Unknown Script: Impact of Script on Cross-Lingual Transfer

TL;DR

This paper investigates how the choice of tokenizer influences cross-lingual transfer performance, especially when the target language's script is not included in the pre-trained model, highlighting tokenizer importance over script similarity.

Contribution

It demonstrates that tokenizer choice significantly impacts cross-lingual transfer, surpassing script similarity and language relatedness, in models with unseen scripts.

Findings

Tokenizer choice is a key factor in transfer performance.

Shared script is less influential than tokenizer.

Model size has minimal effect when tokenizer is optimized.

Abstract

Cross-lingual transfer has become an effective way of transferring knowledge between languages. In this paper, we explore an often overlooked aspect in this domain: the influence of the source language of a language model on language transfer performance. We consider a case where the target language and its script are not part of the pre-trained model. We conduct a series of experiments on monolingual and multilingual models that are pre-trained on different tokenization methods to determine factors that affect cross-lingual transfer to a new language with a unique script. Our findings reveal the importance of the tokenizer as a stronger factor than the shared script, language similarity, and model size.