Tag
This paper audits a multilingual affective generation benchmark and reveals that system rankings are driven by measurement artifacts rather than genuine performance differences, with annotator variability playing a key role.