A Two-Phased Approach for Natural Language Parsing into Formal Logic

Natural language is an intuitive medium for a human to communicate with a robot. Additionally, there are many tasks in areas such as manufacturing, military, and disaster response where communication is limited among the agents performing these tasks. Due to this limited communication, we focus on a protocol where most of the communication is done before and after the mission execution. As a first step in analyzing the effectiveness of this protocol, this thesis presents a two-phased approach to parsing natural language into an arbitrary formal logic. In the first phase, we aim to learn the generic structure of the logical expression associated with a natural language utterance. For example, if the sentence “Approach the target from the west” were to be parsed into the expression Approach(target;west), then the first phase would output a generic structure such as f(c0; c1), where f, c0, and c1 are placeholders for the actual values Approach, target, and west, respectively. In the second phase, we aim the learn how to assign the intended values to these placeholders. The method developed in this thesis is able to achieve an accuracy of 46% and 78% for the first and second phase of our natural language parser, respectively. With the help of our natural language parser, we can use the outputted logical expressions in future work to help in the analysis of the mission execution’s success or failure.