Low-Resource NLP Evaluation Bottlenecked by Scarce Sociolinguistic Expertise

A decade-spanning survey argues that technical capacity to scale multilingual models now far outpaces the human infrastructure needed to authentically evaluate them, particularly for low-resource languages where deep sociolinguistic expertise is inequitably distributed. The Annotation Scarcity Paradox frames this as a structural constraint, not a data problem.