Question Answering is an NLP task that aims to provide an exact answer to a natural language question. Earlier research in this area focussed on short factoid questions that could often be answered just with word matching (e.g. What is the capital of Norway?), but more recent work, such as the Watson Jeopardy Challenge and the CLEF QA for Machine Reading task aims to answer questions that require inference and integration of background knowledge. These sort of questions require more sophisticated processing and a more abstract understanding of the meaning of both the question and potential answer sources.
This project will investigate how the output of two parsers from different traditions (dependency-based MaltParser and HPSG-based PET parser) can be best combined to measure meaning similarity between questions, potential answer candidates and supporting textual evidence for use in a question answering system. A background and interest in syntax would be beneficial for this project. Details and further specification of the project can be discussed with Rebecca Dridan or Lilja Øvrelid.