The first Doha forum .. Multiple dialects of the Arabic language an obstacle to artificial intelligence

Emad Mourad - Doha

The first annual forum of the Qatar Institute for Computing Research, University of Hamad Bin Khalifa, highlighted the development of artificial intelligence in Arabic in the field of acoustics, with the aim of enhancing the prospects and challenges through training the system on the mechanisms in Arabic.

The research group that participated in the forum, which included international companies and universities, focused on three axes, the first is to identify the sound and convert the text into a written, the second to convert the written text to the operative, and the third relates to the recognition of the various dialects of the Arabic language.

The participants agreed that the use of artificial intelligence in the Arabic language faces great challenges, compared to English, since Arabic is difficult in the absence of sufficient information, as well as the multitude of dialects of the Arabic language from one country to another and from one city to another in the same country.

The expert in the Department of Arabic Language Technology at the Qatar Institute for Computer Research Ahmed Ali, said that artificial intelligence depends on the advance information is taught and trained by the system, in order to benefit from the huge development in artificial intelligence and machine learning, and gain experience in this area.

He explained that training the system on a thousand hours of voice in Arabic and a thousand hours of writing, makes us reach the accuracy of 10%, which is the technology used by Al Jazeera Net over the past five years, which converts the written text to read in all news reports.

Ali said in a statement to Al Jazeera Net that the institute is now working on techniques of dealing with artificial intelligence with different Arabic dialects, such as Egyptian, Moroccan, Gulf and Shamia, pointing out that the Arabic language is difficult, in English if you have in the system 5%, of Words do not exist or have not been heard before, the search is in about thirty to sixty thousand words, while the number in the Arabic language may reach one million words.

He added that the challenge is greater in the case of the dialects of the different Arabic language, in the absence of dictionaries or an agreement on information or mechanism on the way to write, and most of the dialects of the Arabic language spoken and unwritten, we do not find a special writing in the dialect Egyptian or Gulf and others, The situation began to change in the presence of new mechanisms such as Twitter, Facebook and others.

The seminar included lectures on the techniques of dealing with artificial intelligence, including the lecture "Understanding emotion and emotion analysis based on speech," which concern service companies in the knowledge of customer satisfaction, as well as the application of health cases that recognize through the sound to the psychological state of the patient, Age, and how happy and sad.

In his turn, the researcher at Columbia University, Rami Iskander, considered that artificial intelligence techniques in the treatment of the Arabic language is very difficult, with limited and insufficient information began to work about 15 years ago, contrary to the information in English or French spread and available to all sixty years ago .

Iskander explained to Al-Jazeera Net that the biggest challenge is in colloquial Arabic because it does not have a unified method of writing. One word is written in more than one way, which leads to overlapping and problems of writing that the system does not understand.

He added that the multiplicity of the dialects of the Arabic language and their differences at the level of countries and cities increases the challenges, which requires a system that is highly efficient to deal with each different dialect, in contrast to the English language, which has few dialects, except English, American, Australian, Not in the form in Arabic, but the way of writing English is almost uniform unlike Arabic.

Among the lectures that received a great response, a lecture by Abdul Rahman Mohammed from Facebook, which reviewed the technology reached by the research laboratory in Facebook and the field of voice recognition.

"This technology has been successfully applied in the field of interpreting English," he said. "This technology can be applied to any other language to the same degree of success.

He revealed that the basic idea in this technology is the ability to access the relations between words and voices apart from some, as well as the ability to know sound, considering that this technology is built on the so-called neural network, systems designed to mimic the human brain in its internal work as part So-called deep learning.

"This technology offers a new approach that is better in terms of speed and efficiency. It also recognizes the sound without being associated with a particular language. It also reduces the mistakes in the English language from 15 to 12 percent," he said.