Estimating human emotions using wording and sentence patterns

Abstract

Recently, with progress in information processing technology, computerization in various business fields has progressed, providing us with more and more chances to interact with computers. Car navigation systems and nursing systems are examples of newly created interactions between computers and humans. However, we still feel much unease while communicating with computers. One reason may be the lack of a computer's ability to deal with affective information such as "language", "voice", "facial expression" and "gesture". Because affective information often includes human emotions, skillfully dealing with emotions is important in realizing more natural communication between computers and humans. Providing a machine with human "sensibility" might make it possible to create an attentive service robot or a friendly nursing robot. Our research group proposes a method to recognize and create human emotions which can be applied to a robot used for welfare services. We also aim to develop an affective computer with enough ability to communicate that people don't sense a disconnect. This paper proposes an "emotion estimation module" based on words for recognizing human emotions. This module extracts "emotion occurrence condition" based on emotions contained in words and sentence meanings, and estimates the emotion of a speaker or writer. The materials used for the module are collected from written quotations from conversations and internet blogs. We constructed a prototype system based on our method and conducted an experiment on sentences taken from internet blogs in order to verify the effectiveness of the method.