Using Adaptive Information Extraction for Effective Human-Centred Document Annotation