Zero-shot Cross-domain Generalization of Dense Retrievers Fine-tuned on Specialized Subdomains

Brazilian legal retrieval is heterogeneous, covering case law, legislation, and question-based search. This makes training dense retrievers a trade-off between stronger domain specialization and broader robustness across retrieval types of search. In this paper, we explore this trade-off using three training setups based on Qwen3-Embedding-4B: a base model with no fine-tuning, a version trained only on legal data, and a mixed setup that combines legal data with SQuAD-pt supervised dataset. We evaluate these models on five legal datasets from the JU leaderboard, along with Quati dataset as an

Research goal: How does the zero-shot cross-domain generalization performance of dense retrievers compare when fine-tuned on specialized subdomains (e.g., biomedical vs. legal) within the BEIR benchmark, as measured by nDCG@10 across all datasets?

Autonomous synthesis report generated by SOVEREIGN Research Kernel. Tribunal consensus score: 9.0/10.