Using Artificial Intelligence to Read Arabic Comics
Edward Fox / 11 Mar 2018
BEIRUT—A computer scientist at the American University of Beirut is using artificial intelligence to classify the content of Arabic comics, applying the computer-based science to this cutting-edge art form in the Arab world.
Artificial-intelligence specialists are always trying to stretch the capabilities of computer brainpower. If artificial intelligence can be used to play the ancient Chinese board game Go, or the American TV quiz game Jeopardy, then Arabic comics are also fair game.
“I try to look for unusual applications for artificial intelligence and machine learning,” explained Mariette Awad, the associate professor in the department of electrical and computer engineering at the American University of Beirut who is leading the project.
“I wanted to move away from the mechanical or routine tasks that you associate with artificial intelligence and look at the level of sentiments, at the level of comprehension, processes which are usually associated with the higher cognitive faculties of human beings,” Awad said.
Applying artificial intelligence to Arabic comics, she said, will help the science adapt to Arabic culture and language and make other regional applications possible.
With their rich mixture of word and image, Arabic comics have in recent years grown into a thriving creative movement, telling stories of contemporary life in Arab countries with compelling imagination, honesty and graphic quality. Each creator of an Arabic comic invents their own way of using sequences of words and images to tell a story, and to convey subtleties of feeling and the texture of human experience. (See a related article, “Arab Comics: Fit for Academic Exploration.”)
“There is a wealth of information in the new Arabic comics,” Awad said. “Just as you would study a civilization by its literature, its painting or its music, so we should look at comics as a form of communication. It tells us a lot. These comics mirror what society in Middle Eastern countries was seeing in terms of oppression, political movements, political structure and gender issues. They give us a history and a view of the society.”
Awad and her colleagues used digitized samples of Arabic comics held in the collection of the AUB library as part of the Mu’taz and Rada Sawwaf Arabic Comics Initiative. They used scanned copies of comics from all over the Arab region, but mostly from the main publishing centers in Iraq, Egypt, Syria and Lebanon.
Awad worked with Lina Ghaibeh, who is director of the comics initiative at the university and also a graphic artist, and with Ghaibeh’s colleague George Khoury to create a searchable database.
To construct a system that could “read” the material, the researchers used a combination of artificial-intelligence processes to identify visual elements such as color and shape.
Awad’s innovation was to devise a process for handling the Arabic text in the comics.
“Can we use machine learning to understand the framework of Arabic? No one has done that,” she said. The Arabic language’s rules of grammar and syntax required a new technique in the branch of artificial intelligence called natural-language processing.
The new technique identifies the root consonants of Arabic words. Most Arabic verbs and nouns are built on a pattern of three consonants; this three-letter root is the kernel of meaning from which verbs and nouns of related meaning are derived.
The resulting artificial-intelligence engine looks for different forms and conjugations of Arabic words, but also makes connections of meaning. “A typical Arabic word has numerous levels of meaning,” Awad said. “There are very deep levels of association between words.” That is, the process can identify nuances of meaning.
Jonathan Guyer, a fellow at the Radcliffe Institute for Advanced Study at Harvard University who is writing a book about Arabic comics, said the state of the art of Arabic comics requires a multidisciplinary approach to begin to capture these subtleties.
“Whether you have a political scientist writing about Arabic comics in an academic journal, or a scientist deploying artificial intelligence, the result will always be imprecise,” Guyer said.
Humor, for instance, is notoriously resistant to analysis. A lot of the humor in Arabic comics is visual humor, Guyer said. “There is a certain brand of black humor, dark humor, gruesome gallows humor.”
In a comic like Tok Tok (an Egyptian comics magazine that is probably the best known publication in the genre) for instance, “there is a lot of morbid joking that would be unseemly to some audiences,” Guyer said.
Mariette Awad’s previous project in artificial intelligence involved analyzing photos posted on social media to derive information that could be used to facilitate timely humanitarian responses to war damage and natural disasters.
She is now working on an application of the Arabic natural language processing system to analyze social media. “This could be used as a tool to measure public opinion,” she said. “In other words, the voice of the Arab street.”