Should you marginalize over possible tokenizations?

TL;DR

This paper investigates whether ignoring tokenization variability in language models significantly affects probability estimates, finding minimal impact for most data but notable differences for complex, long words.

Contribution

The authors develop an importance sampling algorithm to estimate the true string probability by marginalizing over all tokenizations, assessing the impact on model likelihoods.

Findings

Marginalization impact is typically under 0.5% in log-likelihood.

The gap increases for data with long, complex words.

Ignoring marginalization is generally justified for standard datasets.

Abstract

Autoregressive language models (LMs) map token sequences to probabilities. The usual practice for computing the probability of any character string (e.g. English sentences) is to first transform it into a sequence of tokens that is scored by the model. However, there are exponentially many token sequences that represent any given string. To truly compute the probability of a string one should marginalize over all tokenizations, which is typically intractable. Here, we analyze whether the practice of ignoring the marginalization is justified. To this end, we devise an importance-sampling-based algorithm that allows us to compute estimates of the marginal probabilities and compare them to the default procedure in a range of state-of-the-art models and datasets. Our results show that the gap in log-likelihood is no larger than 0.5% in most cases, but that it becomes more pronounced for data with long complex words.