<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="lewis-etal-2021-question">
    <titleInfo>
        <title>Question and Answer Test-Train Overlap in Open-Domain Question Answering Datasets</title>
    </titleInfo>
    <name type="personal">
        <namePart type="given">Patrick</namePart>
        <namePart type="family">Lewis</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Pontus</namePart>
        <namePart type="family">Stenetorp</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Sebastian</namePart>
        <namePart type="family">Riedel</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <originInfo>
        <dateIssued>2021-apr</dateIssued>
    </originInfo>
    <typeOfResource>text</typeOfResource>
    <relatedItem type="host">
        <titleInfo>
            <title>Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics: Main Volume</title>
        </titleInfo>
        <originInfo>
            <publisher>Association for Computational Linguistics</publisher>
            <place>
                <placeTerm type="text">Online</placeTerm>
            </place>
        </originInfo>
        <genre authority="marcgt">conference publication</genre>
    </relatedItem>
    <abstract>Ideally Open-Domain Question Answering models should exhibit a number of competencies, ranging from simply memorizing questions seen at training time, to answering novel question formulations with answers seen during training, to generalizing to completely novel questions with novel answers. However, single aggregated test set scores do not show the full picture of what capabilities models truly have. In this work, we perform a detailed study of the test sets of three popular open-domain benchmark datasets with respect to these competencies. We find that 30% of test-set questions have a near-duplicate paraphrase in their corresponding train sets. In addition, we find that 60-70% of answers in the test sets are also present in the train sets. Using these findings, we evaluate a variety of popular open-domain models to obtain greater insight into what extent they can generalize, and what drives their overall performance. We find that all models perform substantially worse on questions that cannot be memorized from train sets, with a mean absolute performance difference of 61% between repeated and non-repeated data. Finally we show that simple nearest-neighbor models outperform a BART closed-book QA model, further highlighting the role that train set memorization plays in these benchmarks</abstract>
    <identifier type="citekey">lewis-etal-2021-question</identifier>
    <identifier type="doi">10.18653/v1/2021.eacl-main.86</identifier>
    <location>
        <url>https://aclanthology.org/2021.eacl-main.86</url>
    </location>
    <part>
        <date>2021-apr</date>
        <extent unit="page">
            <start>1000</start>
            <end>1008</end>
        </extent>
    </part>
</mods>
</modsCollection>

%0 Conference Proceedings
%T Question and Answer Test-Train Overlap in Open-Domain Question Answering Datasets
%A Lewis, Patrick
%A Stenetorp, Pontus
%A Riedel, Sebastian
%S Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics: Main Volume
%D 2021
%8 apr
%I Association for Computational Linguistics
%C Online
%F lewis-etal-2021-question
%X Ideally Open-Domain Question Answering models should exhibit a number of competencies, ranging from simply memorizing questions seen at training time, to answering novel question formulations with answers seen during training, to generalizing to completely novel questions with novel answers. However, single aggregated test set scores do not show the full picture of what capabilities models truly have. In this work, we perform a detailed study of the test sets of three popular open-domain benchmark datasets with respect to these competencies. We find that 30% of test-set questions have a near-duplicate paraphrase in their corresponding train sets. In addition, we find that 60-70% of answers in the test sets are also present in the train sets. Using these findings, we evaluate a variety of popular open-domain models to obtain greater insight into what extent they can generalize, and what drives their overall performance. We find that all models perform substantially worse on questions that cannot be memorized from train sets, with a mean absolute performance difference of 61% between repeated and non-repeated data. Finally we show that simple nearest-neighbor models outperform a BART closed-book QA model, further highlighting the role that train set memorization plays in these benchmarks
%R 10.18653/v1/2021.eacl-main.86
%U https://aclanthology.org/2021.eacl-main.86
%U https://doi.org/10.18653/v1/2021.eacl-main.86
%P 1000-1008

Markdown (Informal)

[Question and Answer Test-Train Overlap in Open-Domain Question Answering Datasets](https://aclanthology.org/2021.eacl-main.86) (Lewis et al., EACL 2021)

• Question and Answer Test-Train Overlap in Open-Domain Question Answering Datasets (Lewis et al., EACL 2021)

ACL

• Patrick Lewis, Pontus Stenetorp, and Sebastian Riedel. 2021. Question and Answer Test-Train Overlap in Open-Domain Question Answering Datasets. In Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics: Main Volume, pages 1000–1008, Online. Association for Computational Linguistics.