Computer Supported Cooperative Work - A Framework

this chapter published as:
A. J. Dix (1994). Computer-supported cooperative work - a framework. In Design Issues in CSCW Eds. D. Rosenburg and C. Hutchison. Springer Verlag. 23-37.

the framework has been used in various places including:

  • A. Dix (1997). Challenges for Cooperative Work on the Web: An analytical approach. Computer-Supported Cooperative Work: The Journal of Collaborative Computing, 6 pp. 135-156. more ...
  • V. C. Miles, J. C. McCarthy, A. J. Dix, M. D. Harrison and A. F. Monk (1993). Reviewing designs for a synchronous-asynchronous group editing environment. In Computer Supported Collaborative Writing Ed. M. Sharples. Springer-Verlag. pp. 137-160. more ...
  • A. J. Dix (1995). Cooperation without (reliable) Communication: Interfaces for Mobile Applications. Distributed Systems Engineering, 2(3): pp. 171-181. more ...
  • A. Dix, J. Finlay, G. Abowd and R. Beale (1998). Human-Computer Interaction, second edition. Prentice Hall. more ...
  • xxxx
  • xxxx
  • xxxx

Alan Dix

1. Introduction

This chapter presents a framework for considering different aspects of CSCW and different CSCW systems. We are also able to consider systems not normally regarded as having a CSCW element and consider their positive and negative impact on cooperation. The aim of the framework is to enhance understanding of existing areas and hence both to develop them and also to suggest new avenues for research and development. In the framework we shall place existing concepts such as computer mediated communication, deixis and communication through the work itself. We shall also consider many familiar shared tasks from both CSCW and non-CSCW systems, including meeting rooms, shared editing, piano moving and barcodes. The significance of this work is not in the particular examples presented, but in the structure within which they sit.

Although the original purpose of this framework was rhetorical, it has also become a potent tool for design. In Miles et al. (1993) the framework is used to structure the issues surrounding the design of a synchronous/asynchronous group editor. In addition, in Dix et al. (1993) it has been used to structure an introductory survey of groupware systems. Other surveys have been based around the ubiquitous time/space matrix and functional decomposition (e.g., Rodden, 1991). This framework deals with the participants and entities in a cooperative work situation and their relationships. It does not explicitly address temporal issues or the physical location of the entities and it therefore complements the standard time/space matrix. The most similar work is probably Lim and Benbasat's framework which also concentrates on participants and entities (Lim and Benbasat, 1991); however, their approach is based around the computational entities. The emphasis in this chapter is upon the work situation and those objects which, computerized or not, contribute to cooperative work.

The chapter will proceed by dissection. We take the phrase "computer supported cooperative work" and pull it apart - first, looking at cooperation, then at cooperative work and finally at various forms of computer support. This sort of dissection is not really a particularly valid form of analysis. CSCW obviously has a meaning over and above the particular words of which it is composed. My only excuse is that it is useful and draws out the points I wish to make. In particular it helps us to focus on the importance of work - the purpose of cooperation. Also, having started from the words "computer supported cooperative work" one feels somewhat justified in using the resulting framework in a normative fashion; criticising systems which fail to address important areas.

2. Cooperative Work

2.1. Cooperation

The word cooperation suggests two or more participants communicating with one another (fig. 1).

Figure 1. communication

Inter-personal communication is a complex and interesting subject in its own right. For example, conversational analysis and speech-act theory both look at the structure of utterances within normal conversation. For discussion of the role of the former in CSCW, see Morris et al. and Proctor in this volume and for the latter see Winograd and Flores (1986), who used speech-act in the design of the Coordinator message system. At a different level entirely, information theory measures the raw capacity of channels (Shannon and Weaver, 1949), and between these extremes, the author has himself studied other temporal properties of channels (Dix, 1992). Important as communication is, the purpose of this chapter is to show that it is but one part of the study of cooperative work.

