Transfer language selection for zero-shot cross-lingual abusive language detection

作者：

Highlights：

• Zero-shot cross-lingual transfer can be effective in abusive language detection.

• Choosing transfer language only by the language family is not optimal.

• Linguistic similarity metrics can be used to find an optimal transfer language.

• Quantified World Atlas of Language Structures is a robust language similarity metric.

摘要

•Zero-shot cross-lingual transfer can be effective in abusive language detection.•Choosing transfer language only by the language family is not optimal.•Linguistic similarity metrics can be used to find an optimal transfer language.•Quantified World Atlas of Language Structures is a robust language similarity metric.

论文关键词：Abusive language detection,Zero-shot learning,Transfer learning,Linguistics

论文评审过程：Received 2 December 2021, Revised 12 March 2022, Accepted 8 May 2022, Available online 31 May 2022, Version of Record 31 May 2022.

论文官网地址：https://doi.org/10.1016/j.ipm.2022.102981