Pre-Trained Language Models as Zero-Shot Tools for Social Science Research

Record linkage and demographic inference from names are routine demands in social science. Most approaches to these problems require labeled training data that are difficult or costly to acquire.