Part of the job of communicating is coming to a mutual understanding about the subject of conversation. This need not be an agreement about matters of substance, but at least a sufficient understanding about one another's positions. Indeed, communication can be viewed as the process of establishing mutual understanding and Clark and Brennan (1991) refer to this understanding as common ground. Understanding can thus be seen as an aspect of communication, but additionally it will be presented below as a facet of work.

Important as they are, communication and shared understanding are only a beginning. When we talk of cooperation, we need to know what we are cooperating over, what we are communicating about. Cooperation is never just communication, it is communication with a purpose. That purpose is work.

2.2. Work

I'm not sure where politics fits into this spectrum, surely not under intellectual labour - perhaps entertainment?
Work takes many forms: the production of goods (for example, by farmers and factory workers), the use of those goods in service (e.g. bus drivers), intellectual work (management and academia) and entertainment. This work is carried out with, or is performed upon, various artefacts of work: tractors and turnips, business plans and books. These artefacts are denoted A in the diagram. The artefacts of work are controlled and acted upon by the participants, who are also able to perceive the state of the artefacts. This control and observation is denoted by arcs between the participants and the artefacts (fig. 2).

Figure 2. shared artefacts of work

Sometimes an artefact is shared between the participants, that is they both exercise an element of control and observation of the same thing. For instance, (and we shall return to this example several times) when moving a large object such as a piano. In other cases, the different participants each have control of different artefacts.

The sharing of artefacts is not necessary for cooperation. The simplest case is where only one participant is in contact with the artefacts (fig. 3).

Figure 3. one participant controls artefacts

A timetable enquiry is a case in point. Here the enquirer asks for the time of the train, at which the officer looks up the time in the timetable (paper or electronic) and relays the information back to the enquirer.

Another example is in expert-novice situations, where the novice is in physical control of the work and is advised by the expert. Sometimes this is for the purposes of instruction, in these cases the aim of the work is establishing understanding and skill in the novice, the artefact of work is a tool to that end.

There are also many expert-novice situations where there is little or no didactic element. A short while ago I was rung up by a colleague who had a problems with a computer program. While we talked he recompiled the program several times and tried various experiments - eventually finding the bug. The telephone is not the best instrument for such remote debugging; ideally one would like direct contact with the target system. However, despite this, the exercise was surprisingly successful.

I was once at the novice end of a similar, but slightly different, form of interaction. This time the artefacts were a piece of complex electronic equipment and various tools such as logic probes, ammeters and an oscilloscope. The manufacturer of the equipment was on the other end of the telephone line in California with a similar set-up and gave us instructions, performing similar tests on his (working) equipment until the fault in ours was found.

In this example, we see that we each had control of distinct artefacts of work. We could only observe the effects of our own actions, and only had knowledge of each other's through our conversation (fig. 4).

Figure 4. participants control different artefacts

Sometimes the picture may be further complicated as one participant may have control of an object, but have no direct observation of it. The other participant may be able to observe it, but have no control except through the first. An example of this is the control of large ships; the captain on the bridge observes the position and speed of the ship, but has to exercise control via the sailors in the engine room.

In general, a work situation will include all types of artefacts, some shared between all participants, some controlled and observed by just one or a group. Normally, those with control will also receive feedback as to their actions, but as in the ship, not always.

To denote this generic situation, we will use the original diagram with arcs between both participants to the artefacts. However, we should remember that this does not imply all such artefacts are shared, and certainly not that all participants have equal access and control of all artefacts. It also emphasises that even where there is not shared control of the artefacts, the artefacts are the shared focus of the work, and furthermore that the work is the focus of the cooperation.

2.3. Communication through the artefact

Where an artefact is shared, that artefact is not only the subject of communication, it can also become a medium of communication. As one participant acts upon the artefact, the other observes the effects of the action. We can call this observation by the other participants feedthrough. Just as each participant receives feedback as to their own actions they receive feedthrough as to the actions of their colleagues. Because of this feedthrough, the participants are able to communicate through the artefact (fig. 5). Consider, for example, two people moving a piano: as one of them lifts or shifts her end the other will feel the movement in the piano and shift his hold correspondingly. No words need to be spoken, the actions are eloquent enough.

