Setup

Part I of this book dealt with question answering on structured knowledge bases. Now we turn our focus to the other side of the curated Web: large volumes of unstructured textual content in collections like Wikipedia. Although knowledge bases are excellent sources of objective world knowledge, efforts toward their creation and maintenance limits their information coverage at any given point of time. On the other hand, textual sources arguably contain a higher coverage of up-to-date factual knowledge essential for addressing typical questions asked by an average Web user. Unlike canonicalized knowledge bases, text sources often offer substantial redundancy in information. Such redundancy can often be a blessing for boosting confidence in candidate answers derived by the QA system. This chapter focuses on the standard task setups in question answering over text data.