New Publication: One Model to Pronounce Them All

In May 2020, Kaili Vesik submitted a deep-learning model to a grapheme-to-phoneme conversion task as part of the 17th SIGMORPHON (ACL Special Interest Group on Computational Morphology and Phonology) workshop.This model takes as input orthography from any of fifteen unrelated languages and produces its IPA transcription, with a word error rate of 14.99%. The associated conference proceedings paper, co-authored with Dr. Muhammad Abdul-Mageed and Dr. Miikka Silfverberg, is published in the ACL 2020 anthology. To read this publication check out the lab research and publications.