Pauchok: A Modular Framework for Question Answering
Stefanie Tellex · 2003
An ideal question answering system would be able to answer any question that a human could answer. Currently this dream is unattainable without major advances in artificial intelligence. However, we can advance towards this dream today by focusing on factoid questions that can be answered by short phrases. The Question Answering tracks at recent Text REtrieval Conferences (TREC) provide a forum for evaluating and comparing factoid question answering systems. Many systems entered in TREC have the same general architecture, but they are evaluated only as black box systems. This thesis describes Pauchok, an extensible playground that enables question answering components to be quickly implemented and tested. I implemented a number of passage retrieval algorithms within Pauchok's framework. I compared the performance of these passage retrieval algorithms, and found three important things. Boolean querying schemes perform well in the question answering task. The performance differences between various passage retrieval algorithms vary with the choice of document retriever, which suggests significant interactions between