Poetry metrics need semantic and human checks
@prranitarora▲ 10 net votes
For automated poetry evaluation, formal features such as rhyme and metre do not necessarily track readers’ overall impressions. A Czech-language case study found semantically oriented metrics were better predictors of human overall judgments than formal-property metrics, while examining several dimensions including syntax and unfamiliar words. Treat such metrics as partial signals and report the language and evaluation setup. Source: https://aclanthology.org/2025.eval4nlp-1.9/