Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Kiegeland, Samuel, Snæbjarnarson, Vésteinn, Vieira, Tim, Cotterell, Ryan
Format: Preprint
Veröffentlicht: 2026
Schlagworte:
Online-Zugang:https://arxiv.org/abs/2604.28147
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866913077832712192
author Kiegeland, Samuel
Snæbjarnarson, Vésteinn
Vieira, Tim
Cotterell, Ryan
author_facet Kiegeland, Samuel
Snæbjarnarson, Vésteinn
Vieira, Tim
Cotterell, Ryan
contents Surprisal theory links human processing effort to the predictability of an upcoming linguistic unit, but empirical work often leaves the notion of a unit underspecified. In practice, experimental stimuli are segmented into linguistically motivated units (e.g., words), while pretrained language models assign probability mass to a fixed token alphabet that typically does not align with those units. As a result, surprisal-based predictors depend implicitly on ad hoc procedures that conflate two distinct modeling choices: the definition of the unit of analysis and the choice of regions of interest over which predictions are evaluated. In this paper, we disentangle these choices and give a unified framework for reasoning about surprisal over arbitrary unit inventories. We argue that surprisal-based analyses should make these choices explicit and treat tokenization as an implementation detail rather than a scientific primitive.
format Preprint
id arxiv_https___arxiv_org_abs_2604_28147
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle On the Proper Treatment of Units in Surprisal Theory
Kiegeland, Samuel
Snæbjarnarson, Vésteinn
Vieira, Tim
Cotterell, Ryan
Computation and Language
Surprisal theory links human processing effort to the predictability of an upcoming linguistic unit, but empirical work often leaves the notion of a unit underspecified. In practice, experimental stimuli are segmented into linguistically motivated units (e.g., words), while pretrained language models assign probability mass to a fixed token alphabet that typically does not align with those units. As a result, surprisal-based predictors depend implicitly on ad hoc procedures that conflate two distinct modeling choices: the definition of the unit of analysis and the choice of regions of interest over which predictions are evaluated. In this paper, we disentangle these choices and give a unified framework for reasoning about surprisal over arbitrary unit inventories. We argue that surprisal-based analyses should make these choices explicit and treat tokenization as an implementation detail rather than a scientific primitive.
title On the Proper Treatment of Units in Surprisal Theory
topic Computation and Language
url https://arxiv.org/abs/2604.28147