Improving a Language Model Evaluator for Sentence Compression Without Reinforcement Learningстатья