<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="he-etal-2019-unlearn">
    <titleInfo>
        <title>Unlearn Dataset Bias in Natural Language Inference by Fitting the Residual</title>
    </titleInfo>
    <name type="personal">
        <namePart type="given">He</namePart>
        <namePart type="family">He</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Sheng</namePart>
        <namePart type="family">Zha</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Haohan</namePart>
        <namePart type="family">Wang</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <originInfo>
        <dateIssued>2019-nov</dateIssued>
    </originInfo>
    <typeOfResource>text</typeOfResource>
    <relatedItem type="host">
        <titleInfo>
            <title>Proceedings of the 2nd Workshop on Deep Learning Approaches for Low-Resource NLP (DeepLo 2019)</title>
        </titleInfo>
        <originInfo>
            <publisher>Association for Computational Linguistics</publisher>
            <place>
                <placeTerm type="text">Hong Kong, China</placeTerm>
            </place>
        </originInfo>
        <genre authority="marcgt">conference publication</genre>
    </relatedItem>
    <abstract>Statistical natural language inference (NLI) models are susceptible to learning dataset bias: superficial cues that happen to associate with the label on a particular dataset, but are not useful in general, e.g., negation words indicate contradiction. As exposed by several recent challenge datasets, these models perform poorly when such association is absent, e.g., predicting that “I love dogs.” contradicts “I don’t love cats.”. Our goal is to design learning algorithms that guard against known dataset bias. We formalize the concept of dataset bias under the framework of distribution shift and present a simple debiasing algorithm based on residual fitting, which we call DRiFt. We first learn a biased model that only uses features that are known to relate to dataset bias. Then, we train a debiased model that fits to the residual of the biased model, focusing on examples that cannot be predicted well by biased features only. We use DRiFt to train three high-performing NLI models on two benchmark datasets, SNLI and MNLI. Our debiased models achieve significant gains over baseline models on two challenge test sets, while maintaining reasonable performance on the original test sets.</abstract>
    <identifier type="citekey">he-etal-2019-unlearn</identifier>
    <identifier type="doi">10.18653/v1/D19-6115</identifier>
    <location>
        <url>https://aclanthology.org/D19-6115</url>
    </location>
    <part>
        <date>2019-nov</date>
        <extent unit="page">
            <start>132</start>
            <end>142</end>
        </extent>
    </part>
</mods>
</modsCollection>

%0 Conference Proceedings
%T Unlearn Dataset Bias in Natural Language Inference by Fitting the Residual
%A He, He
%A Zha, Sheng
%A Wang, Haohan
%S Proceedings of the 2nd Workshop on Deep Learning Approaches for Low-Resource NLP (DeepLo 2019)
%D 2019
%8 nov
%I Association for Computational Linguistics
%C Hong Kong, China
%F he-etal-2019-unlearn
%X Statistical natural language inference (NLI) models are susceptible to learning dataset bias: superficial cues that happen to associate with the label on a particular dataset, but are not useful in general, e.g., negation words indicate contradiction. As exposed by several recent challenge datasets, these models perform poorly when such association is absent, e.g., predicting that “I love dogs.” contradicts “I don’t love cats.”. Our goal is to design learning algorithms that guard against known dataset bias. We formalize the concept of dataset bias under the framework of distribution shift and present a simple debiasing algorithm based on residual fitting, which we call DRiFt. We first learn a biased model that only uses features that are known to relate to dataset bias. Then, we train a debiased model that fits to the residual of the biased model, focusing on examples that cannot be predicted well by biased features only. We use DRiFt to train three high-performing NLI models on two benchmark datasets, SNLI and MNLI. Our debiased models achieve significant gains over baseline models on two challenge test sets, while maintaining reasonable performance on the original test sets.
%R 10.18653/v1/D19-6115
%U https://aclanthology.org/D19-6115
%U https://doi.org/10.18653/v1/D19-6115
%P 132-142

Markdown (Informal)

[Unlearn Dataset Bias in Natural Language Inference by Fitting the Residual](https://aclanthology.org/D19-6115) (He et al., EMNLP 2019)

• Unlearn Dataset Bias in Natural Language Inference by Fitting the Residual (He et al., EMNLP 2019)

ACL

• He He, Sheng Zha, and Haohan Wang. 2019. Unlearn Dataset Bias in Natural Language Inference by Fitting the Residual. In Proceedings of the 2nd Workshop on Deep Learning Approaches for Low-Resource NLP (DeepLo 2019), pages 132–142, Hong Kong, China. Association for Computational Linguistics.