Language
RobotsMali AI4D Lab is a Bamako-based pedagogical research laboratory focused on developing uses of AI to further the social and economic development of Mali, working on Malian language NLP, educational interventions, the arts, and entrepreneurship. For most of its existence, RobotsMali has been funded primarily by the volunteer labor of its team, though recently Canada’s magnificent IDRC has pitched in with a bit of funding to enable structural support for its activities.
The Linguistic Landscape: Africa, Colonial Languages and Bambara
Until Elias Lönnrot* and the Kalevala, Finnish was not the fully standardized, high-prestige, broadly used written national language it is today. Rarely has a single person shaped a modern language so deeply. Many African languages still wait for such a figure. For Bambara in Mali, AI could be that Lönnrot.
Malians resisted French colonial rule for 80 years in a profound but generally ignored way: the vast majority in this so-called francophone country refused French language and education. Most Malians do not speak French and even among the minority that do, French plays no role in their affective life. That is not to say that the French language does not have an important role in Mali; it is, for all practical purposes, its sole written language and portal to technological civilisation.
In fact, while all written languages are diglossic in relation to their spoken forms, Mali is an extreme case where the written and spoken languages are entirely different languages. We use the term predominantly-oral to describe a very large group of languages in Africa that have a more or less consistent written form, but remain little written because their speakers have ceded the domain of literacy to the colonial languages. While there are exceptions, across Sub-Saharan Africa, similar patterns exist to varying degrees. The extent to which colonial languages have been adopted also varies across the continent. Even where resistance to colonial languages has been less pronounced, most Africans maintain a primary, affective oral language that colonialism has not erased.
We use the designations medium-literacy, low-literacy, and negligible-literacy in speaking about the extent to which literacy has permeated African societies. Only a handful of African languages exist in medium-literacy contexts where a significant minority can read and write fluently and there is a body of literary and functional texts.
A nation must have an educated population to hold its own in the modern world and not to be reduced to global serfhood
Mali has 13 national languages at lower levels of literacy. National languages, not local. Bambara is in the group of mutually-intelligible Manding languages spoken by 40 to 50 million people. Although it’s the most widely-spoken language in Mali, it is low-literacy: the writing system is standardized, and a small number of texts exist, but few people can read the language, and even fewer can write it. Malian national languages are part of regional language networks spoken by 200 to 250 million people across an area larger than Western Europe. The often-cited figure of “2,000 African languages” suggests fragmentation. But every region of the world once had thousands of language varieties before literacy and state formation consolidated them into standardized forms. Africa is no different.
Why Teaching in Bambara Matters
In Africa, it has been considered normal to subject children from their first year of schooling to instruction in a language that they do not speak, presented in a context making no reference to their cultural and social environment. Lack of resources is often given as the primary explanation for the low success rates of African educational systems. Yet, many countries outside of Africa have overcome resource challenges, while no country has built a strong education system on the institutionalised exclusion of children’s language and culture. In Mali, and in many other African countries, a movement has long existed to teach children in the language that they speak. There have been efforts, but there are few demonstrations of success. There hasn’t been an honest assessment of the underlying challenge: you can’t teach children book learning in a language that has few or no books.
For every language with a rich literature, there was a time when no books existed in that language. Historically, this challenge can be overcome, but it requires time and a stimulus—technological, religious, or political. The natural evolution of African languages toward writing may have been slowed by the diglossic use of colonial languages. Nevertheless, developing written forms remains the essential first step for transforming African education.
We are a Malian laboratory working to apply the seismic stimulus of AI to African development. We have focused on education—book learning—on the principle that a nation must have an educated population to hold its own in the modern world and not to be reduced to global serfhood. Education drew us to the problem of language and language to the problem of literacy. We saw an opportunity, using AI, to accelerate the realisation of a national ambition, ensconced in Mali’s new constitution: to use our languages in education and all other facets of life, eventually consigning the colonial language to its proper place as one foreign language among others.
Understanding is easier in the mother tongue
From Text to Image: AI Meets Malian Language and Culture
We used AI in the creation of children’s books, mostly in Bambara, a large addition of over 100 titles to what had been a scant number of existing works. Not the AI-generated pablum that has become omnipresent, but tales crafted by writers aided by AI and a pipeline process to enable the production of high-quality, culturally-specific, Malian-child friendly stories in a language that is barely written and lacks compositional models. When we began, there was no LLM capable of interacting in Malian languages; our writers worked in French to develop texts expressly designed for translation to Bambara. This proved more challenging than one might imagine, so tightly interwoven are the ways that one says things in French with elements of French culture that are incomprehensible for Malian children.
