@inproceedings{kew-sennrich-2023-uncovering,
title = "Uncovering Hidden Consequences of Pre-training Objectives in Sequence-to-Sequence Models",
author = "Kew, Tannon and
Sennrich, Rico",
editor = "Rogers, Anna and
Boyd-Graber, Jordan and
Okazaki, Naoaki",
booktitle = "Findings of the Association for Computational Linguistics: ACL 2023",
month = jul,
year = "2023",
address = "Toronto, Canada",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2023.findings-acl.438",
doi = "10.18653/v1/2023.findings-acl.438",
pages = "7010--7022",
abstract = "Some variants of self-supervised denoising objectives for pre-training encoder-decoder language models have been reported to have a negligible impact on downstream performance. Yet the design of these pre-training objectives leads to behavioural differences that can be uncovered with specific manipulations. We reproduce a recently proposed zero-shot control method and find that it is only successful on a subset of models. To understand what causes the difference in its effectiveness, we perform a set of controlled experiments, varying only the pre-training objective, and find unexpected interactions between the pre-training method and downstream controllability of models after fine-tuning. Our results show that different pre-training objectives have consequences that may not be visible in standard downstream evaluation, but which should be taken into account when developing models with controllability in mind.",
}
<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="kew-sennrich-2023-uncovering">
<titleInfo>
<title>Uncovering Hidden Consequences of Pre-training Objectives in Sequence-to-Sequence Models</title>
</titleInfo>
<name type="personal">
<namePart type="given">Tannon</namePart>
<namePart type="family">Kew</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Rico</namePart>
<namePart type="family">Sennrich</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>Findings of the Association for Computational Linguistics: ACL 2023</title>
</titleInfo>
<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">Jordan</namePart>
<namePart type="family">Boyd-Graber</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Naoaki</namePart>
<namePart type="family">Okazaki</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>Some variants of self-supervised denoising objectives for pre-training encoder-decoder language models have been reported to have a negligible impact on downstream performance. Yet the design of these pre-training objectives leads to behavioural differences that can be uncovered with specific manipulations. We reproduce a recently proposed zero-shot control method and find that it is only successful on a subset of models. To understand what causes the difference in its effectiveness, we perform a set of controlled experiments, varying only the pre-training objective, and find unexpected interactions between the pre-training method and downstream controllability of models after fine-tuning. Our results show that different pre-training objectives have consequences that may not be visible in standard downstream evaluation, but which should be taken into account when developing models with controllability in mind.</abstract>
<identifier type="citekey">kew-sennrich-2023-uncovering</identifier>
<identifier type="doi">10.18653/v1/2023.findings-acl.438</identifier>
<location>
<url>https://aclanthology.org/2023.findings-acl.438</url>
</location>
<part>
<date>2023-07</date>
<extent unit="page">
<start>7010</start>
<end>7022</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Uncovering Hidden Consequences of Pre-training Objectives in Sequence-to-Sequence Models
%A Kew, Tannon
%A Sennrich, Rico
%Y Rogers, Anna
%Y Boyd-Graber, Jordan
%Y Okazaki, Naoaki
%S Findings of the Association for Computational Linguistics: ACL 2023
%D 2023
%8 July
%I Association for Computational Linguistics
%C Toronto, Canada
%F kew-sennrich-2023-uncovering
%X Some variants of self-supervised denoising objectives for pre-training encoder-decoder language models have been reported to have a negligible impact on downstream performance. Yet the design of these pre-training objectives leads to behavioural differences that can be uncovered with specific manipulations. We reproduce a recently proposed zero-shot control method and find that it is only successful on a subset of models. To understand what causes the difference in its effectiveness, we perform a set of controlled experiments, varying only the pre-training objective, and find unexpected interactions between the pre-training method and downstream controllability of models after fine-tuning. Our results show that different pre-training objectives have consequences that may not be visible in standard downstream evaluation, but which should be taken into account when developing models with controllability in mind.
%R 10.18653/v1/2023.findings-acl.438
%U https://aclanthology.org/2023.findings-acl.438
%U https://doi.org/10.18653/v1/2023.findings-acl.438
%P 7010-7022
Markdown (Informal)
[Uncovering Hidden Consequences of Pre-training Objectives in Sequence-to-Sequence Models](https://aclanthology.org/2023.findings-acl.438) (Kew & Sennrich, Findings 2023)
ACL