FST Graduate Research Topic

Research Topic: Semi-supervised learning for semantic role labeling
Suggested By: Dunwei (Grant) Wen
Last Update: September 2nd, 2009

This research project focuses on semantic role labeling, a method of semantic analysis of natural language text, aiming to identify and label the semantic roles of arguments in text, which has wide application in almost all natural language related tasks such as question answering, information extraction and machine translation. There has been some great progress in this area in the last few years. However, the performance and robustness need to be improved due to inaccurate semantically annotated corpora and less data. Current research on semi-supervised learning provides us with a new way to address this problem. Different from both supervised and unsupervised learning, semi-supervised learning is a kind of machine learning techniques that can use both labeled and unlabeled data for training, and thus can be suitable for small training data, which is the case in current semantic role labeling. This project will explore appropriate semi-supervised learning methods and investigate useful constraints and bootstrapping algorithm for semantic role labeling. This research will based on our current research on semantic role labeling and any new inputs and ideas are very welcome. The graduate students who want to pursue the research need to be familiar with natural language processing and machine learning. Background in Linguistics will absolutely an asset for this project. This research topic is suitable for Project or Thesis.