Participatory Research for Low-resourced Machine Translation: A Case Study in African Languages

Research in NLP lacks geographic diversity, and the question of how NLP can\nbe scaled to low-resourced languages has not yet been adequately solved.\n"Low-resourced"-ness is a complex problem going beyond data availability and\nreflects systemic problems in society. In this paper, we focus on the task of\nMachine Translation (MT), that plays a crucial role for information\naccessibility and communication worldwide. Despite immense improvements in MT\nover the past decade, MT is centered around a few high-resourced languages. As\nMT researchers cannot solve the problem of low-resourcedness alone, we propose\nparticipatory research as a means to involve all necessary agents required in\nthe MT development process. We demonstrate the feasibility and scalability of\nparticipatory research with a case study on MT for African languages. Its\nimplementation leads to a collection of novel translation datasets, MT\nbenchmarks for over 30 languages, with human evaluations for a third of them,\nand enables participants without formal training to make a unique scientific\ncontribution. Benchmarks, models, data, code, and evaluation results are\nreleased under https://github.com/masakhane-io/masakhane-mt.\n