Logo des Repositoriums
 

Measuring Gender Bias in German Language Generation

dc.contributor.authorKraft,Angelie
dc.contributor.authorZorn,Hans-Peter
dc.contributor.authorFecht,Pascal
dc.contributor.authorSimon,Judith
dc.contributor.authorBiemann,Chris
dc.contributor.authorUsbeck,Ricardo
dc.contributor.editorDemmler, Daniel
dc.contributor.editorKrupka, Daniel
dc.contributor.editorFederrath, Hannes
dc.date.accessioned2022-09-28T17:10:03Z
dc.date.available2022-09-28T17:10:03Z
dc.date.issued2022
dc.description.abstractMost existing methods to measure social bias in natural language generation are specified for English language models. In this work, we developed a German regard classifier based on a newly crowd-sourced dataset. Our model meets the test set accuracy of the original English version. With the classifier, we measured binary gender bias in two large language models. The results indicate a positive bias toward female subjects for a German version of GPT-2 and similar tendencies for GPT-3. Yet, upon qualitative analysis, we found that positive regard partly corresponds to sexist stereotypes. Our findings suggest that the regard classifier should not be used as a single measure but, instead, combined with more qualitative analyses.en
dc.identifier.doi10.18420/inf2022_108
dc.identifier.isbn978-3-88579-720-3
dc.identifier.pissn1617-5468
dc.identifier.urihttps://dl.gi.de/handle/20.500.12116/39481
dc.language.isoen
dc.publisherGesellschaft für Informatik, Bonn
dc.relation.ispartofINFORMATIK 2022
dc.relation.ispartofseriesLecture Notes in Informatics (LNI) - Proceedings, Volume P-326
dc.subjectgender bias
dc.subjectstereotypes
dc.subjectregard
dc.subjectnatural language generation
dc.subjectgpt-2
dc.subjectgpt-3
dc.subjectgerman
dc.titleMeasuring Gender Bias in German Language Generationen
gi.citation.endPage1274
gi.citation.startPage1257
gi.conference.date26.-30. September 2022
gi.conference.locationHamburg
gi.conference.sessiontitleTrustworthy AI in Science and Society

Dateien

Originalbündel
1 - 1 von 1
Lade...
Vorschaubild
Name:
trustai_03.pdf
Größe:
356.66 KB
Format:
Adobe Portable Document Format