Movement Modeling for History Documents with Hidden Markov Models

Abstract

It is difficult for computers to understand the "meaning" of natural language sentences. To tackle this problem, some existing methods use predicate logic. However, they cannot deal with quantitative data such as geographical distance which is important for understanding historical events. We introduce a new method to construct a simulatable world model from documents. Simulations with this model will help computers to understand the contexts, guess unwritten information, and realize some rules. We experiment with some documents about the Sengoku period in Japanese Wikipedia, and construct a hidden Markov model about people's movement.