BriefGPT.xyz
May, 2020
跨语言词嵌入中，语法性别联系高于主题性别偏见
Grammatical gender associations outweigh topical gender bias in crosslinguistic word embeddings
HTML
PDF
Katherine McCurdy, Oguz Serbetci
TL;DR
研究发现，语义的向量空间模型存在人类文化中不良偏见的问题，特别是主题性别偏见交互作用，并被语法性别偏见效应超过；同时，这些偏见可以通过语料库词形还原得以减轻，这对机器翻译等下游应用有重要的启示。
Abstract
Recent research has demonstrated that
vector space models
of
semantics
can reflect undesirable biases in human culture. Our investigation of crosslinguistic word embeddings reveals that topical
→