Show simple item record

dc.contributor.author
Jin, Zhijing
dc.contributor.author
von Kügelgen, Julius
dc.contributor.author
Ni, Jingwei
dc.contributor.author
Vaidhya, Tejas
dc.contributor.author
Kaushal, Ayush
dc.contributor.author
Sachan, Mrinmaya
dc.contributor.author
Schölkopf, Bernhard
dc.contributor.editor
Moens, Marie-Francine
dc.contributor.editor
Huang, Xuanjing
dc.contributor.editor
Specia, Lucia
dc.contributor.editor
Yih, Scott Wen-tau
dc.date.accessioned
2023-10-23T10:04:04Z
dc.date.available
2022-01-20T13:25:05Z
dc.date.available
2022-02-25T13:14:42Z
dc.date.available
2023-10-23T10:04:04Z
dc.date.issued
2021-11
dc.identifier.isbn
978-1-955917-09-4
en_US
dc.identifier.other
10.18653/v1/2021.emnlp-main.748
en_US
dc.identifier.uri
http://hdl.handle.net/20.500.11850/527298
dc.identifier.doi
10.3929/ethz-b-000527298
dc.description.abstract
The principle of independent causal mechanisms (ICM) states that generative processes of real world data consist of independent modules which do not influence or inform each other. While this idea has led to fruitful developments in the field of causal inference, it is not widely-known in the NLP community. In this work, we argue that the causal direction of the data collection process bears nontrivial implications that can explain a number of published NLP findings, such as differences in semi-supervised learning (SSL) and domain adaptation (DA) performance across different settings. We categorize common NLP tasks according to their causal direction and empirically assay the validity of the ICM principle for text data using minimum description length. We conduct an extensive meta-analysis of over 100 published SSL and 30 DA studies, and find that the results are consistent with our expectations based on causal insights. This work presents the first attempt to analyze the ICM principle in NLP, and provides constructive suggestions for future modeling choices.
en_US
dc.format
application/pdf
en_US
dc.language.iso
en
en_US
dc.publisher
Association for Computational Linguistics
en_US
dc.rights.uri
http://creativecommons.org/licenses/by/4.0/
dc.title
Causal Direction of Data Collection Matters: Implications of Causal and Anticausal Learning for NLP
en_US
dc.type
Conference Paper
dc.rights.license
Creative Commons Attribution 4.0 International
ethz.book.title
Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing
en_US
ethz.pages.start
9499
en_US
ethz.pages.end
9513
en_US
ethz.version.deposit
publishedVersion
en_US
ethz.event
2021 Conference on Empirical Methods in Natural Language Processing (EMNLP 2021)
en_US
ethz.event.location
Punta Cana, Dominican Republic
en_US
ethz.event.date
November 7-11, 2021
en_US
ethz.publication.place
Stroudsburg, PA
en_US
ethz.publication.status
published
en_US
ethz.leitzahl
ETH Zürich::00002 - ETH Zürich::00012 - Lehre und Forschung::00007 - Departemente::02150 - Dep. Informatik / Dep. of Computer Science::02661 - Institut für Maschinelles Lernen / Institute for Machine Learning::09684 - Sachan, Mrinmaya / Sachan, Mrinmaya
en_US
ethz.leitzahl.certified
ETH Zürich::00002 - ETH Zürich::00012 - Lehre und Forschung::00007 - Departemente::02150 - Dep. Informatik / Dep. of Computer Science::02661 - Institut für Maschinelles Lernen / Institute for Machine Learning::09684 - Sachan, Mrinmaya / Sachan, Mrinmaya
en_US
ethz.date.deposited
2022-01-20T13:25:11Z
ethz.source
FORM
ethz.eth
yes
en_US
ethz.availability
Open access
en_US
ethz.rosetta.installDate
2022-02-25T13:14:49Z
ethz.rosetta.lastUpdated
2024-02-03T05:29:15Z
ethz.rosetta.versionExported
true
ethz.COinS
ctx_ver=Z39.88-2004&rft_val_fmt=info:ofi/fmt:kev:mtx:journal&rft.atitle=Causal%20Direction%20of%20Data%20Collection%20Matters:%20Implications%20of%20Causal%20and%20Anticausal%20Learning%20for%20NLP&rft.date=2021-11&rft.spage=9499&rft.epage=9513&rft.au=Jin,%20Zhijing&von%20K%C3%BCgelgen,%20Julius&Ni,%20Jingwei&Vaidhya,%20Tejas&Kaushal,%20Ayush&rft.isbn=978-1-955917-09-4&rft.genre=proceeding&rft_id=info:doi/10.18653/v1/2021.emnlp-main.748&rft.btitle=Proceedings%20of%20the%202021%20Conference%20on%20Empirical%20Methods%20in%20Natural%20Language%20Processing
 Search print copy at ETH Library

Files in this item

Thumbnail

Publication type

Show simple item record