Reading Wikipedia to Answer Open-Domain Questions

This paper proposes to tackle open-domain question answering using Wikipedia as the unique knowledge source: the answer to any factoid question is a text ... ACLAnthology FAQ(current)Corrections(current)Submissions(current) ReadingWikipediatoAnswerOpen-DomainQuestionsDanqiChen, AdamFisch, JasonWeston, AntoineBordesAbstractThispaperproposestotackleopen-domainquestionansweringusingWikipediaastheuniqueknowledgesource:theanswertoanyfactoidquestionisatextspaninaWikipediaarticle.Thistaskofmachinereadingatscalecombinesthechallengesofdocumentretrieval(findingtherelevantarticles)withthatofmachinecomprehensionoftext(identifyingtheanswerspansfromthosearticles).OurapproachcombinesasearchcomponentbasedonbigramhashingandTF-IDFmatchingwithamulti-layerrecurrentneuralnetworkmodeltrainedtodetectanswersinWikipediaparagraphs.OurexperimentsonmultipleexistingQAdatasetsindicatethat(1)bothmodulesarehighlycompetitivewithrespecttoexistingcounterpartsand(2)multitasklearningusingdistantsupervisionontheircombinationisaneffectivecompletesystemonthischallengingtask.AnthologyID:P17-1171Volume:Proceedingsofthe55thAnnualMeetingoftheAssociationforComputationalLinguistics(Volume1:LongPapers)Month:JulyYear:2017Address:Vancouver,CanadaVenue:ACLSIG:Publisher:AssociationforComputationalLinguisticsNote:Pages:1870–1879Language:URL:https://aclanthology.org/P17-1171DOI:10.18653/v1/P17-1171Bibkey:chen-etal-2017-readingCite(ACL):DanqiChen,AdamFisch,JasonWeston,andAntoineBordes.2017.ReadingWikipediatoAnswerOpen-DomainQuestions.InProceedingsofthe55thAnnualMeetingoftheAssociationforComputationalLinguistics(Volume1:LongPapers),pages1870–1879,Vancouver,Canada.AssociationforComputationalLinguistics.Cite(Informal):ReadingWikipediatoAnswerOpen-DomainQuestions(Chenetal.,ACL2017)CopyCitation:BibTeX Markdown MODSXML Endnote Moreoptions…PDF:https://aclanthology.org/P17-1171.pdfCode facebookresearch/DrQA +  additionalcommunitycodeDataCBT, DBpedia, NaturalQuestions, QUASAR-T, SQuAD, SearchQA, WikiMoviesPDF Cite Search CodeExportcitation ×BibTeXMODSXMLEndnotePreformatted@inproceedings{chen-etal-2017-reading, title="Reading{W}ikipediatoAnswerOpen-DomainQuestions", author="Chen,Danqiand Fisch,Adamand Weston,Jasonand Bordes,Antoine", booktitle="Proceedingsofthe55thAnnualMeetingoftheAssociationforComputationalLinguistics(Volume1:LongPapers)", month=jul, year="2017", address="Vancouver,Canada", publisher="AssociationforComputationalLinguistics", url="https://aclanthology.org/P17-1171", doi="10.18653/v1/P17-1171", pages="1870--1879", abstract="Thispaperproposestotackleopen-domainquestionansweringusingWikipediaastheuniqueknowledgesource:theanswertoanyfactoidquestionisatextspaninaWikipediaarticle.Thistaskofmachinereadingatscalecombinesthechallengesofdocumentretrieval(findingtherelevantarticles)withthatofmachinecomprehensionoftext(identifyingtheanswerspansfromthosearticles).OurapproachcombinesasearchcomponentbasedonbigramhashingandTF-IDFmatchingwithamulti-layerrecurrentneuralnetworkmodeltrainedtodetectanswersinWikipediaparagraphs.OurexperimentsonmultipleexistingQAdatasetsindicatethat(1)bothmodulesarehighlycompetitivewithrespecttoexistingcounterpartsand(2)multitasklearningusingdistantsupervisionontheircombinationisaneffectivecompletesystemonthischallengingtask.", } DownloadasFile CopytoClipboard Danqi Chen author Adam Fisch author Jason Weston author Antoine Bordes author 2017-jul text AssociationforComputationalLinguistics Vancouver,Canada conferencepublication Thispaperproposestotackleopen-domainquestionansweringusingWikipediaastheuniqueknowledgesource:theanswertoanyfactoidquestionisatextspaninaWikipediaarticle.Thistaskofmachinereadingatscalecombinesthechallengesofdocumentretrieval(findingtherelevantarticles)withthatofmachinecomprehensionoftext(identifyingtheanswerspansfromthosearticles).OurapproachcombinesasearchcomponentbasedonbigramhashingandTF-IDFmatchingwithamulti-layerrecurrentneuralnetworkmodeltrainedtodetectanswersinWikipediaparagraphs.OurexperimentsonmultipleexistingQAdatasetsindicatethat(1)bothmodulesarehighlycompetitivewithrespecttoexistingcounterpartsand(2)multitasklearningusingdistantsupervisionontheircombinationisaneffectivecompletesystemonthischallengingtask. chen-etal-2017-reading 10.18653/v1/P17-1171 https://aclanthology.org/P17-1171 2017-jul 1870 1879 DownloadasFile CopytoClipboard%0ConferenceProceedings %TReadingWikipediatoAnswerOpen-DomainQuestions %AChen,Danqi %AFisch,Adam %AWeston,Jason %ABordes,Antoine %SProceedingsofthe55thAnnualMeetingoftheAssociationforComputationalLinguistics(Volume1:LongPapers) %D2017 %8jul %IAssociationforComputationalLinguistics %CVancouver,Canada %Fchen-etal-2017-reading %XThispaperproposestotackleopen-domainquestionansweringusingWikipediaastheuniqueknowledgesource:theanswertoanyfactoidquestionisatextspaninaWikipediaarticle.Thistaskofmachinereadingatscalecombinesthechallengesofdocumentretrieval(findingtherelevantarticles)withthatofmachinecomprehensionoftext(identifyingtheanswerspansfromthosearticles).OurapproachcombinesasearchcomponentbasedonbigramhashingandTF-IDFmatchingwithamulti-layerrecurrentneuralnetworkmodeltrainedtodetectanswersinWikipediaparagraphs.OurexperimentsonmultipleexistingQAdatasetsindicatethat(1)bothmodulesarehighlycompetitivewithrespecttoexistingcounterpartsand(2)multitasklearningusingdistantsupervisionontheircombinationisaneffectivecompletesystemonthischallengingtask. %R10.18653/v1/P17-1171 %Uhttps://aclanthology.org/P17-1171 %Uhttps://doi.org/10.18653/v1/P17-1171 %P1870-1879 DownloadasFile CopytoClipboardMarkdown(Informal)[ReadingWikipediatoAnswerOpen-DomainQuestions](https://aclanthology.org/P17-1171)(Chenetal.,ACL2017)ReadingWikipediatoAnswerOpen-DomainQuestions(Chenetal.,ACL2017)ACLDanqiChen,AdamFisch,JasonWeston,andAntoineBordes.2017.ReadingWikipediatoAnswerOpen-DomainQuestions.InProceedingsofthe55thAnnualMeetingoftheAssociationforComputationalLinguistics(Volume1:LongPapers),pages1870–1879,Vancouver,Canada.AssociationforComputationalLinguistics.CopyMarkdowntoClipboard CopyACLtoClipboard