This is the fourth of nine sections.
1. Introduction.
2. The cognitive processing element.
3. Ways of meeting the cognitive needs.

Apologies for the poor quality of some of the Figures.

Lisanne Bainbridge

Section 4.2 presents in detail a real 'sequencer' devised to account for part of the furnace operator's behaviour. The furnace operator's behaviour could be most simply accounted for, not by one big 'sequencer', but by a main 'sequencer', which identified the main contexts of the task, e.g. whether the state of the control task was acceptable or not, and on the basis of this led to several other 'sequencers' which chose the most appropriate behaviour within a sub-context. In the furnace operator's behaviour, the 'sequencers' lead to each other, and they refer to each other for data which is needed by one 'sequencer' but has been found by another.

This Section describes one 'sequencer' in detail as an example. It summarises all the 'sequencers' used in accounting for the furnace operator's behaviour, how they build up and cross refer to an overview of the situation which determines the sequence of behaviour, and how this leads to a contextual model of behaviour.

4.1. Identifying the sequencing determinants

In verbal protocols, a speaker often just changes topic without mentioning why they are now thinking about something different (e.g. Table 3.2.3.2). If the determining factors which affect the choice of topic are not explicitly mentioned, someone analysing the protocol has to use other means to identify what determines the choice.

Table 4.1 : (12K)
Effect of step-change in power usage, and protocol phrases after it (Bainbridge, 1972, Table 5.5b)
(Numbers in lower half indicate sequence of phrases)

The process by which this analysis was done for the steelworks protocol was described in Bainbridge (1972, 1990). Suppose that, in the protocol, topic A is sometimes followed by topic B and sometimes by topic C, and one wants to find out what it is that determines which of these happens. One can look at all the factors, in the process state and in the operator's recent behaviour, which might differ in the two circumstances. As an illustrative example, Table 4.1 shows the only instance of this type of analysis from the steelworks protocol which is simple enough to fit onto one page. The table summarises what happens after a step change (sudden large change) in the amount of electric power being used. The top half of the table shows aspects of the process state. The bottom half of the table shows the topics talked about (the numbers indicate the sequence of phrases). The protocol topics clearly fall into two groups : the first two phrases listed in the table are concerned with making an action, the other phrases with finding out what has happened. So the question is : what determines whether, after a step change, the operator thinks about action, or about identifying what has happened ? Which of the two groups of phrases occurs appears to depend on the size of the New T Value, particularly on how far it deviates from the ideal value (50). So the choice could be as represented in Figure 4.1. (Note that the 'alright' assessment in this Figure would probably have been made by visual judgement, from the angle of the display pointer on the scale, rather than by calculation. This equation just summarises the distinction which is made.)

Figure 4.1 : A suggested mechanism underlying the alternative sequences of behaviour represented in Table 4.1.

This method of analysis was used to identify all the factors which determined the sequence of topics in the furnace operator's verbal protocol.

4.2. Example 'sequencers', and how 'sequencers' are interrelated

Figure 4.2.1 shows an imaginary 'sequencer', in order to start with a simple example. In this example there are three factors which determine the sequence of events : the size and direction of the control error, and whether an action has been made before. The values of these sequencing determinants are found either by a 'routine' or by cross reference (represented by an arrow in the box) to data in the working storage of another 'sequencer'. This 'sequencer' contains elements for which the needs are met by 'routines' called 1, 2 and 3. The decisions in the 'sequencer' lead to other 'sequencers', called A, B and C. So the 'sequencer' determines the order in which the cognitive needs are thought about.

Figure 4.2.1 : Example of a 'sequencer', simplified from Figure 4.2.2.

Figure 4.2.2 shows the 'real' 'sequencer' from which Figure 4.2.1 was extracted. Figure 4.2.2 represents the most complex of the 'sequencers' determining the order in which the steelworks operator thought about topics. This 'sequencer' consists mainly of conditional elements.

Figure 4.2.2 .(8.4K): Sequence when control error unacceptable (Bainbridge, 1972, Figure 7.3.1).

