Politician - An Imitation Game

摘要

We present the question-answering system Politician, which is a chatbot designed to imitate a fictional politician. The chatbot accepts questions on political issues and answers them accordingly. The questions are analyzed using natural language processing techniques, mainly using a custom scenario built in the Treex system, and no complex knowledge base is involved. Once morphological and syntactic annotations added by language tools are available for the question, an appropriate answer template is selected from the manually created set of answer templates based on nouns, verbs, and named entities occurring in the question. Then the answer template is transformed into a grammatically correct reply. So far, two working versions of Politician, Czech and English, have been created. We conducted a Turing-like test to test Politician's intelligence. We also briefly investigated the differences between the two languages and potential generalization of the approach to other topics. Apparently, morphological and syntactic information provides enough data for a very basic understanding of questions on a specific topic.