Charteris Ltd.
rpw@charteris.com

Emotions and language have important features in common:

• They both depend on complex assessments of the current social situation
• They both depend on a theory of mind, to assess how others perceive that situation
• They both serve a purpose of social communication
• They both depend on rapid unconscious calculations
• They are both learned and culturally transmitted

I propose that language and emotion are facets of the same symbolic social intelligence faculty, which we share with other primates. At some time within the last four million years, there has been strong selection pressure to increase the power of this faculty to support rapid fluent language. This increase in the horsepower of social intelligence, and the growth of the theory of mind enabled by language, have had major impacts on our emotional reactions, not always for the best.

In a previous paper at this conference (Worden 1998) I proposed that the language faculty has its origins in primate social intelligence. Evidence in support of this hypothesis is:

• Social intelligence requires us to form structured, composite representations of social situations. These representations are well suited to encode language meanings.
• Social intelligence takes inputs of all sensory modalities, and language meanings can refer to all sensory modalities.
• The primary usage of language is for social communication, which requires social intelligence
• Language requires a theory of mind, which is a facet of social intelligence.

It can be similarly argued that human emotion depends primarily on social intelligence: Apart from a few ‘primitive’ emotions like fear, many emotions serve a purpose of preparing us physiologically for social interactions with our peers. Emotions are driven by a cognitive assessment of our current social situation, depending on sense data of all modalities (Frijda 1986; Oatley 1992). Many emotional reactions produce effects (e.g. in facial expression or posture) which communicate social information to our peers. Finally, complex emotional reactions (e.g pride, shame, embarrassment) depend not just on our assessment of an actual social situation, but also on our assessment of how others see that situation (e.g. Harre 1986, Lutz 1988); they depend on our theory of mind (Carruthers and Smith, 1996).

The common antecedent of human language and human emotion can be seen in primate social behaviour. This largely consists of emotional reactions, driven by social assessments of interactions with their peers.

We can build a computational model of primate social intelligence which then extends to be a computational model of language. In this model, primates represent social situations internally by tree-like feature structures. A representation as complex as a tree structure – rather than, for instance, a simple list of attributes – is needed to capture the composite and structured nature of primate social situations, which primates are sensitive to.

Similarly, the meanings of sentences can all be represented by tree-like feature structures of nodes, links and slots. This is commonly done in computational models of language (Shieber 1986). Doing so enables the feature structures to represent the complex structured meanings of language, and to capture empirical findings about language meanings.

We assume, then, that both language meanings and social situations are represented by the same tree-like feature structures, or ‘scripts’, in the brain. These feature structures are the core of primate social intelligence, and are therefore the drivers of many primate emotional reactions. We can describe a typical primate social/emotional reaction in terms of this feature structure model:

• A primate observes the current social situation and encodes this as a feature structure (e.g. he observes an ally being threatened by another of superior rank).
• By unification with other learned feature structures, which describe the regularities of social life, he or she works out the implications of the current situation for himself (e.g. it may threaten physical danger, or a decrease in rank)
• These implications trigger emotional reactions such as hormone changes, and changes in posture and expression, which both communicate to others and prepare the individual for any required actions (e.g. appeasement).
• This assessment lasts for a long enough period (several seconds or minutes) to affect the animal’s physiology appropriately and to communicate effectively with others by posture and expression

A key concern in primate social life is the achievement of higher rank in the group, as this greatly enhances reproductive chances. Many primate social/emotional reactions are triggered by situations with potential impact on the primate’s rank, and emotional reactions have evolved to improve the chances of achieving high rank (de Waal, 1982; Byrne & Whiten 1988).

Just as many non-human primate social/emotional reactions are triggered by situations with potential impact on rank in the group, so are human emotions. For our assessment of our own rank, we may substitute the phrase ‘self-esteem’. Situations with a bearing on our self-esteem are powerful releasers of emotion. Our self-esteem depends largely on what we think others think about us; so our theory of mind, which enables us to partially understand what others think about us, is closely linked to our emotional reactions. Theory-of-mind knowledge, like other social knowledge, can be represented in tree-like feature structures.

Human emotional reactions have evolved from this primate model and are based on it – but the impact of language and a theory of mind has radically changed them, as discussed in the next section.

The emergence of language has coincided with – and perhaps caused - a great increase in the size and power of the human brain, relative to other primates (Deacon 1997). In order to produce and understand language, the human brain must perform several complex calculations per second. This is probably much faster than the rates of computation required for most primate social intelligence. Also the feature structures required for modern language are probably more complex than those used by most primates.

As we know from introspection, language is capable of serving as a kind of scaffolding for extended trains of conscious thought – which may reach conclusions very different from the simple automatic social inferences made by other primates. Language enables us to be consciously aware of a line of reasoning, to recall and modify it.

Whatever selection pressures drove the emergence and refinement of language, therefore, they have had important side-effects on human social intelligence:

1. They have increased the complexity of the feature structures we can use to represent social situations
2. They have increased the speed with which social inferences can be made
3. They have led to a powerful theory of mind
4. They have enabled conscious, language-based trains of thought

What has been the impact of these major changes on the human emotional system? As for other primates, our emotional reactions are driven by assessments of our social situation - using the same set of internal representations (feature structures) and computations. Increasing the horsepower of this social computation engine must have a powerful impact on those emotional reactions.

I suggest that the first impact of these changes may have been to take a pre-existing system of emotional reactions which had evolved over many millions of years, and which was well-matched to the task it performed, and to de-stabilise it – in that the new speed and power of social cognition was no longer well-matched to the task of driving emotional reactions.

A typical primate brain can form a simple feature structure representation of the current social situation, and can (probably slowly) perform various simple inferences on it which lead to simple emotional reactions. However, since the emergence of language the human brain has been capable of doing more, and doing it faster. This increase in speed and complexity threatens to undermine the previous primate emotional mechanisms, of a relatively slow and simple social computation leading reliably to an appropriate emotional reaction.

If the selection pressure leading to language (and to increased social intelligence) was a strong one, then such a de-stabilisation of emotions will have at least started to occur as the changes required for language began to take place. This de-stabilisation of emotions would lead to decreased fitness in the individuals concerned. In the first instance, there would then be an evolutionary ‘tug of war’ between the two selection pressures, for language proficiency and for stable appropriate emotions.

I suggest that a straight evolutionary tug-of-war may not have taken place, or in any case did not go on for long, because our minds evolved so as somehow to re-stabilise (if only incompletely) our emotional reactions. We have evolved mechanisms which somehow ‘fixate’ our emotional reactions into a comparatively small number of more stable states, leading to sustained emotional reactions, in spite of the rapid possible oscillations of different social inferences and conscious thoughts (see e.g. Stewart and Joines 1987). The cognitive model of emotion is not yet well defined enough to be able to know much about these possible re-stabilisation mechanisms.

The cost of this de-stabilisation and re-stabilisation has perhaps been rather high – resulting in the apparently irrational nature of many of our emotional reactions, with their ability to override conscious thought. Possibly the best way to understand our emotions will be to understand how they share common cognitive mechanisms with the language faculty (which is better understood and easier to study) – and then to understand how the evolution of language has forced the evolution of our emotions. From where we are now, this may be a long and difficult research programme – but given the great potential value of understanding the emotions, it is worth the attempt.

 Conference site: http://www.infres.enst.fr/confs/evolang/