There are three new symbols in Figure 4.2.2 :
i. a dashed box containing a number indicates a 'routine' which is described elsewhere.
ii. a dashed box containing a word indicates transfer of control to another 'sequencer' (this word is a name given to the 'sequencer' by the analyst, for ease of reference).
iii. the lists of numbers in brackets indicate the protocol phrases at the entry and exit points of this 'sequencer'.

It may help to understand what this 'sequencer' does. It has two main functions :
1. It contains one main cognitive need, to find which furnace would be the best one to cut the power to, i.e. to make an action on. This is found by 'routine' number 2.
2. It determines what best to think about next, on the basis of a combination of three aspects of the situation :
i. the size of the control error, and whether it is so large that an action is required,
ii. whether or not the operator has anticipated that an event (change of furnace stage) will happen later during this half-hour,
iii. whether or not any of the furnaces have, or have had, a cut to their power supply,

The operator finds the data needed for this decision making, by cross reference to working storage, or by carrying out a routine. The decisions made by the conditional elements lead on to one of four other 'sequencers' :
i. OVERALL, in which the operator assesses the size of the control error, and chooses an appropriate activity on the basis of it,
ii. ACTION, which chooses the size of, and assesses the effect of making, an action,
iii. ASSESS, in which the operator predicts the effects of an anticipated event, and its implications for control decisions,
iv. ONGOING, in which the operator reviews the state of all the furnaces considered together,
Each of these 'sequencers' also calls on 'routines' to find data needed, and then leads on to other 'sequencers', depending on the evidence obtained.

Several main 'sequencers' were needed to describe the factors determining the order in which the operator mentioned topics in his protocol. The most frequently occurring 'sequencer' was OVERALL. This is not surprising, as the operator's task is to control the power usage, and the OVERALL 'sequencer' finds out what the control error is, and determines what should be done next on the basis of it. Figure 4.2.3 summarises how the 'sequencers' lead to each other. There is no need for a higher level executive or deus ex machina to generate the sequence of behaviour.

Figure 4.2.3 : Any one 'sequencer' may be followed by several other 'sequencers'.

4.3. The general properties of a 'sequencer'

A 'sequencer' contains some main cognitive needs, and calls on 'routines' to meet them. It also makes decisions about what to do next, on the basis of various aspects of the situation. These decisions lead to other 'sequencers', which call 'routines', and so on.

The decisions are represented as conditional elements. A conditional element's need, for the data on which a decision is made, may be met either by cross reference to working storage or by carrying out a routine (see Section 3, part 3.2.2 on the two-arrow box).

As the data needs which are met by reference to working storage are never mentioned in the verbal protocol, this suggests that the decisions represented by the 'sequencer' might be done by unconscious parallel processing, rather than by a sequence of distinct decisions each of which considers only one variable. The two-arrow box implies that the conditional processing is done in parallel if the data required is available in working storage, and as serial processing if a 'routine' has to be carried out to find the data. This flexibility is represented in some diagrams by dashed lines (see Figure 4.6.1).

There is an important point to make about the conditional elements and the 'sequencer' diagrams. Carrying out the 'routines' is explicitly mentioned in the protocol. However the sequencing decisions are not explicitly mentioned : the presence of a decision, and the variables and criteria used in the decision, were inferred by the method described in Section 4.1. So there is no direct evidence about the decisions in a 'sequencer'. The structure given in the diagram just represents the simplest form for describing the choices identified by this type of analysis. The diagram shows the influencing factors, and the combinations of them which lead to particular outcomes. It is convenient to use conditional elements to describe the decisions, not only because this improves the internal consistency of this approach to cognitive modelling, but also because the choices do have to be expressed in a way which represents how 'routines' are called on if the data needed is not already available, and to show how other 'routines' fit into the sequencing. If there was not this requirement, the decisions could just as well be described by a multidimensional table, by comparing lists, by states in working storage which lead to transitions, or in several other formats. The importance of these provisos will appear in Section 6, part 6.4.

