Improving Multilingual Models with Language-Clustered Vocabularies
This addresses the challenge of vocabulary design for massively multilingual applications, offering a practical improvement for NLP practitioners working with diverse languages.
The paper tackles the problem of suboptimal vocabularies in multilingual models by introducing a language-clustered vocabulary generation method, resulting in improvements such as +2.9 F1 on TyDi QA and an 8x reduction in out-of-vocabulary rate without increasing model size or data.
State-of-the-art multilingual models depend on vocabularies that cover all of the languages the model will expect to see at inference time, but the standard methods for generating those vocabularies are not ideal for massively multilingual applications. In this work, we introduce a novel procedure for multilingual vocabulary generation that combines the separately trained vocabularies of several automatically derived language clusters, thus balancing the trade-off between cross-lingual subword sharing and language-specific vocabularies. Our experiments show improvements across languages on key multilingual benchmark tasks TyDi QA (+2.9 F1), XNLI (+2.1\%), and WikiAnn NER (+2.8 F1) and factor of 8 reduction in out-of-vocabulary rate, all without increasing the size of the model or data.