A Minimum Message Length Approach for Discourse Interpretation

The ability to communicate with computer systems in Natural Language has great potential to improve the overall experience for users. However, current systems support only limited means of communication. In this project, we propose to investigate the application of model-selection techniques for interpreting human discourse. In particular, we will consider situations where the computer interprets users' discourse in the context of its own knowledge. The versatility of our approach will be demonstrated by using it to 1)interpret discourse in a human-computer dialogue, 2) provide feedback to short essays, and 3) determine the impact of a document on a model.