@inproceedings{bauwens-etal-2025-grampa,
title = "{GR}a{MP}a: Subword Regularisation by Skewing Uniform Segmentation Distributions with an Efficient Path-counting {M}arkov Model",
author = "Bauwens, Thomas and
Kacz{\'e}r, David and
de Lhoneux, Miryam",
editor = "Che, Wanxiang and
Nabende, Joyce and
Shutova, Ekaterina and
Pilehvar, Mohammad Taher",
booktitle = "Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)",
month = jul,
year = "2025",
address = "Vienna, Austria",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2025.acl-long.1180/",
doi = "10.18653/v1/2025.acl-long.1180",
pages = "24228--24257",
ISBN = "979-8-89176-251-0",
abstract = "Stochastically sampling word segmentations from a subword tokeniser, also called subword regularisation, is a known way to increase robustness of language models to out-of-distribution inputs, such as text containing spelling errors. Recent work has observed that usual augmentations that make popular deterministic subword tokenisers stochastic still cause only a handful of all possible segmentations to be sampled. It has been proposed to uniformly sample across these instead, through rejection sampling of paths in an unweighted segmentation graph. In this paper, we argue that uniformly random segmentation in turn skews the distributions of certain segmentational properties (e.g. token lengths and amount of tokens produced) away from uniformity, which still ends up hiding meaningfully diverse tokenisations. We propose an alternative uniform sampler using the same segmentation graph, but weighted by counting the paths through it. Our sampling algorithm, GRaMPa, provides hyperparameters allowing sampled tokenisations to skew towards fewer, longer tokens. Furthermore, GRaMPa is single-pass, guaranteeing significantly better computational complexity than previous approaches relying on rejection sampling. We show experimentally that language models trained with GRaMPa outperform existing regularising tokenisers in a data-scarce setting on token-level tasks such as dependency parsing, especially with spelling errors present."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="bauwens-etal-2025-grampa">
<titleInfo>
<title>GRaMPa: Subword Regularisation by Skewing Uniform Segmentation Distributions with an Efficient Path-counting Markov Model</title>
</titleInfo>
<name type="personal">
<namePart type="given">Thomas</namePart>
<namePart type="family">Bauwens</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">David</namePart>
<namePart type="family">Kaczér</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Miryam</namePart>
<namePart type="family">de Lhoneux</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2025-07</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)</title>
</titleInfo>
<name type="personal">
<namePart type="given">Wanxiang</namePart>
<namePart type="family">Che</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Joyce</namePart>
<namePart type="family">Nabende</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ekaterina</namePart>
<namePart type="family">Shutova</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Mohammad</namePart>
<namePart type="given">Taher</namePart>
<namePart type="family">Pilehvar</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Vienna, Austria</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-251-0</identifier>
</relatedItem>
<abstract>Stochastically sampling word segmentations from a subword tokeniser, also called subword regularisation, is a known way to increase robustness of language models to out-of-distribution inputs, such as text containing spelling errors. Recent work has observed that usual augmentations that make popular deterministic subword tokenisers stochastic still cause only a handful of all possible segmentations to be sampled. It has been proposed to uniformly sample across these instead, through rejection sampling of paths in an unweighted segmentation graph. In this paper, we argue that uniformly random segmentation in turn skews the distributions of certain segmentational properties (e.g. token lengths and amount of tokens produced) away from uniformity, which still ends up hiding meaningfully diverse tokenisations. We propose an alternative uniform sampler using the same segmentation graph, but weighted by counting the paths through it. Our sampling algorithm, GRaMPa, provides hyperparameters allowing sampled tokenisations to skew towards fewer, longer tokens. Furthermore, GRaMPa is single-pass, guaranteeing significantly better computational complexity than previous approaches relying on rejection sampling. We show experimentally that language models trained with GRaMPa outperform existing regularising tokenisers in a data-scarce setting on token-level tasks such as dependency parsing, especially with spelling errors present.</abstract>
<identifier type="citekey">bauwens-etal-2025-grampa</identifier>
<identifier type="doi">10.18653/v1/2025.acl-long.1180</identifier>
<location>
<url>https://aclanthology.org/2025.acl-long.1180/</url>
</location>
<part>
<date>2025-07</date>
<extent unit="page">
<start>24228</start>
<end>24257</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T GRaMPa: Subword Regularisation by Skewing Uniform Segmentation Distributions with an Efficient Path-counting Markov Model
%A Bauwens, Thomas
%A Kaczér, David
%A de Lhoneux, Miryam
%Y Che, Wanxiang
%Y Nabende, Joyce
%Y Shutova, Ekaterina
%Y Pilehvar, Mohammad Taher
%S Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)
%D 2025
%8 July
%I Association for Computational Linguistics
%C Vienna, Austria
%@ 979-8-89176-251-0
%F bauwens-etal-2025-grampa
%X Stochastically sampling word segmentations from a subword tokeniser, also called subword regularisation, is a known way to increase robustness of language models to out-of-distribution inputs, such as text containing spelling errors. Recent work has observed that usual augmentations that make popular deterministic subword tokenisers stochastic still cause only a handful of all possible segmentations to be sampled. It has been proposed to uniformly sample across these instead, through rejection sampling of paths in an unweighted segmentation graph. In this paper, we argue that uniformly random segmentation in turn skews the distributions of certain segmentational properties (e.g. token lengths and amount of tokens produced) away from uniformity, which still ends up hiding meaningfully diverse tokenisations. We propose an alternative uniform sampler using the same segmentation graph, but weighted by counting the paths through it. Our sampling algorithm, GRaMPa, provides hyperparameters allowing sampled tokenisations to skew towards fewer, longer tokens. Furthermore, GRaMPa is single-pass, guaranteeing significantly better computational complexity than previous approaches relying on rejection sampling. We show experimentally that language models trained with GRaMPa outperform existing regularising tokenisers in a data-scarce setting on token-level tasks such as dependency parsing, especially with spelling errors present.
%R 10.18653/v1/2025.acl-long.1180
%U https://aclanthology.org/2025.acl-long.1180/
%U https://doi.org/10.18653/v1/2025.acl-long.1180
%P 24228-24257
Markdown (Informal)
[GRaMPa: Subword Regularisation by Skewing Uniform Segmentation Distributions with an Efficient Path-counting Markov Model](https://aclanthology.org/2025.acl-long.1180/) (Bauwens et al., ACL 2025)
ACL