Leading Whitespaces of Language Models' Subword Vocabulary Pose a Confound for Calculating Word Probabilities
This addresses a methodological issue for researchers using language models to model human incremental processing, though it is incremental as it corrects an existing approach rather than introducing a new paradigm.
The paper identifies a confound in aggregating subword probabilities from Transformer-based language models into word probabilities due to leading whitespaces in tokens, which violates probability axioms and misallocates surprisal in modeling human reading difficulty. It proposes a simple decoding technique to correct this, resulting in lower estimates of garden-path effects and poorer fits to reading times in experiments.
Predictions of word-by-word conditional probabilities from Transformer-based language models are often evaluated to model the incremental processing difficulty of human readers. In this paper, we argue that there is a confound posed by the most common method of aggregating subword probabilities of such language models into word probabilities. This is due to the fact that tokens in the subword vocabulary of most language models have leading whitespaces and therefore do not naturally define stop probabilities of words. We first prove that this can result in distributions over word probabilities that sum to more than one, thereby violating the axiom that $\mathsf{P}(Ω) = 1$. This property results in a misallocation of word-by-word surprisal, where the unacceptability of the end of the current word is incorrectly carried over to the next word. Additionally, this implicit prediction of word boundaries incorrectly models psycholinguistic experiments where human subjects directly observe upcoming word boundaries. We present a simple decoding technique to reaccount the probability of the trailing whitespace into that of the current word, which resolves this confound. Experiments show that this correction reveals lower estimates of garden-path effects in transitive/intransitive sentences and poorer fits to naturalistic reading times.