Improving Zero-shot Cross-lingual Transfer between Closely Related Languages by injecting Character-level Noise

Abstract

Cross-lingual transfer between a high-resource language and its dialects or closely related language varieties should be facilitated by their similarity, but current approaches that operate in the embedding space do not take surface similarity into account. In this work, we present a simple yet effective strategy to improve cross-lingual transfer between closely related varieties by augmenting the data of the high-resource parent language with characterlevel noise to make the model more robust towards spelling variations. Our strategy shows consistent improvements over several languages and tasks: Zero-shot transfer of POS tagging and topic identification between language varieties from the Germanic, Uralic, and Romance language genera. Our work provides evidence for the usefulness of simple surface-level noise in improving transfer between language varieties.

Subword Evenness (SuE) as a Predictor of Cross-lingual Transfer to Low-resource Languages

“…For example, our sample of target languages does not include any Indo-European languages, such as Germanic or Romance low-resource languages. These languages have been studied before and it has been shown that the best choice for them is transferring from a genealogically related rich-resource language (Aepli and Sennrich, 2021). It might be interesting to see how our proposed measure would compare with other measures in these cases, but this would require a different study design, which we leave for future work.…”