Evaluation of Text Generation A Survey

从3个维度将评价指标分类

1 Human-Centric Evaluation Methods

​ gold standard expensive to execute

2 Untrained Automatic Evaluation Metrics

widely used

汇总

property： 应该是说这个方法的关注点

3 Untrained Automatic Evaluation Metrics

overfitting and `gaming of the metric.’

参考

https://arxiv.org/abs/2006.14799