In the field of emotion detection and interpretation, determining the valence of the stimulus has proven one of the most difficult obstacles to overcome. Measuring intensity is easier than valence. Although humans are good at recognizing valence in the facial expression of others [Ward Winton, Lois Putnam, and Robert Krauss, 1984], cameras and computer vision are not yet capable of this extrapolation. The trend towards portable computing also currently inhibits the use of vision recognition, as vision systems are generally bulky, heavy, fragile, and it is hard to wear a camera that looks at your face.
The external method of emotion stimulus employed by the IAPS database over an internally generated emotion routine, as employed by Dr. Clynes, helps us explore naturally occurring emotions. The use of a computer as the source of the stimulus allows us to observe the kinds of emotions that can be induced by a computer in ordinary situations. This natural setting where the computer provokes the user does not provide an exhaustive set of all the variants on emotions capable in a human computer interaction. The draw back to using slides is that low level arousal images are difficult to manufacture. There is a significant deficit in the IAPS database in these categories. One possible reason for this is the goal of advertising to produce a high arousal response in the viewer has caused a flood of high arousal pictures.
Language and labeling are big obstacles for emotion research. Peter Lang avoided language barriers by using pictorial cues. Dr. Manfred Clynes described in explicit detail exactly what he wanted the subject to feel for each emotion label. Many subjects in this experiment remarked after the fact that they had difficulty deciding if they ``liked'' or ``disliked'' some stimuli. They also talked about their inner struggle extracting what they really felt from what they thought they were supposed to feel. In the next incarnation of this experiment, the pictorial representations from Lang's work should be used rather than the verbal labeling of the slider bars.
The internal sense of time that each subject experiences will presumably change for the different stimuli. The relationship between duration and valence, is not studied here, but is an avenue for future exploration.
It has been shown in this pilot study, that valence information can be captured by a pressure sensitive mouse. Future research will produce a reliable, non-invasive, method for detecting the valence of a stimulus. Building an agent to monitor and interpret these signals in real-time could yield a valence aware program. Such a pattern recognition program would enable computers to be more responsive to the user's likes and dislikes. Finally, by adding intensity information to this agent, by means of further data manipulation, it can map the valence arousal space and learn how to respond to a variety of emotional states.