What Tracks Hard-Negative Transfer? A Study of Lexical and Semantic Signals
The study investigates which properties of hard-negative supervision predict dense retriever transfer before training. It compares lexical noncoverage, frozen encoder similarity statistics, and hardness proxies across different retrieval sources like BM25 and LLMs. The key finding is that a simple lexical association best tracks the aggregate transfer ordering, while complex semantic statistics do not improve rank agreement.