2020/09/30 by Chantal Amrhein, Rico Sennrich, Amrhein, Chantal +1 · 1 citation
Computer Science · #Computation and Language (cs.CL) #FOS: Computer and information sciences #Natural Language Processing Techniques #Speech Recognition and Synthesis #Topic Modeling
paper · pdf · doi:10.48550/arxiv.2009.14824
openalex publication_date 2020/09/30 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
Transfer learning is a popular strategy to improve the quality of\nlow-resource machine translation. For an optimal transfer of the embedding\nlayer, the child and parent model should share a substantial part of the\nvocabulary. This is not the case when transferring to languages with a\ndifferent script. We explore the benefit of romanization in this scenario. Our\nresults show that romanization entails information loss and is thus not always\nsuperior to simpler vocabulary transfer methods, but can improve the transfer\nbetween related languages with different scripts. We compare two romanization\ntools and find that they exhibit different degrees of information loss, which\naffects translation quality. Finally, we extend romanization to the target\nside, showing that this can be a successful strategy when coupled with a simple\nderomanization model.\n