@inproceedings{hao-linzen-2023-verb,
title = "Verb Conjugation in Transformers Is Determined by Linear Encodings of Subject Number",
author = "Hao, Sophie and
Linzen, Tal",
editor = "Bouamor, Houda and
Pino, Juan and
Bali, Kalika",
booktitle = "Findings of the Association for Computational Linguistics: EMNLP 2023",
month = dec,
year = "2023",
address = "Singapore",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2023.findings-emnlp.300",
doi = "10.18653/v1/2023.findings-emnlp.300",
pages = "4531--4539",
abstract = "Deep architectures such as Transformers are sometimes criticized for having uninterpretable {``}black-box{''} representations. We use causal intervention analysis to show that, in fact, some linguistic features are represented in a linear, interpretable format. Specifically, we show that BERT{'}s ability to conjugate verbs relies on a linear encoding of subject number that can be manipulated with predictable effects on conjugation accuracy. This encoding is found in the subject position at the first layer and the verb position at the last layer, but distributed across positions at middle layers, particularly when there are multiple cues to subject number.",
}
<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="hao-linzen-2023-verb">
<titleInfo>
<title>Verb Conjugation in Transformers Is Determined by Linear Encodings of Subject Number</title>
</titleInfo>
<name type="personal">
<namePart type="given">Sophie</namePart>
<namePart type="family">Hao</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Tal</namePart>
<namePart type="family">Linzen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2023-12</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Findings of the Association for Computational Linguistics: EMNLP 2023</title>
</titleInfo>
<name type="personal">
<namePart type="given">Houda</namePart>
<namePart type="family">Bouamor</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Juan</namePart>
<namePart type="family">Pino</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Kalika</namePart>
<namePart type="family">Bali</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Singapore</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
</relatedItem>
<abstract>Deep architectures such as Transformers are sometimes criticized for having uninterpretable “black-box” representations. We use causal intervention analysis to show that, in fact, some linguistic features are represented in a linear, interpretable format. Specifically, we show that BERT’s ability to conjugate verbs relies on a linear encoding of subject number that can be manipulated with predictable effects on conjugation accuracy. This encoding is found in the subject position at the first layer and the verb position at the last layer, but distributed across positions at middle layers, particularly when there are multiple cues to subject number.</abstract>
<identifier type="citekey">hao-linzen-2023-verb</identifier>
<identifier type="doi">10.18653/v1/2023.findings-emnlp.300</identifier>
<location>
<url>https://aclanthology.org/2023.findings-emnlp.300</url>
</location>
<part>
<date>2023-12</date>
<extent unit="page">
<start>4531</start>
<end>4539</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Verb Conjugation in Transformers Is Determined by Linear Encodings of Subject Number
%A Hao, Sophie
%A Linzen, Tal
%Y Bouamor, Houda
%Y Pino, Juan
%Y Bali, Kalika
%S Findings of the Association for Computational Linguistics: EMNLP 2023
%D 2023
%8 December
%I Association for Computational Linguistics
%C Singapore
%F hao-linzen-2023-verb
%X Deep architectures such as Transformers are sometimes criticized for having uninterpretable “black-box” representations. We use causal intervention analysis to show that, in fact, some linguistic features are represented in a linear, interpretable format. Specifically, we show that BERT’s ability to conjugate verbs relies on a linear encoding of subject number that can be manipulated with predictable effects on conjugation accuracy. This encoding is found in the subject position at the first layer and the verb position at the last layer, but distributed across positions at middle layers, particularly when there are multiple cues to subject number.
%R 10.18653/v1/2023.findings-emnlp.300
%U https://aclanthology.org/2023.findings-emnlp.300
%U https://doi.org/10.18653/v1/2023.findings-emnlp.300
%P 4531-4539
Markdown (Informal)
[Verb Conjugation in Transformers Is Determined by Linear Encodings of Subject Number](https://aclanthology.org/2023.findings-emnlp.300) (Hao & Linzen, Findings 2023)
ACL