Figure 5. communication through the artefact

Communication through the artefact can be very powerful, as the piano moving example demonstrates. The communication medium is obviously well suited to the task in hand. How successful it is depends on the nature of the task. Solid physical objects are obviously particularly well suited. At the other extreme, if there are no shared artefacts there is no possibility for communication through the artefact. In between these extremes are shared artefacts which are more or less tightly knit.

Imagine our piano movers again, let's call them Sue and Keith. Now they have finished getting the piano into the house and are bringing in the mattress. This time, as Sue moves her end Keith does not feel what has happened because the mattress bends in the middle. Each gets feedback as to their own actions and tends to pace their work in relation to that feedback. However, there is little feedthrough informing them of the other's actions.

To simplify the situation imagine two people (blindfold and silent) holding opposite ends of a solid pole - they could communicate using simple Morse code and develop quite a speed at it. Now swap the pole for a 'slinky' spring. Possibly one could feel the waves propagate through it, but the pace of communication, and particularly the pace of interaction would drop significantly. The pole and the piano are examples of tightly knit artefacts where feedthrough is almost as strong as feedback. The slinky spring and the mattress are less tightly knit; feedthrough is far weaker.

2.4. Understanding - Soft and hard artefacts.

We have already mentioned understanding in the light of communication for cooperation. This may take two major forms. There is understanding about the work. This is most closely tied to the communication itself, and is a means to the completion of the work. On the other hand, understanding may be a goal of the work. This is the case in academic work: one communicates to share or discover some understanding of a domain and understanding is the sole purpose of the work. We may thus see understanding as a form of artefact of work (fig. 6).

Figure 6. shared understanding

Understanding and other intellectual-ware may be seen as soft artefacts as opposed to the physical or hard artefacts. It is often the case that during the process of intellectual work one externalizes the understanding in some form - the production of academic papers, management reviews, use of a whiteboard or demonstrations using models or computer simulations. This representation of soft artefacts as hard ones may be merely the communication medium where the real aim is the understanding itself, or may be seen (as with the production of a shared paper) as a primary goal of cooperation. Further, the existence of the external reified form may serve to facilitate understanding among the participants.

2.5. Deixis

As we have seen cooperation cannot be seen as communication alone, but as communication with a purpose. In particular it is communication about work and about the artefacts of work. During this process the participants will use various means to refer to the particular artefacts - this is deixis. Deixis is shown in figure 7 as an arc from the communication to the artefacts of work.

Figure 7. deictic reference

In conversation where the artefacts of work are present, one is able to indicate the different objects by pointing, or even just by movement of the eye. These indicative movements may be accompanied by phrases such as "put that one there" or "let's use this". If one is writing, using a telephone, or just talking about the objects when they are elsewhere, deixis is more difficult and one is forced to use more complex expressions - "turn the left hand knob with the red dot on it". This is not confined to physical objects: in academic papers, one cites other papers, and organisations usually have sophisticated labelling schemes to enable them to refer to specific letters, bills, people etc.

3. Computer support - communication

We can use this diagrammatic representation of cooperative work to investigate different aspects of CSCW. For any CSCW system, we can ask precisely which aspects are computerized?

We shall move through the diagram looking at the different arcs (representing communication or control) and see what sort of systems computerize those arcs. We begin at the top of the diagram with the inter-participant communication and establishment of understanding. Then in later sections we shall look at the artefacts of work, whether or not they are themselves computerized, and at the results of automating the control of the artefacts.

Figure 8. candidates for computer support

Looking at figure 8, we see that cooperative work has many facets. If the term CSCW is at all meaningful, we should expect that any system or discussion which claims to address CSCW should relate to the larger picture, not just one aspect or another. This may not be entirely fair, part of the valid study of any discipline is the study of quite specific aspects of it. It is at least useful, however, to assess any particular system in the light of the whole - does it support cooperative work?

