University of Southern California

Emotion Twenty Questions

Quick Links ...

News ...

About ...

Emotion Twenty Questions (EMO20Q) is a dialog-based guessing game. It is much like the ordinary twenty questions game, but instead of picking an object, player one (the answerer) picks an emotion word. Player two (the questioner ) then tries to guess the emotion in less than 20 questions.

This game is meant to be fun, but it will also benefit science, technology, and society. Science, like this game, often starts with a good question. Scientifically, we want to study how people understand emotions by observing their natural language behavior. For technology, it will help computers understand human language that refers to emotions. For society, it will help people "get in touch with" their emotions and learn to express these emotions articulately. To these ends, your chat data will be saved securely and studied anonymously.

For The Curious ...

The study of computers and emotions is known as affective computing. At the sail lab, we have a big group that studies emotions. In particular, this project focuses on emotions in text. In contrast to other most other approaches to emotions in text, this does not study the emotion that the speaker/writer feels, but rather the emotional meaning that they are trying to convey. This is useful in cases when the speaker/writer is refering to 2nd hand emotional experiences like gossip or emotional experiences in the past (emotional self-report).

The human-human chat site is based on XMPP/Jabber instant messaging technology. The server runs ejabberd and this web-client is based on iJab. XMPP is an open standard, and the client and server software are free and open source. Likewise, we endeavor to make our data and methodologies available to the community. Currently these are hosted at code.google.com . Here is a description of how EMO20Q was inspired.

It is an observed fact that normally developing children ask lots of questions when they are in the period of rapidly increasing vocabulary. We feel that we can use this behavior as a model for making computers that can learn language. From a theoretical point of view, asking a question can be seen as the colloquial equivalent of making a hypothesis. The logician Charles Sanders Peirce identified hypothesis, or abduction, as one of the three main forms of reasoning. The tradition of question-asking has a long history that goes back to the Socratic Method.

The automated agent (computer) player of EMO20Q is powered by a graph-based structure that allows us to analyze emotional concepts in terms of graph connectivity. This can make for some interesting visualizations like this animation (warning: can be slow to load). Other graph visualizations can be found here.

Another aspect of understanding EMO20Q is determining how people conceptualize answers to yes/no questions, as can be seen in Fuzzy Yes/Yo graph 1 and Fuzzy Yes/No graph 2.

Publications ...

  • A Sequential Bayesian Dialog Agent for Computational Ethnography
    Abe Kazemzadeh, James "Jimmy" Gibson, Juanchen Li, Sungbok Lee, Panayiotis Georgiou, and Shrikanth Narayanan.
    In Proceedings of Interspeech, Portland, OR, Sept. 2012
    .
  • Emotion Twenty Questions (EMO20Q): Toward a Crowd-sourced Theory of Emotions
    Abe Kazemzadeh, Sungbok Lee, Panayiotis Georgiou, and Shrikanth Narayanan.
    In Proceedings of Affective Computing and Intelligent Interaction (ACII2011),Memphis, TN, Oct. 2011
  • EMO20Q Questioner Agent
    Abe Kazemzadeh, James Gibson, Panayiotis Georgiou, Sungbok Lee, and Shrikanth Narayanan.
    In Proceedings of Affective Computing and Intelligent Interaction (ACII2011),Memphis, TN, Oct. 2011
  • Determining Which Question To Ask, with the Help of Spectral Graph Theory
    Abe Kazemzadeh, Sungbok Lee, Panayiotis Georgiou, and Shrikanth Narayanan.
    In Proceedings of Interspeech, Florence, Italy. August, 2011.

People ...