Thesis (M.Sc.) - Cairo University - Faculty of Computers and Information - Departmen of Information Technology

In natural language, the same idea can be expressed by using di{uFB00}erent words. This phe-nomenon is the main reason why natural language processing (NLP) is such a challenging task, which has become highly evident in question answering (QA) where the task is to answer a natural language question. Furthermore, the words and syntax of the answer do not necessarily match those of the question. As a result, QA is considered to be one of the primary research areas in NLP and information retrieval. From an answer-retrieval perspective, QA can generally be divided into two branches. One is where the answer is created from one or multiple raw text documents. The other branch assumes that a given question corresponds to one of a set of questions that have already been answered, and the task is to retrieve the answer (s). In Arabic language, QA involving answer generation from raw text documents has received, to some extent, more interest than QA based on frequently asked questions (FAQ). However, it still in its {uFB02}edgling stages, since it has not addressed all types of questions, especially those which are phrased in a descriptive way. Furthermore, the existing methods are still unable to generate precise answers. Unlike the {uFB01}rst branch, the second one is likely to retrieve more precise answers, since the answers have already been manually generated. Moreover, this type of QA is able to answer all types of questions. Accordingly, QA based on FAQ is gaining more and more attention. This type is the focus of this thesis: answering a newly posed arabic natural language question based on pre-answered ones

Issued also as CD