Treffer: Simple and Effective Question Processing using Regular Expressions and WordNet

Natural Language Processing is a computationally expensive and complex area of Computer Science. Question Answering in particular is known for having time and resources thrown at it, to achieve a system that is suitable only for a limited domain. Is this really necessary? Simplicity and efficiency are just as important as accuracy. A fast system is easier to test, debug and experiment with; it is also more attractive to an end user. This report details attempts to improve upon the question processing component of the Dalhousie University Jellyfish question answering system. To achieve speed as well as efficiency, our system utilizes a part-of-speech tagger instead of a full natural language parser. First, Regular expressions distinguish the syntactic structure, then entities and relations in a question are found and processed, and finally the question is assigned a category with the help of WordNet. Our approach is based on the syntactic structure of a question and is independent of any particular word groupings.