Dialog systems (computer systems that can converse with humans) usually use only a deterministic template or a probabilistic model to control the dialog flow. In this project, we will explore a dialog management system that uses the combination of templates and stochastic models to parse queries and generates responses. We also plan to divide the system into two layers: one for domain determination and one for semantic tagging. We will initially deploy deterministic models to collect training data, and gradually shift to the hybrid models. Finally, we will evaluate the model using a user simulator and human testers.

I have worked in the linguistics department on automatically measuring and annotating vowel formant frequencies from casual English speech. I have worked in Media Lab on developing an AI system that can reflect on its own actions. I have studied Japanese for 3 years, and I hope to do an NLP-related internship in Japan next summer.