@inproceedings{sharma-etal-2026-zipf,
title = "{Z}ipf Made Me Do It: A Causal Account of Outlier Dimensions",
author = "Sharma, Kartik and
Rai, Shreshth and
Gupta, Nikunj and
Vishwakarma, Dinesh K",
editor = "Arad, Dana and
Belinkov, Yonatan and
Chen, Hanjie and
Feldhus, Nils and
Mueller, Aaron and
Mohebbi, Hosein and
Sarti, Gabriele and
Tutek, Martin",
booktitle = "Proceedings of the 9th {B}lackbox{NLP} Workshop: Analyzing and Interpreting Neural Networks for {NLP}",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.blackboxnlp-1.33/",
pages = "561--572",
ISBN = "979-8-89176-476-7",
abstract = "Outlier dimensions are a recurring geometric property of transformer representations that have been linked to token frequency in pretrained language models. However, studies on pretrained models cannot disentangle token frequency from semantic and optimization confounds, preventing a causal understanding of whether frequency itself drives outlier formation. These distortions also degrade probing accuracy and hinder model compression, motivating a controlled study of their origin. We address this issue by isolating frequency skew as a controlled variable in a synthetic transformer trained on data with a tunable Zipfian token distribution. We find that outlier magnitude remains stable under low skew but increases sharply beyond a moderate threshold. A frozen frequency-based output bias fails to reproduce this behavior, model capacity exhibits a non-monotonic U-shaped effect, and an $L_2$-normalized cosine softmax suppresses outlier geometry only under a large temperature not reached by standard training. These results identify frequency skew as a causal driver of outlier dimensions and show that removing their geometry requires an intervention beyond standard training."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="sharma-etal-2026-zipf">
<titleInfo>
<title>Zipf Made Me Do It: A Causal Account of Outlier Dimensions</title>
</titleInfo>
<name type="personal">
<namePart type="given">Kartik</namePart>
<namePart type="family">Sharma</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Shreshth</namePart>
<namePart type="family">Rai</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Nikunj</namePart>
<namePart type="family">Gupta</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Dinesh</namePart>
<namePart type="given">K</namePart>
<namePart type="family">Vishwakarma</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-10</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 9th BlackboxNLP Workshop: Analyzing and Interpreting Neural Networks for NLP</title>
</titleInfo>
<name type="personal">
<namePart type="given">Dana</namePart>
<namePart type="family">Arad</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yonatan</namePart>
<namePart type="family">Belinkov</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Hanjie</namePart>
<namePart type="family">Chen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Nils</namePart>
<namePart type="family">Feldhus</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Aaron</namePart>
<namePart type="family">Mueller</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Hosein</namePart>
<namePart type="family">Mohebbi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Gabriele</namePart>
<namePart type="family">Sarti</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Martin</namePart>
<namePart type="family">Tutek</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Budapest, Hungary</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-476-7</identifier>
</relatedItem>
<abstract>Outlier dimensions are a recurring geometric property of transformer representations that have been linked to token frequency in pretrained language models. However, studies on pretrained models cannot disentangle token frequency from semantic and optimization confounds, preventing a causal understanding of whether frequency itself drives outlier formation. These distortions also degrade probing accuracy and hinder model compression, motivating a controlled study of their origin. We address this issue by isolating frequency skew as a controlled variable in a synthetic transformer trained on data with a tunable Zipfian token distribution. We find that outlier magnitude remains stable under low skew but increases sharply beyond a moderate threshold. A frozen frequency-based output bias fails to reproduce this behavior, model capacity exhibits a non-monotonic U-shaped effect, and an L₂-normalized cosine softmax suppresses outlier geometry only under a large temperature not reached by standard training. These results identify frequency skew as a causal driver of outlier dimensions and show that removing their geometry requires an intervention beyond standard training.</abstract>
<identifier type="citekey">sharma-etal-2026-zipf</identifier>
<location>
<url>https://aclanthology.org/2026.blackboxnlp-1.33/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>561</start>
<end>572</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Zipf Made Me Do It: A Causal Account of Outlier Dimensions
%A Sharma, Kartik
%A Rai, Shreshth
%A Gupta, Nikunj
%A Vishwakarma, Dinesh K.
%Y Arad, Dana
%Y Belinkov, Yonatan
%Y Chen, Hanjie
%Y Feldhus, Nils
%Y Mueller, Aaron
%Y Mohebbi, Hosein
%Y Sarti, Gabriele
%Y Tutek, Martin
%S Proceedings of the 9th BlackboxNLP Workshop: Analyzing and Interpreting Neural Networks for NLP
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-476-7
%F sharma-etal-2026-zipf
%X Outlier dimensions are a recurring geometric property of transformer representations that have been linked to token frequency in pretrained language models. However, studies on pretrained models cannot disentangle token frequency from semantic and optimization confounds, preventing a causal understanding of whether frequency itself drives outlier formation. These distortions also degrade probing accuracy and hinder model compression, motivating a controlled study of their origin. We address this issue by isolating frequency skew as a controlled variable in a synthetic transformer trained on data with a tunable Zipfian token distribution. We find that outlier magnitude remains stable under low skew but increases sharply beyond a moderate threshold. A frozen frequency-based output bias fails to reproduce this behavior, model capacity exhibits a non-monotonic U-shaped effect, and an L₂-normalized cosine softmax suppresses outlier geometry only under a large temperature not reached by standard training. These results identify frequency skew as a causal driver of outlier dimensions and show that removing their geometry requires an intervention beyond standard training.
%U https://aclanthology.org/2026.blackboxnlp-1.33/
%P 561-572
Markdown (Informal)
[Zipf Made Me Do It: A Causal Account of Outlier Dimensions](https://aclanthology.org/2026.blackboxnlp-1.33/) (Sharma et al., BlackboxNLP 2026)
ACL
- Kartik Sharma, Shreshth Rai, Nikunj Gupta, and Dinesh K Vishwakarma. 2026. Zipf Made Me Do It: A Causal Account of Outlier Dimensions. In Proceedings of the 9th BlackboxNLP Workshop: Analyzing and Interpreting Neural Networks for NLP, pages 561–572, Budapest, Hungary. Association for Computational Linguistics.