Pre-Trained Language Models as Zero-Shot Tools for Social Science Research
For record linkage, I introduce EnsembleLink, an algorithm that uses pre-trained embeddings alongside string-based scorers to retrieve a short list of candidates and then identify the likeliest matches.
ProofPaper ↗
Key points
- Record linkage and demographic inference from names are routine demands in social science.
- I show that pre-trained language models perform both tasks accurately without requiring task-specific training.
- For ethnicity inference from names, I show that large language models (LLMs) achieve up to 83.8% accuracy on U.S. voter files, compared to 78.0% for name-supplemented BISG.
- Open-weight models offer reproducible implementations that approach these models in accuracy.
Sources (1)
- [1]Pre-Trained Language Models as Zero-Shot Tools for Social Science ResearcharXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 13, 06:43 PM
For record linkage, I introduce EnsembleLink, an algorithm that uses pre-trained embeddings alongside string-based scorers to retrieve a short list of candidates and then identify the likeliest matches.
Record linkage and demographic inference from names are routine demands in social science.
Extractive summary: sentences quoted from the sources.
Before this
- Oct 13, 2026inLoRA-MINT: Internal Representations for Auditing Training Data in Domain-adapted LLMs
- Oct 13, 2026Checking the Checkers: A Synthesis of Verifier Noise and Robustness in RLVR
- Oct 13, 2026Beyond the Verdict: Agentic Code Judging with Verifiable Feedback
- Oct 13, 2026Are We Testing Too Much? A Survey of Benchmark Compression for Efficient Evaluation of Large Language Models
- Oct 12, 2026SchemaWeaver: Accelerating On-Device Mobile Agentic AI via Schema-Guided Parallel LLM Decoding
- Oct 12, 2026Meet the Anthropic Engineer Keeping Its AI Agent Trustworthy
