Learning Adaptable Patterns for Passage Reranking