Language Model Tokenizers Introduce Unfairness Between Languages

TL;DR

This paper reveals that current multilingual tokenizers cause significant disparities in tokenization lengths across languages, leading to unfair treatment and suggesting the need for fairer multilingual tokenization methods.

Contribution

It identifies and analyzes disparities in tokenization lengths across languages caused by current tokenizers, advocating for the development of fairer multilingual tokenization techniques.

Findings

Tokenization length disparities can be up to 15 times between languages.

Disparities persist even with multilingual-trained tokenizers.

Character and byte-level models also show significant length differences.

Abstract

Recent language models have shown impressive multilingual performance, even when not explicitly trained for it. Despite this, there are concerns about the quality of their outputs across different languages. In this paper, we show how disparity in the treatment of different languages arises at the tokenization stage, well before a model is even invoked. The same text translated into different languages can have drastically different tokenization lengths, with differences up to 15 times in some cases. These disparities persist even for tokenizers that are intentionally trained for multilingual support. Character-level and byte-level models also exhibit over 4 times the difference in the encoding length for some language pairs. This induces unfair treatment for some language communities in regard to the cost of accessing commercial language services, the processing time and latency, as well as the amount of content that can be provided as context to the models. Therefore, we make the case that we should train future language models using multilingually fair subword tokenizers.