Tasks Overview

STEB evaluates style text embeddings across six task types. Each task probes a different aspect of how well embeddings capture stylistic properties.

Task	What It Measures	Primary Metric
Clustering	Style-based cluster formation	V-measure
All-to-All Pair Classification	Same-vs-different style discrimination	EER, AUC
Pre-defined Pair Classification	Verification on pre-defined pairs	EER, AUC
Order Alignment	Preservation of graded style ordering	Accuracy
Retrieval	Style-matched text retrieval	MRR, Recall@K
Probing	Linguistic property encoding	Accuracy