This can then be a touchstone with which to judge the various approaches to CSCW which emerge as we focus on the specific areas of the framework.

3.1. Computer-mediated communication

The participants themselves are assumed to be human, so the first candidate for computer support is the communication between the participants (fig. 9).

Figure 9. computer-mediated communication

There are many systems which fall into this category: email, bulletin boards and some electronic conferencing systems. There are many important issues in this area: how to compensate for lack of face-to-face meetings (heaven forbid substitute) or how to improve existing communication-at-a-distance such as letters.

If we stay just with this arc, we are talking about computer-mediated communication (CMC) not CSCW. That is not to say that CMC is not a part of CSCW. It is a valid CSCW activity in the same way as, say, screen design is part of HCI. However, it does make one wonder about the very large place CMC has in the CSCW literature.

A CMC system will have an impact on the cooperative work situation whether or not it is designed with that intent. The introduction of new media has profound effects on the nature of conversation. Different conversational structures encouraged by the media, perhaps branching conversations, or the permanence of the media will be more or less useful for different activities and may change the whole social dynamics of the conversation (Clark and Brennan, 1991). Thus while the role of CMC as such is debatable, the impact of CMC is most definitely a central CSCW issue.

There are another set of CSCW systems which go beyond CMC. These are of the ideas-generating kind typified by various forms of electronic meeting rooms such as Colab (Stefik et al., 1987). These systems include the support of mutual understanding as well as communication. The activities they support have understanding as their principal or sole activity. Understanding is the artefact of work. These systems can then justifiably be called CSCW systems with a (or several) capital C, in that they support cooperation not merely communication.

3.2. Anti-CMC

So, despite the feeling one might get from many 'CSCW' systems, CMC is not sufficient. Neither is it necessary: many good CSCW systems do nothing to support direct communication. Such systems which have no communication element, one could call anti-CMC systems (fig. 10).

Figure 10. anti-CMC

Examples of such systems include Timbuktu (Farallon, 1987), Shredit (CSMIL, 1989) and other shared screen applications. These systems all implicitly assume that the participants will have some additional means of communication: face-to-face, audio or audio/visual. Indeed, there is no particular reason why such systems could not be combined with an on-line electronic conferencing system. Nevertheless, the systems in themselves only support a shared artefact. There is, of course, communication through the artefact: the participants can read each others typing and share a focus of attention. Indeed, there have been attempts to use such systems as the sole means of communication. But, this is not their design intent.

The electronic meeting rooms are similar in some ways to shared screens. However, the emphasis in the meeting rooms is that the shared surface is the focus for establishing common understanding and externalising that understanding. It is therefore explicitly supporting communication even though additional (verbal and non-verbal) forms of communication are expected. On the other hand, the emphasis of Shredit is on the document as an artefact, the establishment of common understanding is a side-effect of collaborative editing.

The anti-CMC systems have been introduced here as an example of the fact that a CSCW system need not be a CMC system. The example of Shredit is one where the artefact of work is computerized. We look at general issue of the computerisation (and not) of artefacts in the next two sections.

In Dix (1991), I examined the distinction between communication via. messages as compared to using shared data. At first sight these categories appear to correspond to the direct communication between the participants (which is often message based) and through the shared application objects. This is, however, an oversimplification Looking at direct communication, systems such as bulletin boards and electronic whiteboards act as direct inter-personal channels, but are a shared information source. On the other hand the movement of another participant's cursor in a closely coupled shared editor clearly acts through the artefact and yet may act as a message. This should therefore again be seen as a closely related but distinct categorisation of communication.

4. Computerized artefacts of work

So, we have seen that the communication between the participants, about the work, may or may not be computerized. If we look down the diagram to the artefacts of work, these too may or may not be computerized. This in itself is not a CSCW issue, but one about automation in general. However, if the artefacts are part of a cooperative work situation (as are most), then we must look at the impact of automation on cooperation.

