Karsten Homann and Anita Lulay Abstract
The research towards the development of mathematical assistants for high level mathematical problem solving has led to environments integrating specialized mathematical packages by combining techniques from artificial intelligence and symbolic mathematical computations.
Although claiming to assist in any kind of mathematical problem, these environments can not understand and solve algebra story problems. Such problems can be tackled by understanding, abstraction, and transformation of their representation into symbolic equational form which can be solved by algebraic algorithms. We report on a hybrid architecture in which the process of understanding, representation, and answer generation is done by neural networks.
The integration of the learning capabilities of neural networks into natural language processing systems combined with the symbolic algorithms of computer algebra systems provide a powerful tool for solving algebra story problems.