4.4. The working storage overview

The main items which analysis identified as determining the order in which topics were thought about are listed on the right (B) side of Table 3.2.3.1. It is clear that the two columns of this Table (the left side showing the cognitive needs met by the main 'routines' and the right side showing the items which determine the sequence of behaviour) which were arrived at by independent analyses, are very similar. This led me to infer that the data found to meet the originating cognitive needs of the main 'routines' (left side of table) are also the data in working storage which is referred to by the 'sequencer' decisions (right side of table).

It is possible to suggest that these main data, about the main cognitive needs, are in a general overview form of working storage, which is continuously available to all processing rather than local to particular 'routines'. Figure 4.4.1 below shows evidence from the furnace operator simulation which supports this. Figure 4.2.3, which summarises how 'sequencers' follow on from each other, shows no very clear distinction between processing when the control state (as identified by the OVERALL 'sequencer') is acceptable (ONGOING and STEP CHANGE) or unacceptable (UNACCEPT). In Figure 4.4.1, the 'sequencers' are laid out in the same positions as in Figure 4.2.3, but the arrows show how the 'sequencers' make use of, i.e. cross refer to, data found by other 'sequencers'. This shows clearly that processing when the control state is unacceptable makes extensive use of data obtained when the control state was acceptable (such as which furnaces are doing what, what is likely to happen in the near future, what effect this will have on the control situation, and what to do about it). This suggests that the 'sequencers' do not fall into distinct groups, and that they all operate within the same general overview. However, there is not one big 'sequencer', as all the 'sequencers' in Figure 4.2.3. are independent in the sense that they can either follow or be followed by several other 'sequencers' (including UPDATE, see Figure 5.1.1.1).

Figure 4.4.1 : Sequencers refer to data in working storage which was found by other 'sequencers'. Compare this to Figure 4.2.3.

In a processing element representation of cognitive processing, the overview is made up of what is in the working storage boxes associated with the main cognitive needs, which are located in the 'sequencers'. The 'sequencers' refer to these items when they make decisions about what best to do next. This overview is called 'situation awareness' in many studies of complex tasks.

If the contents of working storage are available in parallel, it is possible to think of the working storage boxes as a type of structured blackboard. The processing element representation suggests working storage is inherently structured, relative to the cognitive needs, and by cross-references between cognitive needs. These cross-references have been symbolised in the Figures either explicitly, e.g. Figure 2.3.2, or by boxes containing arrows (e.g. Figure 4.2.1). The arrow is interpreted as pointing to the data needed, which is already available either earlier in the present processing or elsewhere with another major cognitive need.

The points made here supplement the points on working storage in Section 2, and on the main cognitive needs and the overview in Section 3, part 3.2.3. The element mechanism shows clearly that what is in working storage is a construct : what is kept in mind temporarily is the result of thinking about the task, not an untransformed representation of input data. Indeed, it has been called working 'storage' to underline this difference from some models of working memory. For more discussion of working storage and the nature of the overview, see Section 5.

Figure 4.5.1 (9.3K) : The basic contextual processing cycle.

4.5. Contextual processing

This account leads to a contextual model of cognitive processing in complex tasks. Sequencing decisions, made by the conditional elements in the 'sequencers', lead to the use of 'routines', which update the working storage overview. This overview then acts as the context for (i.e. provides the data for) the sequencing decisions, and so on. This can be summarised by the cycle in Figure 4.5.1. Figure 4.5.2 shows, greatly simplified, how this basic cycle interacts with the environment and with knowledge, and how highly salient information in the environment can override the working storage (attract attention). The actual operation of the cycle can be represented by the cognitive elements in the 'sequencers', calling on 'routines'. So context effects which determine the flexible sequencing of behaviour can be accounted for by the proposed mechanism.

Figure 4.5.2 : A simple representation of the contextual cycle in relation to the knowledge base and the environment.

This representation of the basic nature of cognitive processing, as an interplay between context (in the overview) and processing, contrasts with sequential models of processing, which may be used in experimental psychology and in human factors/ ergonomics. In sequential models, processing proceeds in one direction only, from activation of the senses to motor response, each stage of processing reacts only to the output of previous stages, and the task goals are given. This sort of representation does not include a contextual overview, and cannot account for active search for information, for thinking about topics in a flexible sequence, or for thinking out what the optimum goals should be. So a sequential model is unable to account for some important features of behaviour in complex dynamic tasks. Hollnagel, 1993, has an interesting discussion of these issues.

