Question Answering for Portuguese: How Much Is Needed?

The task of a Question Answering (QA) system is to automatically answer a question in natural language, searching for information in a given data source, such as structured databases or unstructured natural language documents. In this paper we investigate how much is needed in terms of tools and resources for a QA system for Brazilian Portuguese. In particular we assess the impact of shallow and deep methods and the contribution of different tools and resources in terms of the performance of the system in the task. We argue that a combination of deep and shallow strategies results in optimal performance, but that shallow methods alone may already give adequate results.