Advanced Techniques For Personalized, Interactive Question Answering

Using a computer to answer questions has been a human dream since the beginning of the digital era. A first step towards the achievement of such an ambitious goal is to deal with natural language to enable the computer to understand what its user asks. The discipline that studies the connection between natural language and the representation of its meaning via computational models is computational linguistics. According to such discipline, Question Answering can be defined as the task that, given a question formulated in natural language, aims at finding one or more concise answers in the form of sentences or phrases. Question Answering can be interpreted as a sub-discipline of information retrieval with the added challenge of applying sophisticated techniques to identify the complex syntactic and semantic relationships present in text. Although it is widely accepted that Question Answering represents a step beyond standard information retrieval, allowing a more sophisticated and satisfactory response to the user’s information needs, it still shares a series of unsolved issues with the latter. First, in most state-of-the-art Question Answering systems, the results are created independently of the questioner’s characteristics, goals and needs. This is a serious limitation in several cases: for instance, a primary school child and a History student may need different answers to the question: When did the Middle Ages begin?. Moreover, users often issue queries not as standalone but in the context of a wider information need, for instance when researching a specific topic. Although it has recently been proposed that providing Question Answering systems with dialogue interfaces would encourage and accommodate the submission of multiple related questions and handle the user’s requests for clarification, interactive Question Answering is still at its early stages. Furthermore, an issue which still remains open in current Question Answering is that of efficiently answering complex questions, such as those invoking definitions and descriptions (e.g. What is a metaphor?). Indeed, it is difficult to design criteria to assess the correctness of answers to such complex questions. These are the central research problems addressed by this thesis, and are solved as follows. An in-depth study on complex Question Answering led to the development of classifiers for complex answers. These exploit a variety of lexical, syntactic and shallow semantic features to perform textual classification using tree-kernel functions for Support Vector Machines. The issue of personalization is solved by the integration of a User Modelling component within the the Question Answering model. The User Model is able to filter and re-rank results based on the user’s reading level and interests. The issue of interactivity is approached by the development of a dialogue model and a dialogue manager suitable for open-domain interactive Question Answering. The utility of such model is corroborated by the integration of an interactive interface to allow reference resolution and follow-up conversation into the core Question Answering system and by its evaluation. Finally, the models of personalized and interactive Question Answering are integrated in a comprehensive framework forming a unified model for future Question Answering research.