Show simple item record

dc.contributor.authorSaunders, danielle
dc.contributor.authorsallis, rosie
dc.contributor.authorByrne, William
dc.date.accessioned2022-03-17T00:30:34Z
dc.date.available2022-03-17T00:30:34Z
dc.identifier.urihttps://www.repository.cam.ac.uk/handle/1810/335074
dc.description.abstractGenerating machine translations via beam search seeks the most likely output under a model. However, beam search has been shown to amplify demographic biases exhibited by a model. We aim to address this, focusing on gender bias resulting from systematic errors in grammatical gender translation. Almost all prior work on this problem adjusts the training data or the model itself. By contrast, our ap- proach changes only the inference procedure. We explore two techniques: constraining beam search to improve gender diversity in n-best lists, and reranking n-best lists using gender features obtained from the source sentence. Combining these strongly improves WinoMT gender translation accuracy for three language pairs without additional bilingual data or re- training. We also demonstrate our approach’s utility for consistently gendering named enti- ties, and its flexibility to handle new gendered language beyond the binary.
dc.description.sponsorshipEPSRC grants EP/M508007/1 and EP/N509620/1 and performed using resources from the Cambridge Tier-2 system operated by the University of Cambridge Research Computing Service funded by EPSRC Tier-2 cap- ital grant EP/P020259/1.
dc.rightsAll Rights Reserved
dc.rights.urihttp://www.rioxx.net/licenses/all-rights-reserved
dc.titleFirst the worst: Finding better gender translations during beam search
dc.typeConference Object
dc.publisher.departmentDepartment of Engineering
dc.date.updated2022-03-15T19:45:44Z
dc.identifier.doi10.17863/CAM.82514
dcterms.dateAccepted2022-02-24
rioxxterms.versionofrecord10.17863/CAM.82514
rioxxterms.versionAM
pubs.conference-nameFindings of ACL 2022
cam.orpheus.counter26*
cam.depositDate2022-03-15
pubs.licence-identifierapollo-deposit-licence-2-1
pubs.licence-display-nameApollo Repository Deposit Licence Agreement
rioxxterms.freetoread.startdate2023-03-16


Files in this item

Thumbnail

This item appears in the following Collection(s)

Show simple item record