Discover Le Grand Continent. Ideas and analyses you won’t find anywhere else. But you will see them everywhere soon. In your inbox starting at 8 euros per month
That the French language has spread across the world and that today it is spoken by hundreds of millions of people around the globe is not the result of a single decisive event, but of a series of complex mutations that were not solely the work of the prince. It must be recalled because, in French mythology, everything is indeed assumed to proceed from the prince and his will. Sometimes, when one says “prince,” one also says “edict,” “ordinance” – the Ordinance of Villers-Cotterêts signed in 1539 by François I.
Amending the Narrative
Because he wishes to reform the legal and notarial regime of his realm, François I enacts a law that redefines the respective responsibilities of royal and ecclesiastical courts. It requires notaries to record all births and deaths across the entire kingdom, and guarantees defendants the right to respond personally to the charges during their trial. For the rest, the ordinance requires that all judgments, sentences, contracts, wills, and other legal acts be written in a French language so clear, precise, and devoid of ambiguity and uncertainty that any need for interpretation becomes superfluous.
Even if one regards this ordinance as proof of the monarch’s ability to impose his language on his people, and even if François I opened the path to eradicating regional idioms, Villers-Cotterêts does not necessarily explain how French could impose itself beyond the kingdom’s borders. Forcing subjects to change their idiom is one thing. Carrying this idiom to the far reaches of the earth is another.
Such an effort requires more than kings and emperors, or even soldiers. By themselves, neither the force of arms nor the sovereign’s will suffices to widen the frontiers of a language. It had to be that at some moment ordinary bearers of their language (sailors, poets, artisans, writers, scholars, merchants, teachers, and missionaries) left their homeland, settled elsewhere, temporarily or permanently, and dispersed the language through interactions of all kinds with distant lands and peoples. The mobility of people is therefore a condition of the migration of languages, their circulation, even their planetary spread.
This migration can take diverse forms. It has taken the form, for instance, of colonial conquest. For us, indeed, everything begins with the military occupation of distant territories, the subjugation of the defeated peoples, the seizure of land and its rearrangement in service of metropolitan interests, the ideological grid by means of schooling or the evangelization of pagans, and so on. This partly explains why the debate about the past and future of French in Africa remains dominated by ideological considerations.
A relatively linear narrative would have it that every gain achieved by this language in Africa automatically results in equivalent losses for indigenous languages. Moreover, for many, French is still seen as one of the main causes of African alienation. Francophonie would, for its part, be nothing more than an instrument of dulling the mind through which “Françafrique” would continue to exercise its harmful grip on hearts and minds, in order to better plunder the continent’s resources.
The moment may have arrived to amend this narrative, not to deny the reality of asymmetrical relationships between Africa and the powers of the world, but to better account for the complexity of processes of cultural creativity under domination. First, just like English, Arabic, or Portuguese, French in all its variants has become a language, if not African, at least an Africa-ality—an “African language” in spirit. Going backward is unlikely. Indeed, in contact with the continent, the French language turned out, like indigenous languages, to be plastic, adaptable, and malleable, open to all kinds of grafts.
Beyond the undeniable effects of imposition, Africans did not merely adapt to the French language. They have, so to speak, conquered it. They have measured what they could obtain from it. Some saw in it a measure of their own value and their own power. That is why they adopted it. To adopt is to participate in what one adopts. We have thus moved from an exclusively colonial encounter characterized by brute force to a different kind of relationship—undeniably ambivalent and paradoxical, but whose generativity cannot be denied.
Moreover, if the intrusion of French into the African linguistic space occurred by break-in, it did not entail the systematic death of local languages. Far from restricting their possibilities, this process has enriched them, and likewise enriched the French language. A new ecology of languages has emerged, as have new norms and new modes of use and functioning of languages on both sides. Far from a zero-sum game, the mixing of languages initiated by colonization continues with renewed vigor. With the strength and resilience of indigenous languages assisting, creolization—more than multilingualism—is our future. It feeds on three factors that compel us to rethink our relationship to French and its institutions in a new light.
The New Infrastructures of Language
The first is the refinement of artificial intelligence technologies, aided by the growing power of computational instruments and the existence of immense data bases. This project of full automation of existence through the algorithmic processing of massive data is carried by a few giant enterprises.
French in all its variants has become a language, if not African, at least an African language.
Achille Mbembe
These aim in particular to replace organizations and social institutions with connected objects, services, and digital applications. This is one of the reasons why the technological transformations of the last quarter of the 20th century and the early 21st century have disrupted work, but also the relations between humans and machines, the production of knowledge, and the entire set of activities through which humans exteriorize their being-in-the-world.
