Procedure

I videotaped three actors eating and showing six different emotional states. The video tapes for each emotional state were 23 seconds long.

	Neutral
	Sadness
	Stress
	Anger
	Joy
	Boredom

Then, I designed two experiments. The first one intended to measure how well people can detect the emotional state of a person eating when the video is blurred. The purpose of this experiment is to measure how well people can detect the emotional state of the person eating when the most important feature they are looking at is body parts motion.

the blurring of the video eliminated the following details from the video.

	Face gestures
	Fingers movement
	surrounding environment
	amount of food picked by the fork
	rotation of body parts (hands, etc)
	Type of food

The subjects where asked to watch the videos and select the emotional state that best represented the emotional state of the actor eating.

The second experiment was intented to measure how well can people detect the emotional state of a person eating when the actors have the face blurred. The purpose is to understand what are the important features that people consider when detecting the emotional state of the actors eating, and how well is people able to detect the correct emotional state just having body language clues.

The subjects where asked to watch the videos and select the emotional state that best represented the emotional state of the actor eating.

The subjects were allowed to watch the videos only once in both experiments. The videos were seen by the subjects with no sound (I checked this for all 20 subjects). Another important thing to mention is that the videos were shown randomly and with no fixed number of videos shown. For example, a subject could see the same video during one experiment, or also could see two videos of boredom and none of anger for example. I decided to do this in order to give less clues to the subjects about what the correct answers were.

After finishing experiment 2, the subjects were given multiple choices showing features that probably they considered when trying to detect the emotional state of the actors. finally, there was a comments section in which the subjects were allowed to type additional features that they consider important to detect the emotional state of the actors.

I designed a web site for the experiment. Each time a person opens the web page of the experiment, the link to the videos is generated randomly so that all subjects answering the experiment will see a different sequence of videos for experiment one and two.

Click here to Start the Experiments

Twenty MIT Students (21-38 years old) took the experiment.