Ph.D. thesis.
The rapid growth in the development of Internet-based information systems increases the demand for natural language interfaces that are easy to set up and maintain. Unfortunately, the problem of understanding natural language queries is far from being solved. Therefore this research proposes a simpler task of matching a one-sentence-long user question to a number of question templates, which cover the knowledge domain of the information system, without in-depth understanding of the user question itself.The research started with development of an FAQ (Frequently Asked Question) answering system that provides pre-stored answers to user questions asked in ordinary English. The language processing technique developed for FAQ retrieval does not analyze user questions. Instead, analysis is applied to FAQs in the database long before any user questions are submitted. Thus, the work of FAQ retrieval is reduced to keyword matching without understanding the questions, and the system still creates an illusion of intelligence.
Further, the research adapted the FAQ answering technique to a question-answering interface for a structured database, e.g., relational database. The entity-relationship model of the database is covered with an exhaustive collection of question templates - dynamic, parameterized "frequently asked questions" - that describe the entities, their attributes, and the relationships in form of natural language questions. Unlike a static FAQ, a question template contains entity slots - free space for data instances that represent the main concepts in the question. In order to answer a user question, the system finds matching question templates and data instances that fill the entity slots. The associated answer templates create the answer.
Finally, the thesis introduces a generic model of template-based question answering which is a summary and generalization of the features common for the above systems: they (i) split the application-specific knowledge domain into a number of question-specific knowledge domains, (ii) attach a question template, whose answer is known in advance, to each knowledge domain, and (iii) match the submitted user question to each question template within the context of its own knowledge domain.
automated question answering, FAQ answering, question-answering system, template-based question answering, question template, natural language based interface