Languages themselves have not escaped these mutations. For instance, notable advances have been recorded in natural language processing. Notably, freely accessible neural machine translation or deep learning tools have become available to the public, such as Google Translate or DeepL.
Furthermore, the opportunities to use AI in teaching and learning languages have multiplied. This is visible in the domains of oral expression, writing, and reading. The use of chatbots is increasingly common. As a personal tutor, ChatGPT provides instant feedback on writings, helps correct errors, and offers writing ideas. Automatic evaluation systems, so‑called intelligent tutoring systems, conversational robots, “smart virtual environments,” and affective computing all now participate in the destinies of the major languages and influence how cognitive and linguistic capital is built at a global scale. The integration of generative artificial intelligence into language teaching and learning does not merely transform methods of instruction and learning. It also has a direct impact on the acquisition of linguistic skills and on the language itself—the vocabulary, the syntax, the grammar, the pronunciation, what a text is, what literature is.
To properly gauge the consequences of these upheavals, it must be recalled that languages are not merely tools of communication. They elude, a priori, any mathematical-computational definition and any definitive encoding within a program. In truth, language is what, by unfurling its own wings, enables humans to launch themselves into the vast world, to embrace it, and in so doing, to stay alive. Sharing the same language with others means, for the human subject, becoming part of a common country—the country of the living—and of a common city, the linguistic city. Through language, what each individual produces both in reality and in the imagination nourishes the production of others and makes possible a shared life because it is open to the sharing of meanings.
Every language is, finally, an intellectual production, through which intelligence takes form and circulates. In this sense, it is not merely an activity, but a modality of giving and participating. Because it is truly public and the quintessential symbol of sharing and togetherness, it must escape the games of monetary exchange.
Automation and Commodification
The general ecology of languages is therefore undergoing an irreversible transformation on a global scale. One of the defining features of our time is indeed the emergence of a global market for the language industries. These are increasingly dominated not by states or major public and cultural institutions, but by large corporations. Language is becoming less a marker of identity or an instrument of influence and political sovereignty and more and more a raw material that platform capitalism, in its endless expansion, has rightly seized as a commercial resource.
Sharing with others the same language is, for the human subject, becoming part of a common country—the land of the living—and of a common city, the linguistic city.
Achille Mbembe
The deployment by large corporations of exploitation strategies aimed at turning languages into merchandise like any other is explained above all by the desire to lay hold of humanity’s immense cognitive capital. It is not out of the question that the market for language-processing technologies could, in the medium term, transform into an oligopoly dominated by the major American and Chinese companies whose priority is to extend their grip on online human interactions.
Take, for example, the case of intelligent personal assistants. In 2020, Cortana, a virtual assistant developed by Microsoft and designed to provide voice support on various devices (Windows computers, smartphones, and smart speakers), was available in seven languages dominated by English. It was deployed for natural language processing and machine learning, aiming to understand large masses of language data (whether spoken or textual) and to respond to user commands. To interpret voice commands, Cortana relied on advanced algorithms. It used Microsoft’s cloud services to process information and provide personalized recommendations. Essentially, this was about encoding messages and formalizing language in service of computation.
It is important to note that although English was its dominant language, Cortana did not distinguish between Australian, British, American, Indian, Nigerian, or Jamaican variants of English, nor between Mexican, Argentine, or Spanish variants of Spanish. Meanwhile, Siri’s settings on a Mac computer offered different choices. There were 41 languages, including nine variants of English, four variants of Chinese, but only one variant of Arabic, whereas ISO standards count more than 30. The shift toward neural networks in the 2010s had already made it possible to incorporate greater complexity into language processing and language learning.
Today, these tools boast vastly increased capabilities. To better grasp the meaning of discourse, they can simultaneously integrate speech and gesture. The accelerated hybridization between learning systems and symbolic systems is underway. The multiplication of such devices makes greater circulation of languages online possible.
In turn, this creates, de facto, an unprecedented multilingualism. But while greater efficiency in automatic language processing does enable more content to be disseminated in more languages, it is not the equivalent of polyglossia. Does polyglossia not require levels of competence and degrees of mutual intelligibility among the various languages and their respective variants? What we witness is nothing less than the creation of languages that are not artificial, but compressed and defined above all by indicators of quantity, visibility, and accessibility.
