Mubasher TV
Contact Us Advertising   العربية

Story behind first Arabic voice assistant ‘Salma’ – Interview

Story behind first Arabic voice assistant ‘Salma’ – Interview
Salma is the Arabic version of Google Home and Amazon Alexa

By: Ingy ElSafy & Moslem Ali

Cairo – Mubasher: Salma, the first Arabic personal digital assistant developed by Mawdoo3, was among tech startup success stories featured in the recent RiseUp Summit, held in Cairo on 5-7 December.

Mubasher visited Salma’s booth on the exhibition floor and spoke to some members of the team behind the artificial intelligence (AI) project that became the Arabic version of Google Home and Amazon Alexa.

“Salma was the next step for Mawdoo3 to move from only publishing online Arabic content to developing the first Arabic AI solutions, thus, presenting the platform’s content through a virtual assistant. [The project] began with Moustafa Samir, our Senior Machine Learning Research Engineer,” Ahmed Ragab, Machine Learning Research Engineer at Mawdoo3 told Mubasher.

Collaborating with Banks

The Jordan-based startup is expanding in business to business (B2B) services, focusing on commercializing Natural Language Processing (NLP) and AI services, especially for the banking sector, according to Hussien Al Natsheh, Chief AI Officer at Mawdoo3.

Banking is a closed domain, in which there are certain frequently asked questions and orders, such as balance inquiry, bill payment, and ATM locator,” Al Natsheh explained.

“Privacy and security are two main challenges that we face since the banking industry has to comply with certain rules and regulations. In my opinion, the most important thing is trust, which is enhanced through non-disclosure agreements (NDAs) and confidentiality agreements.

“We are operating as a post-authentication voice interface to users, so everything is safe and secured, but instead of pressing buttons, you just utter your request, and all data is anonymized to protect privacy,” Al Natsheh added.

Expanding in the Region

The Chief AI Officer additionally revealed to Mubasher that in January 2020, the company will announce its first bank client in Jordan. “We will develop the voice assistance services for a Jordanian bank. It is a big project that would take around 12 months to be completed. The agreement has been finalized, but the deal has not been announced yet.”

Moreover, the Amman-based company is in talks with a bank in Iraq, and will soon inaugurate its office in Egypt, where many of its engineers and specialists are from. “We look forward to partner with the government and banks in Egypt once Salma strongly covers the Egyptian dialect. We are also establishing in the UAE where we already have one of our teams,” he added.

When asked about other business areas where Salma is seeking partnerships, Al Natsheh mentioned call centres, especially within the telecom sector.

Business Model

“Salma’s competitive edge is the Arabic language. We have the biggest team of AI specialists and Arabic language experts. Our model is still learning more accents and dialects, and we need linguistic specialists to compensate the technical work,” said Farah Al Zu'bi, Business Analyst at Mawdoo3, who works within the Salma Commercial Team that is focusing on how to automate business services and make business-customer interactions more intuitive.

One of the core challenges for a voice assistant is perfecting numerous dialects, especially since the Arabic language has diverse vocabulary and grammatical conjugations that are more complicated than English. This makes the job of search engines harder, Ahmed Ragab noted.

“In each country, there are many dialects, from southern to norther and so on, and to cater to the whole market Salma needs to understand all these variations. We currently have the Levant and Saudi dialects, with Egyptian set to be offered next year,” Al Zu'bi stated.

“Machine learning never stops. It has to keep updating with new solutions and information,” Ragab noted.

Moreover, the machine learning process is improved through different data that enhances Salma’s understanding of dialects and accents, such as customer service call recordings from internet companies, Al Zu'bi added, stressing that such inputs are chosen not to reveal personal details and protect the user privacy.

When asked about accessibility applications for the disabled and visually impaired, who can really benefit from an Arabic voice assistant, Al Zu'bi noted that Salma’s user interface is yet to cater to this segment but remains on the roadmap.


It is worth noting that Mawdoo3 started in 2010 with the goal of enriching Arabic online content. It won first place in the 2011 Queen Rania National Entrepreneurship Competition (QRNEC), and officially launched in 2012, co-founded by Mohammad Jaber and Rami Al Qawasmi.

Other than Salma, Mawdoo3 has various projects, such as the Mawdoo3.com portal; Mowjaz, a news application, and Ujeeb, an Arabic question-and-answer website, like Quora. “Salma is being trained with various inputs to be able to answer the users’ questions about subjects such as world capital cities, currency exchange, or weather conditions,” Abeer Abu Touq, Social Media Manager at Mawdoo3 told Mubasher.

Mawdoo3 is the most visited Arabic website in the Middle East and North Africa (MENA) region with 50 million unique visitors per month, and it has expanded during the past years from two employees in 2012 to around 300 employees now, Rami Al Qawasmi, Co-founder and CEO of Mawdoo3 said during a panel discussion on the third and final day of RiseUp 2019.