We have already seen one example of an electronic artefact, the document in Shredit. Other examples come from virtually any area of computing: databases, spreadsheets, financial information, stock control, anywhere were the participants' focus of work is on the computer.

The final example, stock control, is interesting as it reminds us that although the computerized artefacts are the local focus of work, they are often (probably always) a means to an end. If one looks deep enough there are physical artefacts of work behind the electronic ones. Stock control systems remind us by their name that they are for the efficient recording of physical stock. At some different level of analysis we may prefer to think of the stock control systems as a communication medium about the stock, or as medium of control of the stock.

Traditionally, cooperation has not been a major concern in the computerisation of work objects. They were either centralized in a DP department, or more recently individual property. To the extent that they fitted within a cooperative framework, this was an externally imposed, organisational framework and not an attribute of the systems themselves.

Systems can be built which recognize that they are part of a wider, group context:

Support for sharing - It is now accepted that information systems, especially databases, are likely to be shared between various members of an organisation. Both databases and network file-systems, are built with a range of locking mechanisms designed to minimize effects of this sharing. The intention is to stop damage to the data or confusion of the users from inadvertent sharing. They thus support sharing, but not cooperation.

Environments for cooperation - A long term aim is to look towards environments where the sharing of information is a focus for communication and cooperation. This means not having a conferencing window where one can communicate, but having communication a seamless part of the electronic environment. In particular, we need support for deixis, within the electronic medium, with references from conversations to artefacts, and notes attached to the artefacts themselves.

We have started along this path in the design of a 'Group Editing Environment' (Miles et al., 1993). Attaching conversations to sections of document makes the document a focus for cooperation as well as the outcome of the work. However, if similar mechanisms of cooperation are to become an integral part of all systems the whole architectural design may have to change. Locking mechanisms introduced to deal with clashes between users must be rethought completely to support cooperation between users. These mechanisms must become pervasive if users are to be allowed to cooperate over their work, rather than periodically use special cooperative applications. By analogy, we would find it rather hard if we were only allowed to talk to our colleagues in special 'cooperation rooms' and not allowed to take in any of our normal working equipment and papers with us.

There are examples of systems which move towards the ideal of an environment for cooperation. The Quilt co-authoring tool (Leland et al., 1988) allows email messages to contain navigable references to documents and automatically posts email upon certain changes to the database. Quilt, of course, addresses a specific task, however, more generic tools must cope with the fact that most applications are not group aware.

Despite the limitations imposed by existing single-user applications, some progress can be made. Some commercial email systems allow enclosures - the inclusion of arbitrary application objects such as spreadsheets or documents. These enclosures can be navigated by the recipient invoking the appropriate application. At the interface level, shared window systems such as Shared X (Gust, 1988) allow users to cooperatively use single-user applications.

The computerisation of artefacts of work may allow sharing of data where there was none before. This then becomes a focus for cooperation and, of course, allows communication through the artefact. For instance, a shared calendar system, such as that described by Beard et al. (1990), makes information available to the group which previously was isolated in individual diaries. Network based object linking such as publish and subscribe under Apple's System 7 operating system will allow and encourage sharing of data between different applications and users. Although these facilities are not really group aware at the moment, one can imagine them evolving to include specific group features whilst third party applications gradually adapt. Within research systems DistEdit (Knister and Prakash, 1990) is a good example of this approach as it supplies an underlying group text object which existing editors can be relatively easily adapted to support.

5. Non-computerized artefacts

Despite the current emphasis away from traditional manufacturing industry towards financial, clerical and information jobs, the ultimate end of most of these is physical. We eat bread, not EPOS and fly in aeroplanes, not flight control systems. As an information technologist (that's what my job says I am), one is hardly used to moving outside the electronic and magnetic medium, and yet that is where real life is lived.

