Zipfian regularities in “non-point” word representations
作者：
Highlights：
• Variances of Gaussian embeddings can be used to quantify semantic uncertainty.
• There exist Zipfian regularities between word frequencies and semantic breadth/uncertainty.
• Zipfian patterns: more frequent words tends to be generic while less frequent ones tend to be specific.
• Zipfian patterns can be leveraged to increase entailment detection performance.
摘要
•Variances of Gaussian embeddings can be used to quantify semantic uncertainty.•There exist Zipfian regularities between word frequencies and semantic breadth/uncertainty.•Zipfian patterns: more frequent words tends to be generic while less frequent ones tend to be specific.•Zipfian patterns can be leveraged to increase entailment detection performance.
论文关键词：Word variances,Word frequencies,Zipf’s law,Meaning-frequency relation,Zipfian regularities,Word entailment,Semantic breadth
论文评审过程：Received 29 August 2020, Revised 4 January 2021, Accepted 4 January 2021, Available online 19 January 2021, Version of Record 19 January 2021.
论文官网地址：https://doi.org/10.1016/j.ipm.2021.102493
