MIT EECS — Amazon Undergraduate Research and Innovation Scholar
Automatically Generating Accurate Annotations for High Precision Question Answering
START is a natural language question answering system developed by Boris Katz and the InfoLab at MIT’s CSAIL. START is able to answer questions by matching them to annotations, which are natural language descriptions of non-parseable content such as tabular data, images, and unstructured text. Currently, these annotations are manually created and loaded into START. It is very important to create a system to generate these annotations automatically, as it is infeasible to manually create annotations for a large corpus of data such as Wikipedia (5M articles). Automatic annotations can greatly increase the number of questions START can answer. The techniques developed in my research will also help push the state-of-the-art for other question answering systems such as Siri and Google Now.
I became interested in this project because of my desire to learn about natural language processing and apply it in a practical setting. I’m excited about the possibility of making a significant contribution to START and learning a lot while doing so.