The Script Tax: Measuring Tokenization-Driven Efficiency and Latency Disparities in Multilingual Language Models

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Dixit, Aradhya, Dixit, Shreem
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917269385248768
author Dixit, Aradhya
Dixit, Shreem
author_facet Dixit, Aradhya
Dixit, Shreem
contents Pretrained multilingual language models are often assumed to be script-agnostic, yet their tokenizers can impose systematic costs on certain writing systems. We quantify this script tax by comparing two orthographic variants with identical linguistic content. Across mBERT and XLM-R, the higher-fragmentation orthography shows a ~3.4x increase in fertility (6.73-6.85 vs. 2.10-2.35 tokens/word), leading to a 16.5x inference slowdown (0.23 vs. 3.8 sentences/second) on identical hardware. Using bits per character (BPC) to avoid the "NLL paradox" from subword fragmentation, we find a substantial increase in information cost: +19.7% for mBERT (8.06->9.65) and +47.1% for XLM-R (12.19->17.94). A round-trip conversion check (CER_rt=0.31) suggests these gaps reflect orthography-conditioned processing rather than mapping noise. Our results highlight tokenization as a key source of inequity in multilingual NLP and motivate script-aware tokenization and pretraining.
format Preprint
id arxiv_https___arxiv_org_abs_2602_11174
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle The Script Tax: Measuring Tokenization-Driven Efficiency and Latency Disparities in Multilingual Language Models
Dixit, Aradhya
Dixit, Shreem
Computation and Language
Artificial Intelligence
Pretrained multilingual language models are often assumed to be script-agnostic, yet their tokenizers can impose systematic costs on certain writing systems. We quantify this script tax by comparing two orthographic variants with identical linguistic content. Across mBERT and XLM-R, the higher-fragmentation orthography shows a ~3.4x increase in fertility (6.73-6.85 vs. 2.10-2.35 tokens/word), leading to a 16.5x inference slowdown (0.23 vs. 3.8 sentences/second) on identical hardware. Using bits per character (BPC) to avoid the "NLL paradox" from subword fragmentation, we find a substantial increase in information cost: +19.7% for mBERT (8.06->9.65) and +47.1% for XLM-R (12.19->17.94). A round-trip conversion check (CER_rt=0.31) suggests these gaps reflect orthography-conditioned processing rather than mapping noise. Our results highlight tokenization as a key source of inequity in multilingual NLP and motivate script-aware tokenization and pretraining.
title The Script Tax: Measuring Tokenization-Driven Efficiency and Latency Disparities in Multilingual Language Models
topic Computation and Language
Artificial Intelligence
url https://arxiv.org/abs/2602.11174