Chapter 2 EMOTIONAL DIMENSIONS
Chapter 2
EMOTIONAL DIMENSIONS
2.1 Emotional process
The emotional process, like any other cognition process, is complex and requires multiple steps. The first step is activation or arousal. Activation is followed by evaluation of the situation. For example, whether a given situation poses a threat or no threat, as well as whether one is facing a prey or a predator. Based on the assessment, either the inhibitory or energizing process is activated and then appropriate emotional behaviours are expressed. Each stage of the emotional process is associated with physiologic changes.
a) Activating mechanisms. Activation, a core concept in emotions, is defined as readiness for emotional behaviour.
b) Evaluation functions. The evaluating system of animals, which can work with or without cognitive awareness (presuming that only humans have cognitive awareness), works in three stages:
1) assesses events for their relevance,
2) directs action in response, and
3) feeds information to the hypothalamus (the head ganglion of the autonomic system) to regulate physiologic response.
For example, an animal assesses the size and strength of another animal, and determines whether the response should be aggression or withdrawal. In terms of emotions, evaluation functions help to distinguish whether the emotion is positive or negative.
Based on these aspects of the emotional mechanism, the idea of representing emotions as a multidimensional space arises.
2.2 Dimensionality of emotions
Emotion dimensionality is a simplified description of basic properties of emotional states. According to Osgood, Suci and Tannenbaum’s theory [Osg57] and subsequent psychological research (s. [Dat64, Meh74])], the communication of affect is conceptualised as three-dimensional with three major dimensions of connotative meaning, arousal, pleasure and power. The three different dimensions are regularly mentioned in the literature related to emotion analysis and are defined as follows:
· Activation or Arousal: Refers to the degree of intensity of the affect and ranges “from sleep to frantic excitement” [Pit93]. It is also related to the degree of readiness to act. Research from Darwin on has recognised that emotional states involve dispositions to act in certain ways. A basic way of reflecting that theme turns out to be surprisingly useful. States are simply rated in terms of the associated activation level, i.e. the strength of the person’s disposition to take some action rather than none. Activation dimension differentiates, for instance, anger from boredom. While the first one possess a high activation level, the last one presents lower disposition to act, and therefore lower activation level.
· Evaluation or Valence: Determines how positive or negative, liking or disliking the affect is. This dimension reflects the clearest common element of emotional states; an individual is materially influenced by feelings that are ‘valenced’, i.e. they are centrally concerned with positive or negative evaluations of people or things or events. The link between emotion and valence is widely agreed, although authors describe it in different terms. Arnold refers to the ‘judgement of weal or woe’; Tomkins, describes affect as what gives things value - ‘without its amplification, nothing else matters, and with its amplification, anything else can matter’; Rolls sees emotional processing as where "reward or punishment value is made explicit in the representation". For instance, happy is considered to be positive, i.e. it has a high evaluation level, while angry presents a negative value, i.e. low evaluation level.
· Power or Control: Relates to the degree of power or sense of control over the affect, and helps distinguish emotions initiated by the subject from those elicited by the environment e.g. contempt versus fear. It is also related to the degree of dominance/submission.
The main advantage of this theory is that it provides taxonomy allowing simple distance measures between emotion categories, as well as a continuous framework for expressing gradual, non-extreme emotional states. However, three dimensions do not still capture all the relevant aspects of an emotional state.
2.3 Theory of the activation – evaluation space
Activation-evaluation space is a particularization of the dimensionality introduced in section 2.2, which results in a representation that is both simple and capable of capturing a wide range of significant issues in emotion. It rests on a simplified treatment of two key themes: pleasure (evaluation) and arousal (activation).
Figure 2.1. Graphic representation of the activation-evaluation theory of emotions.
The axes of activation-evaluation plane reflect those themes. The vertical axis shows activation level, the horizontal axis evaluation. A basic attraction of that arrangement is that it provides a way of describing emotional states which is more tractable than using words, but which can be translated into and out of verbal descriptions. Translation is possible because emotion-related words can be understood, at least to a first approximation, as referring to positions in activation-emotion space. Various techniques lead to that conclusion, including factor analysis, direct scaling, and others.
Activation-evaluation space is a surprisingly powerful device, and it has been increasingly used in computationally oriented research (s. [Sca97, Cow99a, Cow99b]). However, it has to be emphasised that representations of that kind depend on collapsing the structured, high-dimensional space of possible emotional states into a homogeneous space of two dimensions. There is inevitably loss of information; and worse still, different ways of making the collapse lead to substantially different results.
Research suggests that the activation-evaluation space is naturally circular, i.e. states that are at the limit of emotional intensity define a circumference. The states that are the limit of emotional intensity are equidistant from an emotional neutral point (see figure 2.1). Many techniques converge on the conclusion that to a first approximation, emotion terms can be understood as referring to points in a space defined by those two axes.
[Per00] performed an experiment based on listening tests of some emotional utterances (cold anger, hot anger, happiness, neutrality and sadness). Results concluded that all the emotions were significantly different from each other on at least two of the dimensions. Findings in [Per00] show that the concept of the dimensions of emotion is useful to describe and distinguish emotions; and that emotions with a similar level of arousal, and sometimes a similar level of power, share acoustic characteristics in terms of F0 range and mean, and particularly intensity mean. It is also suggested that this contributes to perceived similarity between emotions, and consequently confusions, especially in hearing-impaired. It was observed that, in the majority of the cases, the two emotions confused were closer on the arousal dimension than on the other two dimensions.
In the framework of the PHYSTA project a system called FEELTRACE [Cow00] is designed to represent emotions on the activation-evaluation plane. The function of this tool is to record the way the signs are perceived by representative observers through a system that they find easy to use, and that gives reasonably reliable outcomes. Results of the test of FEELTRACE are consistent with the theoretical approach.
2.4 Assumption of the present work
Most of the prior work on emotion recognition [Bat00, Hub98]) makes only use of acoustic prosodic features, due to their ease of handling, which give mainly information concerning the arousal dimension of emotions. This study deals with a second dimension of the emotional space, i.e. pleasure, and its relation to quality features, i.e. auditory features that arise from variation in the source signal and vocal tract properties (see section 3.3). For instance, happiness and anger are emotions expressed with high intensity, i.e. with a high level in the arousal dimension, which makes them very difficult to classify based uniquely on prosodic features. However, they are situated opposite to each other on the pleasure axis, and therefore, quality features will contribute effectively to enhance this classification.
The assumption of this work is that both dimensions of the emotional space relate to different features of the speech, with different degrees of complexity [Tat02]. Therefore, it makes sense to divide the problem and apply hierarchical classification accomplished in terms of levels, from high to low, looking at a subspace of the features (prosodic or quality features).
25