@inproceedings{nastase-merlo-2023-grammatical,
title = "Grammatical information in {BERT} sentence embeddings as two-dimensional arrays",
author = "Nastase, Vivi and
Merlo, Paola",
editor = "Can, Burcu and
Mozes, Maximilian and
Cahyawijaya, Samuel and
Saphra, Naomi and
Kassner, Nora and
Ravfogel, Shauli and
Ravichander, Abhilasha and
Zhao, Chen and
Augenstein, Isabelle and
Rogers, Anna and
Cho, Kyunghyun and
Grefenstette, Edward and
Voita, Lena",
booktitle = "Proceedings of the 8th Workshop on Representation Learning for NLP (RepL4NLP 2023)",
month = jul,
year = "2023",
address = "Toronto, Canada",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2023.repl4nlp-1.3",
doi = "10.18653/v1/2023.repl4nlp-1.3",
pages = "22--39",
abstract = "Sentence embeddings induced with various transformer architectures encode much semantic and syntactic information in a distributed manner in a one-dimensional array. We investigate whether specific grammatical information can be accessed in these distributed representations. Using data from a task developed to test rule-like generalizations, our experiments on detecting subject-verb agreement yield several promising results. First, we show that while the usual sentence representations encoded as one-dimensional arrays do not easily support extraction of rule-like regularities, a two-dimensional reshaping of these vectors allows various learning architectures to access such information. Next, we show that various architectures can detect patterns in these two-dimensional reshaped sentence embeddings and successfully learn a model based on smaller amounts of simpler training data, which performs well on more complex test data. This indicates that current sentence embeddings contain information that is regularly distributed, and which can be captured when the embeddings are reshaped into higher dimensional arrays. Our results cast light on representations produced by language models and help move towards developing few-shot learning approaches.",
}
<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="nastase-merlo-2023-grammatical">
<titleInfo>
<title>Grammatical information in BERT sentence embeddings as two-dimensional arrays</title>
</titleInfo>
<name type="personal">
<namePart type="given">Vivi</namePart>
<namePart type="family">Nastase</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Paola</namePart>
<namePart type="family">Merlo</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2023-07</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 8th Workshop on Representation Learning for NLP (RepL4NLP 2023)</title>
</titleInfo>
<name type="personal">
<namePart type="given">Burcu</namePart>
<namePart type="family">Can</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Maximilian</namePart>
<namePart type="family">Mozes</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Samuel</namePart>
<namePart type="family">Cahyawijaya</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Naomi</namePart>
<namePart type="family">Saphra</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Nora</namePart>
<namePart type="family">Kassner</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Shauli</namePart>
<namePart type="family">Ravfogel</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Abhilasha</namePart>
<namePart type="family">Ravichander</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Chen</namePart>
<namePart type="family">Zhao</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Isabelle</namePart>
<namePart type="family">Augenstein</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Anna</namePart>
<namePart type="family">Rogers</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Kyunghyun</namePart>
<namePart type="family">Cho</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Edward</namePart>
<namePart type="family">Grefenstette</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Lena</namePart>
<namePart type="family">Voita</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Toronto, Canada</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
</relatedItem>
<abstract>Sentence embeddings induced with various transformer architectures encode much semantic and syntactic information in a distributed manner in a one-dimensional array. We investigate whether specific grammatical information can be accessed in these distributed representations. Using data from a task developed to test rule-like generalizations, our experiments on detecting subject-verb agreement yield several promising results. First, we show that while the usual sentence representations encoded as one-dimensional arrays do not easily support extraction of rule-like regularities, a two-dimensional reshaping of these vectors allows various learning architectures to access such information. Next, we show that various architectures can detect patterns in these two-dimensional reshaped sentence embeddings and successfully learn a model based on smaller amounts of simpler training data, which performs well on more complex test data. This indicates that current sentence embeddings contain information that is regularly distributed, and which can be captured when the embeddings are reshaped into higher dimensional arrays. Our results cast light on representations produced by language models and help move towards developing few-shot learning approaches.</abstract>
<identifier type="citekey">nastase-merlo-2023-grammatical</identifier>
<identifier type="doi">10.18653/v1/2023.repl4nlp-1.3</identifier>
<location>
<url>https://aclanthology.org/2023.repl4nlp-1.3</url>
</location>
<part>
<date>2023-07</date>
<extent unit="page">
<start>22</start>
<end>39</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Grammatical information in BERT sentence embeddings as two-dimensional arrays
%A Nastase, Vivi
%A Merlo, Paola
%Y Can, Burcu
%Y Mozes, Maximilian
%Y Cahyawijaya, Samuel
%Y Saphra, Naomi
%Y Kassner, Nora
%Y Ravfogel, Shauli
%Y Ravichander, Abhilasha
%Y Zhao, Chen
%Y Augenstein, Isabelle
%Y Rogers, Anna
%Y Cho, Kyunghyun
%Y Grefenstette, Edward
%Y Voita, Lena
%S Proceedings of the 8th Workshop on Representation Learning for NLP (RepL4NLP 2023)
%D 2023
%8 July
%I Association for Computational Linguistics
%C Toronto, Canada
%F nastase-merlo-2023-grammatical
%X Sentence embeddings induced with various transformer architectures encode much semantic and syntactic information in a distributed manner in a one-dimensional array. We investigate whether specific grammatical information can be accessed in these distributed representations. Using data from a task developed to test rule-like generalizations, our experiments on detecting subject-verb agreement yield several promising results. First, we show that while the usual sentence representations encoded as one-dimensional arrays do not easily support extraction of rule-like regularities, a two-dimensional reshaping of these vectors allows various learning architectures to access such information. Next, we show that various architectures can detect patterns in these two-dimensional reshaped sentence embeddings and successfully learn a model based on smaller amounts of simpler training data, which performs well on more complex test data. This indicates that current sentence embeddings contain information that is regularly distributed, and which can be captured when the embeddings are reshaped into higher dimensional arrays. Our results cast light on representations produced by language models and help move towards developing few-shot learning approaches.
%R 10.18653/v1/2023.repl4nlp-1.3
%U https://aclanthology.org/2023.repl4nlp-1.3
%U https://doi.org/10.18653/v1/2023.repl4nlp-1.3
%P 22-39
Markdown (Informal)
[Grammatical information in BERT sentence embeddings as two-dimensional arrays](https://aclanthology.org/2023.repl4nlp-1.3) (Nastase & Merlo, RepL4NLP 2023)
ACL