurate information in a widespread misinformation world.