Prune or Retrain: Optimizing the Vocabulary of Multilingual Models for Estonian
This work addresses computational cost and performance issues for Estonian language processing, but it is incremental as it builds on existing multilingual models.
The study tackled adapting multilingual language models to Estonian by modifying their vocabulary to improve efficiency and performance on Named Entity Recognition, finding that pruning unused tokens had no negative effects while retraining the tokenizer degraded performance.
Adapting multilingual language models to specific languages can enhance both their efficiency and performance. In this study, we explore how modifying the vocabulary of a multilingual encoder model to better suit the Estonian language affects its downstream performance on the Named Entity Recognition (NER) task. The motivations for adjusting the vocabulary are twofold: practical benefits affecting the computational cost, such as reducing the input sequence length and the model size, and performance enhancements by tailoring the vocabulary to the particular language. We evaluate the effectiveness of two vocabulary adaptation approaches -- retraining the tokenizer and pruning unused tokens -- and assess their impact on the model's performance, particularly after continual training. While retraining the tokenizer degraded the performance of the NER task, suggesting that longer embedding tuning might be needed, we observed no negative effects on pruning.