Narrowing the Loop: Integration of Resources and Linguistic Dataset Development with Interactive Machine Learning

Abstract

This thesis proposal sheds light on the role of interactive machine learning and implicit user feedback for manual annotation tasks and semantic writing aid applications. First we focus on the cost-effective annotation of training data using an interactive machine learning approach by conducting an experiment for sequence tagging of German named entity recognition. To show the effectiveness of the approach, we further carry out a sequence tagging task on Amharic part-of-speech and are able to significantly reduce time used for annotation. The second research direction is to systematically integrate different NLP resources for our new semantic writing aid tool using again an interactive machine learning approach to provide contextual paraphrase suggestions. We develop a baseline system where three lexical resources are combined to provide paraphrasing in context and show that combining resources is a promising direction.