Learning Generation Dictionaries from Product Reviews

Language generation systems rely on dictionaries describing the properties of a particular domain. Manual dictionary creation is a costly and lengthy process. The purpose of this research is to mine restaurant user reviews to automatically learn expressions and phrases characteristic of the restaurant domain. The induced domain ontology will be tested in the regeneration experiments. This work will build on the unsupervised learning technique introduced by M.Walker and R. Prasad in “Learning to generate naturalistic utterances using reviews in spoken dialog systems”, 2006.

Mentor: Elena Rishes, Graduate Student
Secondary Mentor: Prof. Marilyn Walker