Bambara had made its way into Google and other AI translators, and these tools were able to complete 80 to 90 percent of the translation. That still left substantial correction and replacement with idiomatic language. Yet automated translation did more than handle most of the text: it provided structural scaffolding that would otherwise have required extensive experimentation and revision by human translators with limited experience writing in Bambara.
The books were richly illustrated using text-to-image AI generation, the collection producing about 900 images, painstakingly worked by illustrators to ensure accurate depictions of the physical environment of Mali, Malian dress and physiognomy, character consistency, beauty, child-friendliness and general delightfulness.
As with producing the source texts in French, it proved extremely difficult to get the tools to generate images that were not European caricatures of Africa, or to produce elements such as hairstyles and clothing that bore any real resemblance to what one sees in Mali. One often speaks of bias in the data used to train AI systems. This is, of course, a significant problem for Africa. There are more European caricatures of Africa on the internet than actual images from Africa, and generative AI produces according to its training. Humans too, though. In fact, we spent considerable time examining storybooks published in Africa, mainly in colonial languages, and came away with the impression that human illustrators follow caricatural conventions almost as readily as AI does.
Building Stories in Bambara: The Children’s Book Project
We set out to create literature—and we believe we succeeded to a significant degree: works artfully constructed, layered in meaning, stimulating imagination, language development, and moral and even scientific reflection, while remaining appropriate to the levels of understanding of Malian children across different age groups.
We did not, however, confine ourselves to the themes and narrative devices of African oral tradition. We felt that Malian children had the right to a modern literature: stories set in urban environments and shaped by contemporary concerns; fantasy and science; humor; Malian adaptations of world literature; the worlds of work and childhood; family dynamics; even, ever so lightly, politics. This has not been without controversy. The same people who argue that our languages must be preserved often insist that Tales from Grandmother are the only legitimate subject matter for African children’s literature.
We were not aiming at creating school curricula, but we nonetheless wanted to support teachers who would have the courage to use literature in language arts classes. Generative AI excels at analysing texts so we fed our stories back into the beast and produced multiple-choice vocabulary and comprehension questions accompanying every story and Teacher’s Guides with exercises, games, and thematic exploration to guide classroom discussion.
What is the advantage of using AI—and, in using AI, aren’t we taking work away from writers, illustrators, and educators? While we agree that our books could have been produced by humans unaided by AI, this had not been done in Mali. In creating a literature where there was virtually none, we created a market—and we taught Malian writers, illustrators, and educators to do so in the way that books will be developed everywhere in the age of AI. We solved nearly impossible problems: forging a literary language, producing a volume of literature that could have taken a decade to assemble, if it were produced at all, and achieving a level of depth and quality that is rarely found in African children’s literature.
Technology changes every second, but we tried to use technology to create something that would endure 100 years
What Works, What’s Next
The education community wants evidence-based interventions, meaning assessment methodologies derived from literate cultures that are ill-equipped to measure qualities such as cultural resonance or … delight. The AI community wants novel machine learning techniques that yield improvements in precision scores. The AI-for-Education community wants… well … apps.
So we built apps—five of them. The full collection of books is now available on smartphones; there is a reading tutor with an AI voice model that provides pronunciation feedback; and an app designed for early reading initiation. We have gathered evidence using formal assessment methodologies. We have measured and achieved improvements in precision.
What we have not been able to ignite—largely for lack of resources—is the cultural revolution that would place these tools in the hands of creators. The real challenge of producing literature in a low-literacy, predominantly oral language remains poorly understood by the technical communities. The idea of using AI to support cultural evolution has, so far, proved difficult to fund.
Yet, it works. We’ve seen many hundreds of kids learn to read and to read their first, second and fifth book in Bambara, with obvious delight … and pride. Technology changes every second, but we tried to use technology to create something that would endure 100 years. Lönnrot used immense creativity and the tools of his time to make Finnish a modern national language; we are working with the tools of our time, neural networks and massive stores of human knowledge, to do the same for African languages.
On the Field: a Conversation with Banafa Doumbia
(Baga, Commune of Dialakoroba)
This article presents insights from an interview with Banafa Doumbia, a CED educator from Baga, Mali. The fieldwork, adaptation of questions, translation, and video production were carried out by Mamadou Dembélé, part of RobotsMali. Conducted in Bambara and translated into English via French by Michael Leventhal also part of RobotsMali. The conversation captures the realities of teaching and learning in the region.
Banafa Doumbia is a CED (Centre of Education for Development) educator in Baga, Commune of Dialakoroba.
“I am from Baga—my father is from here and my mother as well.
I want to start on one important point: there are many people who do not know French, yet today they manage to read and write in Bambara thanks to a program that the former DNAFLA (National Agency for Functional Literacy, mutated into the current National Agency for Non-Formal Education and National Languages) had. Personally, I think we should not abandon our mother tongue, whatever the influence of foreign languages may be. If the majority are equipped with functional literacy through the mother tongue, it is a light against great darkness. That’s why we must promote the Bambara language, also in education.”
