Tokenization and Morphology in Multilingual Language Models: A Comparative Analysis of mT5 and ByT5

TL;DR

The impact of tokenization is captured by contrasting two multilingual language models: mT5 and ByT5 by showing that the models learn the morphological systems of some languages better than others and that morphological information is encoded in the middle and late layers.

Abstract

Morphology is a crucial factor for multilingual language modeling as it poses direct challenges for tokenization. Here, we seek to understand how tokenization influences the morphological knowledge encoded in multilingual language models. Specifically, we capture the impact of tokenization by contrasting two multilingual language models: mT5 and ByT5. The two models share the same architecture, training objective, and training data and only differ in their tokenization strategies: subword tokenization vs.\@ character-level tokenization. Probing the morphological knowledge encoded in these models on four tasks and 17 languages, our analyses show that the models learn the morphological systems of some languages better than others and that morphological information is encoded in the middle and late layers. Finally, we show that languages with more irregularities benefit more from having a higher share of the pre-training data.

Figures (2)

• Figure 1: Overview of our experimental procedure
• Figure 2: Probing accuracy of ByT5 (left) and mT5 (right) across layers grouped by languages and tasks. Each line represents a language (top) or a task (bottom). Each data point is the accuracy scores at each layer of each language.