|
Article Excerpt
The information that is decoded from a face allows an immediate assessment of the mood signalled by that face. The decoding process involved in the perception of facial expressions of emotion can inform models of perceptual and cognitive face processing in addition to informing models of emotion and social communication. Every face is a spatial array of similarly shaped features (mouths, eyes, brows, and nose) with similar spatial relationships among those features (distance and contour). Although information about emotional state is conveyed by the same features and relationships used to identify individuals, it is assumed that separate analysis of identity and emotion must occur at some level of processing (Bruce & Young, 1986; Farah, 1996; Tovee & Cohen-Tovee, 1993; Valentine, 1988). Certainly, when people are asked to examine the face for expressions of emotion, they are capable of ignoring the information about facial features or relationships that determine identity (Calder, Young, Keane, & Dean, 2000; Ekman & Friesen, 1976; Etcoff, 1984; Etcoff & Magee, 1992; McKelvie, 1995; Prkachin & Prkachin, 1994; but see also Schweinberger & Soukup, 1998). Different locations of the neuronal mechanisms involved in the processing of identity and the changeable aspects of a face, such as expression, have consistently been reported (Hoffman & Haxby, 2000).
One might think that the decoding of facial expressions of emotion would be more dependent upon the specific individual features of the face that make up the expression rather than the relationships among the features. In fact, caricatures of happy faces can be produced with a simple upturn of the lips and sadness with a downturn of the lips. It has, however, recently become clear that the global analysis involved in facial identity processing also applies to the analysis of facial expressions of emotions. The processing of facial expressions of emotion and other communicated expressions is similar to face recognition in being dependent upon relationships among facial features. Explicit manipulations of the configuration of the face disrupted the identification of facial expressions of emotion (Calder et al., 2000).
While facial expressions of emotion (FEE) and facial identity (Fl) require similar mechanisms which utilize the spatial relationships of the features, there is a difference in the processes. In the case of recognition of an individual, the process must involve an assessment of the unique features and unique relationships among the features of the face. In the case of the recognition of facial expressions of emotion, the process must involve an abstraction of the features and spatial relationships that are invariant relations for that expression. These consistent feature changes and relationships must then be generalized from one individual's face to another. Simply put, a smile is a smile regardless of whose face it appears on. It is perceived as a smile regardless of being made by a big or small mouth that is close to the eyes in a small face or is distant from the eyes in a large face.
The abstraction of variable and invariant relationships may require different strategies. Normally when confronted with a face, we either recognize it or we do not. The process is probably similar to signal detection, in that we do not really discriminate it from all other faces that we recognize. It is more likely that we first detect a face, then classify it as known/not known, and finally scrutinize the unique features and relationships that would lead to identification of the face. In general, experiments have required a same-different discrimination or used a match-to-sample task. We are rarely required to compare or discriminate one facial expression from another except in research designs. It seems more likely that we first detect a face, determine if there is an expression on it, and finally scrutinize it more closely to identify the invariant relationships of the features for that expression.
The ease with which we recognize objects is dependent upon the degree to which single features define or classify the object. If the size of the face is a distinctive feature, then further search of the face is unnecessary for identification. If the curve of the mouth is a distinctive feature of a happy expression, then this feature would be sufficient to recognize a happy face, and further search would be unnecessary. At issue for FEE is the number of dimensions and levels of classification that are involved in defining the category to which the expression belongs and that must be perceived in order to recognize it. Facial expression recognition can be seen as determining: stimulus-class membership (face vs. non-face), categorization (expression vs. non-expression), and subordinate categories (happy vs. sad). The difference between FI and FEE most probably is that FI requires more subordinate categories to accommodate the subtle differences that allow recognition of an individual without unique features, whi le fewer subordinate categories are involved in the recognition of FEE (at least for the well-known facial expressions).
We could also expect that the number of subordinate categories might vary for the expressions. For example, the shape of the features involved in the expression of happiness is quite distinct from those involved in other expressions of emotion. The uniqueness or lack of overlapping changes in the feature or spatial relations should predict the ease with which the expression is recognized. Some of the common expressions of emotion have more obvious overlap of feature changes and relationships among the features. For example, anger and disgust expressions share changes around the nose, mouth, and middle of the face and could be confused. Surprise and fear expressions also share an apparent opening of the face and change around the eyes and could be confused.
Inversion makes recognition of the individual disproportionally less accurate than the drop in recognition of other inverted objects (Yin, 1969, 1970). When the rationale outlined above is applied to the perception of facial expressions, it leads to a number of expectations. For example, if subjects use specific features to perceive the facial expressions, then inverting those features should severely disrupt identification of the expression. The degree of disruption should depend upon the particular facial expression that is inverted, and the expression should be confused with other expressions of emotion. A specific instance would be: happy faces are perceived as happy when the lips are turned up. Sad faces are seen as sad when the corners of the lips are down. Inverted happy faces should be confused with sad faces (and vice versa) if the direction of the lips is the critical feature necessary for the detection of these emotional expressions. If the relation between the direction of the lips and eyes is uni que and defines the expression, then inversion should be less disruptive. If there is overlap in the features or configuration of features, and this is the process that dominates the perception of the facial expression, then inversion of the face should be disruptive. Those expressions that require a distinction between the overlapping features and relation of those features should be more disrupted by inversion in terms of either recognition of the expression or confusion with other expressions that have overlapping changes in the relationship of the features.
A demonstration by Diamond and Carey (1986) of the face inversion effect was obtained for dog faces, but only by dog experts. The experts in dog recognition may have the ability to abstract configural information or relations that were ignored by nonexperts. The non-experts ignored the relationships important to recognition of the dogs and were consequently not subject to the face inversion effect. The importance of this demonstration is the idea that some relations were not being abstracted by the nonexperts. It seems reasonable to expect that expertise, the task employed to examine object or face recognition, and the time allowed to view the expression could either facilitate or not facilitate the abstraction of relations. If these factors do affect face expression perception, we should expect enhanced sensitivity to upright facial expressions that are more familiar. We should also expect that expressions with more discrete feature changes or less overlap in spatial relations should be less disrupted by inv ersion and that those with more overlapping feature changes will be more disrupted by inversion. Perception may also be influenced by...
|
