Evaluating Generative Language Models in Information Extraction as Subjective Question Correction