In this contextual account, the main cognitive needs fall into two main groups, concerned with working out what is going on, and with working out what to do about it. These are not synonymous with processing sensory inputs and producing motor outputs. The processes required to understand what is going on may include actions (for example in process operation - walking to another part of the building to find some necessary information), and devising a response can include getting information (for example about what actions are available) (Bainbridge, 1981, 1993a). This sort of flexibility can be dealt with by the cognitive element and 'sequencers' approach.

There are tasks (such as experimental laboratory tasks which attempt to exclude context effects as far as possible) in which behaviour may be adequately represented by a sequential model. However, in my view it would be both more parsimonious and more interesting to have only one model for both simple and complex tasks. This would be possible if the model which accounts for complex situations could also account for simple ones : in the present case, if the contextual model could behave like a serial model in the appropriate circumstances. In fact, we have already seen it do this. Section 3, part 3.2.2 describes the two-arrow box, which symbolises a mechanism by which, if the data required is not already available in working storage, then a 'routine' is carried out to find it. Section 3.2.3 describes how this two-arrow box accounts for the way in which, in a new situation for which an overview has not yet been built up, the operator has to go through a standard sequence of thinking. When no context is (yet) available, then the contextual mechanism generates a standard sequence of behaviour. So standard sequences of behaviour are generated by, or emerge from, the contextual mechanism in certain circumstances, even though they are not (necessarily) explicitly represented in it. (For more discussion of this, see Bainbridge, 1993a.)

Figure 4.6.1 : An example of conditional elements in a 'routine'.

4.6. A complex routine

The difference between 'routines' and 'sequencers' is that 'routines' meet a cognitive need, while 'sequencers' determine the order in which needs are thought about The difference is not that 'sequencers' consist of conditional elements and 'routines' do not. 'Routines' may contain conditional elements and 'sequencers' may not. For example, the furnace operator assesses the present control state by a 'routine' which uses three different methods, depending on circumstances. Figure 4.6.1 is a simplified description of the choice between these alternative methods, represented using conditional elements. Figure 4.6.2 is the full 'routine'.

Figure 4.6.2 (12.5K) : 'Routine' describing the assessment of control error. Different methods of assessment are used at different times in the half-hour (Bainbridge, 1972, Figure 7.2.1).

In 'sequencers', conditional elements have been used to represent the choice of the most appropriate cognitive need to meet next. In 'routines', conditional elements have been used to choose between alternative methods for meeting the originating cognitive need. The two types of choice are not actually clearly distinct, as discussed in Section 6.

Summary of main points in Section 4

* Doing a complex dynamic task involves meeting several main classes of cognitive need
* These main cognitive needs may be thought about in a very varied order.
* 'Sequencers', made up of conditional and ordinary processing elements, make the decisions which determine the sequence of thinking about the main cognitive needs
* The 'sequencers' contain the main cognitive needs, i.e. the elements which call on the main 'routines'.
* The main data items in the working storage in the 'sequencers' form a general overview of the state of the task, which is continually available to all processing.
* The data found to meet the main cognitive needs in the 'sequencers' are also the items which are referred to by the sequencing decisions, so they act as the context for those decisions.
* There is a cycle in processing : the overview determines which cognitive needs are thought about. These needs call on 'routines', which update the overview, and so on.
* The factors involved in making a decision about what to do next are considered in parallel, unless the necessary data is not available, when a 'routine' is used to find it. This flexibility between parallel and serial operation is implemented by the two arrow box.
* So this cyclic contextual processing can act as a serial processing device when there is no context or when the contextual overview has not yet been built up, even though this standard sequence of activity is not explicitly represented in the working methods.
* A mechanism is also needed by which highly salient environmental information can override the overview and its control of processing.

©1998 Lisanne Bainbridge