Conversational software, that is software with which a user can converse in a natural language such as English or Dutch, can be classified into two distinct categories: chatbots and dialogue systems. Both categories of systems have their advantages and disadvantages. Chatbot systems are trivial to build and maintain, but are too simplistic for applications that do more than answering frequently asked questions. Dialogue systems on the other hand are harder todevelop and maintain, can deal with less variety in user input, but are capable of handling and generating all kinds of linguistic phenomena such as grounding and information revision.This thesis presents a system that combines ideas from chatbots and dialogue system. This hybrid system doesn’t parse natural language into a semantic presentation, but contains a dialogue manager that can handle natural language directly. This system would make it easy to implement and maintain new domains even when the developer has little or no knowledge of computational linguistics. A statistics tutor has been implemented using this hybrid system with which three students interacted. This shows that the statistics tutor is able to deal with phenomena such as grounding, question accommodation and information revision.
展开▼
