Improving Transfer Learning for Sequence Labeling Tasks by Adapting Pre-trained Neural Language Models