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

Findings of the Association for Computational Linguistics: EMNLP 2020, pages 2144–2160 November 16 - 20, 2020. c⃝2020 Association for Computational Linguistics 2144 Participatory Research for Low-resourced Machine Translation: A Case Study in African Languages ∀∗, Wilhelmina Nekoto1, Vukosi Marivate2, Tshinondiwa Matsila1, Timi Fasubaa3, Tajudeen Kolawole4, Taiwo Fagbohungbe5, Solomon Oluwole Akinola6, Shamsuddee Hassan Muhammad7,39, Salomon Kabongo4, Salomey Osei4, Sackey Freshia8, Rubungo Andre Niyongabo9, Ricky Macharm10, Perez Ogayo11, Orevaoghene Ahia12, Musie Meressa13, Mofe Adeyemi14, Masabata Mokgesi-Selinga15, Lawrence Okegbemi5, Laura Jane Martinus16, Kolawole Tajudeen4, Kevin Degila17, Kelechi Ogueji12, Kathleen Siminyu18, Julia Kreutzer19, Jason Webster20, Jamiil Toure Ali1, Jade Abbott21, Iroro Orife3, Ignatius Ezeani38, Idris Abdulkabir Dangana23,7, Herman Kamper24, Hady Elsahar25, Goodness Duru26, Ghollah Kioko27, Espoir Murhabazi1, Elan van Biljon12,24, Daniel Whitenack28, Christopher Onyefuluchi29, Chris Emezue40, Bonaventure Dossou31, Blessing Sibanda32, Blessing Itoro Bassey4, Ayodele Olabiyi33, Arshath Ramkilowan34, Alp ¨Oktem35, Adewale Akinfaderin36, Abdallah Bashir37 ∗Masakhane, Africa 1Independent, 2University of Pretoria, 3Niger-Volta LTI, 4African Masters in Machine Intelligence, 5Federal University of Technology, Akure, 6University of Johannesburg, 7Bayero University, Kano, 8Jomo Kenyatta University of Agriculture and Technology, 9UESTC, 10Siseng Consulting Ltd, 11African Leadership University, 12InstaDeep Ltd, 13Sapienza University of Rome, 14Udacity, 15Parliament, Republic of South Africa, 16Explore Data Science Academy, 17UCD, Konta, 18AI4Dev, 19Google Research, 20Percept, 21Retro Rabbit, 23Di-Hub, 24Stellenbosch University, 25Naver Labs Europe, 26Retina, AI, 27Lori Systems, 28SIL International, 29Federal College of Dental Technology and Therapy, Enugu, 31Jacobs University, 32Namibia University of Science and Technology, 33Data Science Nigeria, 34Praekelt Consulting, 35Translators without Borders, 36Amazon, 37Max Planck Institute for Informartics, University of Saarland, 38Lancaster University, 39University of Porto, 40Technical University of Munich masakhane-nlp@googlegroups.com Abstract Research in NLP lacks geographic diver- sity, and the question of how NLP can be scaled to low-resourced languages has not yet been adequately solved. “Low- resourced”-ness is a complex problem go- ing beyond data availability and reflects systemic problems in society. In this paper, we focus on the task of Ma- chine Translation (MT), that plays a cru- cial role for information accessibility and communication worldwide. Despite im- mense improvements in MT over the past decade, MT is centered around a few high- resourced languages. ∗∀ to represent the whole Masakhane community. As MT researchers cannot solve the prob- lem of low-resourcedness alone, we pro- pose participatory research as a means to involve all necessary agents required in the MT development process. We demon- strate the feasibility and scalability of par- ticipatory research with a case study on MT for African languages. Its imple- mentation leads to a collection of novel translation datasets, MT benchmarks for over 30 languages, with human evalua- tions for a third of them, and enables par- ticipants without formal training to make a unique scientific contribution. Bench- marks, models, data, code, and evaluation results are released at https://github. com/masakhane-io/masakhane-mt. 2145 1 Introduction Language prevalence in societies is directly bound to the people and places that speak this language. Consequently, resource-scarce lan- guages in an NLP context reflect the resource scarcity in the society from which the speak- ers originate (McCarthy, 2017). Through the lens of a machine …

Content truncated to 500 words