Abstract
Computer science education has promised open access around the world, butaccess is largely determined by what human language you speak. As youngerstudents learn computer science it is less appropriate to assume that theyshould learn English beforehand. To that end we present CodeInternational, thefirst tool to translate code between human languages. To develop a theory ofnon-English code, and inform our translation decisions, we conduct a study ofpublic code repositories on GitHub. The study is to the best of our knowledgethe first on human-language in code and covers 2.9 million Java repositories.To demonstrate CodeInternational's educational utility, we build an interactiveversion of the popular English-language Karel reader and translate it into 100spoken languages. Our translations have already been used in classrooms aroundthe world, and represent a first step in an important open CS-educationproblem.