AllSpeak: A Language-Agnostic Runtime for Computational Literacy in Multilingual Communities

Language is the foundation of human development. It is what enables us to convey ideas, share knowledge, and build on the discoveries of those who came before us. Since the industrial age, we have extended language into new forms—programming languages that instruct and control the machines we build to serve us. Software engineering, expressed in these languages, now underpins the whole of modern society. We are now creating new forms of intelligence, expected before long to transform every part of our existence. If humans are to retain a relevant role, we must learn to interact usefully with these entities. The traditional programming languages cannot serve this purpose alone: they exclude most of the world's population from the process, because they are expressed in English and designed for human authors, not human readers. AllSpeak is an open-source runtime designed to be a lingua franca—a common language comprehensible to both humans and AI, available in any human tongue. Its programs are intentionally simple, constrained, and readable, not because the problems they address are trivial, but because legibility is the design goal that makes human oversight possible. A codebase that no human can meaningfully inspect is not a managed system; it is an autonomous one, operating on trust. In high-stakes domains—automation, finance, health, infrastructure—that is a governance problem as much as a technical one. This paper describes AllSpeak's architecture, its current implementation in four languages, and the methodology developed for extending it to new linguistic communities. It argues that as AI-generated code becomes ubiquitous, the humans who need to understand it will be a different population from those who once wrote it—less technically trained, more linguistically diverse—and that the tools available to them should reflect that reality.