Integrating diverse corpora for training an endangered language machine translation system
Amis–English and Amis–Mandarin translation using diverse corpora, synthetic English data, OpenNMT, and NLLB. Synthetic data helps when translating into English; effects vary by direction and model.
Paper PDF ACL Anthology Recorded talk Translation demo (restricted)
Citation
Hunter Scheppat, Joshua Hartshorne, Dylan Leddy, Éric Le Ferrand, and Emily Prud’hommeaux. 2025. “Integrating diverse corpora for training an endangered language machine translation system.” Proceedings of ComputEL-8, pp. 162–169.