Abstract
Natural language processing (NLP) systems have become a central technology incommunication, education, medicine, artificial intelligence, and many otherdomains of research and development. While the performance of NLP methods hasgrown enormously over the last decade, this progress has been restricted to aminuscule subset of the world's 6,500 languages. We introduce a framework forestimating the global utility of language technologies as revealed in acomprehensive snapshot of recent publications in NLP. Our analyses involve thefield at large, but also more in-depth studies on both user-facing technologies(machine translation, language understanding, question answering,text-to-speech synthesis) as well as more linguistic NLP tasks (dependencyparsing, morphological inflection). In the process, we (1) quantify disparitiesin the current state of NLP research, (2) explore some of its associatedsocietal and academic factors, and (3) produce tailored recommendations forevidence-based policy making aimed at promoting more global and equitablelanguage technologies.