Unpacking Tokenization: Evaluating Text Compression and its Correlation with Model Performance

TL;DR

This paper investigates the role of compression in BPE tokenization, demonstrating its empirical correlation with language model performance across tasks and languages, and emphasizing its importance for future tokenizer development.

Contribution

It provides both theoretical and empirical evidence that compression quality in tokenization correlates with downstream model performance, highlighting its significance for improving language models.

Findings

Higher compression in tokenizers correlates with better downstream performance.

The correlation is stronger for generation tasks and smaller models.

Results are consistent across English and Turkish languages.

Abstract

Despite it being the cornerstone of BPE, the most common tokenization algorithm, the importance of compression in the tokenization process is still unclear. In this paper, we argue for the theoretical importance of compression, that can be viewed as 0-gram language modeling where equal probability is assigned to all tokens. We also demonstrate the empirical importance of compression for downstream success of pre-trained language models. We control the compression ability of several BPE tokenizers by varying the amount of documents available during their training: from 1 million documents to a character-based tokenizer equivalent to no training data at all. We then pre-train English language models based on those tokenizers and fine-tune them over several tasks. We show that there is a correlation between tokenizers' compression and models' downstream performance, suggesting that compression is a reliable intrinsic indicator of tokenization quality. These correlations are more pronounced for generation tasks (over classification) or for smaller models (over large ones). We replicated a representative part of our experiments on Turkish and found similar results, confirming that our results hold for languages with typological characteristics dissimilar to English. We conclude that building better compressing tokenizers is a fruitful avenue for further research and for improving overall model performance.