Document Type

Honors Project - Open Access

Abstract

Consider a personal assistant who lives with physical therapy patients, reminding them to perform various exercises and giving patients feedback about how well they have enacted such exercises. The assistant has learned, through the patient's interactions with his or her physical therapist, to personalize its commands and responses to the patient. Having such an assistant would decrease the time it takes for patients to recover via a more regimented exercise routine and would give physical therapists a much more detailed account of the activities of their patients. Here I demonstrate a system that, given an encoding of a patient's exercise, generates emotionally-manipulated sentence feedback about how to better perform the exercise. In essence, the system is composed of two primary modules: a reasoning module and a natural language generation module. For my results, I present a technique to generating instances of emotionally-altered exercise feedback on a stern-nurturing axis. I show a method for relating feedback words to deficiencies in exercise performance. Finally, I demonstrate a proof of concept technique for generating feedback given real-time skeleton mapping software. In future work, the system will be integrated into a physical therapy social avatar, which will interact as an assistant to patients within their home.

Share

COinS
 
 

© Copyright is owned by author of this document