He presented the current situation of education in Mali.
“In my opinion, children like to study—every child you meet, you find that they like thinking, they like studying. That being the case, if our society would truly place a high value on education, we will really move things forward. The challenge is that everyone must have perseverance. When you are an educator, you must be prepared to work in an environment that does not value your role. When you are a learner, it is the same, your family may not support you. The social difficulties are especially acute for girls. Their mothers expect them to spend most of their time helping with household tasks.”
These challenges also affect teaching.
“Teaching, like everything, has its difficulties. A teacher must be equipped with patience. Not everyone has the ability to teach. For example, there were 3 special cases in my class. There was a deaf child; when they tried to chase him away, I said no. There was a girl who stuttered very strongly; the students started to make fun of her, but I forbade it. A teacher must find solutions so that these kinds of people do not become discouraged. You must be careful about how you correct children—even if what they did is not good, you must correct them without humiliating them, and with encouragement.”
And the language you choose for teaching plays a role.
“The difference between learning in Bamara and in French is that Bambara is our mother tongue. The main observation is that even when regular [French-language] school students have no classes, when they come and I invite them to enter my classroom, they refuse. Why? If they enter and find us doing math, for example, they disappear one by one because they know they cannot keep up with my students. That shows that understanding is easier in the mother tongue.”
Banafa sees technology increasingly shaping learning.
“During our training, we saw several approaches, including techniques for progressively learning sigini, kanɲɛ, daɲɛ, kumasen (letters, syllables, words, sentences), but also learning the days of the week, everyday objects, etc. But your [RobotsMali] visit the other day to introduce new techniques with new technologies left its mark here. I felt surpassed by the children because they are more comfortable with touchscreen phone tools than I am. For us, that is what needs to be added to learning methods.
If this is put in place widely, I wouldn’t say that there would be no illiteracy in Mali, but it would be very little. It would even motivate adults to learn. In fact, if your apps are added to our teaching methods, you will see the number of people who drop out and abandon education decrease. Since you left here last time, the children keep asking when you will return; I answer them: it is God who decides.
Once they get home the children can work quite a lot with the technological tools. The children were so fascinated by how you can learn letters, move to syllables, then to words, and simply like that. I think that with these tools, within 3 years you will see our language come back to life.”
But it’s not that easy to find a balance between tradition and innovation.
“‘If a farmer finds a new daba [Malian hoe] in addition to the old one, it only multiplies the means and makes plowing easier.‘ To achieve the best outcome, adding these tools should not make us forget traditional methods: when you add these tools to learning, you will see it expand even more. As a teacher, we alternate the methods: sometimes we will work with chalk and the board, other times with notebooks, and other times with technological tools.
However, to evaluate the effectiveness of these tools you still need traditional assessment. For example, at some point you can say: turn off the phones. As a teacher, you can ask: who can come write this letter, syllable, or word on the board? A volunteer comes to write, then you ask the class again whether it is correctly written; if not, who will come correct it. If they have difficulties with a word, you write it on the board and then ask them to go look again at the word on the phone.
If we have tools like these, before making them available to students, we must establish rules and conditions. Sometimes it is necessary to visit the children at home in the village. If you see a child using the phone for something other than learning, you must confiscate it. Solicit the collaboration of the parents.”
Because students, teachers and the schooling community must remain central to the learning process.
“ It is the children’s ability that makes it so that even if you don’t feel like it, you still teach classes. If it were a day when we were supposed to have class and I did not teach, you would see the children crying, begging me to teach. They ask me, when they go home and their parents check their notebooks, what will they say [when there is nothing there]? I sometimes feel sorry for them, so even if I don’t feel like working, I work. As an adult, I learned something: children want to learn, and I have the obligation to pass things on to them with the means available.
As we say in Bambara: “Fɛn dɔ jɛnkɛna a ma bɔn” — “A pot slipping does not mean all its contents have been spilled.” In other words, even if progress has been slow or partial, it doesn’t mean all is lost. If you put a solid system in place, you will see people eagerly embracing the chance to learn in their own languages.”
Michael Levanthal
Founder and Principal Investigator at RobotsMali
After failing at becoming a philosopher, poet, and linguist, Michael Leventhal found middling success in computing, working on knowledge representation, core internet technologies, and novel computing architectures. As one does, he abandoned the fevered plentitude of Silicon Valley, bags laden with tech lore, to work alongside bright and ambitious young Malians in Sahelian-dust-choked fields of opportunity, as they strive to win their nation a place of honor on the world’s stage. RobotsMali AI4D Lab is the fruit of our toil.