Based mainly in the United States, the largest software publishers first turn to American English, and only subsequently, to other languages corresponding to lucrative markets. Their strategies stem from a complex tangle of distinct logics: identifying markets, data availability, the linguistic knowledge of decision-makers, the costs involved in adding a new language, etc. They also depend on cultural representations and historical stereotypes, including racist and sexist stereotypes. Within a given language, language-processing technologies can thus constrain linguistic diversity.
The continued growth in the number of languages online goes hand in hand with the exponential rise in the need for automatic translation. In 2019, the global translation market was valued at $46.9 billion. It was expected to grow by 19% between 2020 and 2024. The boom in translation demand fuels, in turn, the demand for effective and cheap language-processing tools. It may well be that translation, in the sense used by the strategic language policies of large companies, is limited to providing easier access to content in their mother tongue, and thus paradoxically reduces the motivation to truly learn other languages. If that were the case, linguistic diversity on the Web would be only a veneer of monolingualism rather than genuine multilingualism. Moreover, the kind of multilingualism favored by large companies does not necessarily reflect offline linguistic reality, which remains largely monolingual even though English usage is widespread in spots.
Rapid translation across multiple language pairs has the potential, de facto, to increase global multilingualism. Yet it is not certain that this will guarantee linguistic diversity, let alone multilingualism. From a technical and cognitive point of view, the very logic of neural machine translation rests on a double rupture. First, AI is primarily a technique of information compression, and the design of its tools and their usage is constrained by economic imperatives. It is also constrained by epistemological and cognitive limits—the limits of the training corpus itself, in terms of recurrence, categories and taxonomies, and the presence of biases of all kinds. There is also the fact that by encoding language as mathematical representations, AI seeks to free itself from the constraints of natural languages. Yet linguistic function is far from being the same as the statistical function as deployed in particular algorithms. The result is a phenomenon of loss—or even de-symbolization—that would be dangerous to ignore.
The reservoirs of meaning and symbols necessary for democratic life and the survival of humanity are stored in human languages.
Achille Mbembe
We must finally ask whether multilingualism and linguistic diversity are enriched or impoverished by automatic language processing. In other words, what are the effects of automatic language processing tools on the global landscape of online languages? To answer this question, one must consider the various facets of the technical and economic ecosystem of language industries—the techniques used, the performance indicators, the interaction between humans and machines, the latest market developments, etc. Take, for instance, performance indicators. Today, the evaluation of neural machine translation is done with metrics that privilege non-linguistic criteria, such as user engagement rate or the time users spend on a page or interface. Rather than rendering the source text as faithfully as possible into a target language, the aim is to maximize clicks. In the current technological conjuncture, mutual intelligibility is indeed enhanced by automatic translation, but the latter is not designed to account for intranslatables, for all the nuances and specifics of a given linguistic community.
Let’s Not Do to Languages What We Have Done to Nature
From the reflections above, three conclusions emerge.
As today everything tends to pass through information machines on a planet now populated with mobile devices, the linguistic function is, more than ever, externalized. There no longer exists syntax, grammar, or dictionaries that do not rely on technologies. The very nature of language and its work within culture and the field of symbolic creation are deeply transformed. We must therefore think of the French language and Francophonie not as mere “instruments of influence,” but truly in terms of an open, migratory, ecumenical linguistic city, and in light of the project of re-symbolizing the world, without which substantive democracy scarcely exists.
To do this, we must develop language proficiency levels in both French and mother tongues, because both French and mother tongues could suffer from the omnipresence of linguistic technologies that simplify and standardize both general creation and symbolic expression.
Second, languages are not commodities and they are far more than vehicles of communication. They are a modality of giving. In this sense, they form part of a broad economy, that of the commons, alongside air, water, the biosphere, resources, or living instruments. This is why we must not leave to a handful of dominant economic actors the task of wielding, without accountability, the large-scale diffusion of language-processing technologies. In the spirit of the commons, platforms, educational, cultural, and legal institutions, and user communities must all be stakeholders in the debate on the role and consequences of language processing technologies in our AI-era societies.
Finally, there exists a close link between the ecological crisis, the crisis of meaning, and the ongoing process of language drying up and impoverishment. The reservoirs of meaning and symbols necessary for democratic life and the survival of humanity are stored in human languages. We cannot transplant onto the domain of languages the kind of relationship we have established with nature—one marked by a will to instrumentalize all forms of life on Earth, in addition to an unfaltering faith in the market mechanism. Bringing human languages into the realm of economic value and technological problem-solving, without safeguards and in an abstract, disembodied form, will sooner or later lead to their extinction and to the disappearance of symbolic habitats without which it is impossible to re-symbolize democracy and to write the world.