Can Large Language Models Follow Concept Annotation Guidelines? A Case Study on Scientific and Financial Domains