If the artefacts of work themselves are not computerized how is the process of cooperative work supported by computer. First, we may use electronic means to talk about the real world. For instance, a shipping company may use email (but more likely fax!) to discuss cargo details. With no added connection to the job at hand, this is CMC pure and simple. In addition, one may have electronic representations of the real world, as with the stock control system. But, as we noted previously, if the focus of work is the stock itself, then the stock control system begins to become a means of communication. It does differ from the simple email system in that it is a medium of communication which directly supports the task. That is, it is an embodiment of understanding of the task.

Moving on from the communication about and understanding of the artefact, we have two more candidates for computerisation: the communication with the artefacts (control and feedback) and deixis (reference to the artefacts). We will look at these below, however, to some extent, the stock control system can be said to support deixis, as it creates an electronic counterpart of the physical artefact which can then be used as a focus for communication about the artefact.

5.1. Prosthesis

Extending the meaning of the word, I use prosthesis to mean the automation (whether mechanical or electronic) of the control of the artefacts of work. Blurring the computerisation/mechanisation distinction somewhat, we can look at various examples of prosthesis. The forklift truck is a good example. Rather than lift boxes directly, the driver lifts them indirectly using the truck. For a slightly more high-tech example, there are many situations, such as handling radioactive material or working on the sea bed, where it is dangerous or impossible to work directly. Often mechanical arms are used, where a mechanical arm in the dangerous environment follows the movements of a control arm moved by the operator.

These aids are, of course, very necessary, but they are clearly not the same as directly handling the objects. In particular, there is a loss of feedback and feedthrough.

Feedback - The operators rarely have any tactile feedback of their actions. Instead, the operator must rely on visual feedback. In remote situations, even the visual contact is mediated by television cameras removing many of the clues (e.g. stereo vision) of direct visual contact. This in itself is a problem for individual operator control of the environment. In this respect it is likely to be recognized and some attention paid to it.

Feedthrough - As well as experiencing the effects of our own actions, we experience the effects of others' actions on the environment. If two users cooperate using prostheses then as well as their individual feedback being reduced, they will also loose the feedthrough through the artefact. That is, their ability to communicate through the artefact is reduced. This effect is much less likely to be recognized than simple loss of feedback.

Compare two persons moving an object, say a piano, by hand with them moving it with the aid of two (small) forklift trucks. The tasks will be very different for many reasons: the piano is heavy, the forklifts haven't many degrees of freedom. However, if you listen to the carriers other differences become apparent. In the manual case, much of the time you hear little (except grunts). As they manoeuvre it round corners, one of them (probably the one in front) may tip it and the other, feeling the movement, moves in concert. With the forklift trucks, (over the roar of the engines) the carriers tell each other their intentions, and perhaps rely more on visual feedback as well. This is an over simplification, in both cases they would talk about strategic decisions "shall we tip it on end to go through the door". The difference is most apparent at the tactical level of minor movements.

In the above example, we saw the principal effect of replacing direct control with prosthesis. Communication through the artefact was replaced with communication about the artefact.

5.2. Implications for design

Realising that prostheses may reduce feedthrough can guide us in our design of systems where the control of artefacts, or the artefacts themselves are computerized.

For an example, compare a traditional filing cabinet containing paper files with the equivalent system on database. This is an example where the whole artefact is automated however some of the most important points are due to the effect of prosthesis.

Consider first the original paper system. Imagine you go to the filing cabinet to retrieve a file, and find it missing. Depending on the urgency you may either leave it till later, or go and find the file on a colleague's desk. In either case you will have obtained useful information: in the former that someone else had an interest in the same file; in the latter who that person was. In addition, in the latter case, you would have an opportunity to discuss the mutual interest in the record. In both cases, the physical nature of the record acted as a communication mechanism registering interest about the particular record. Even if the record were not in use, a paper record lends itself to the addition of ad hoc comments as a communication between different users of the record. Also, the very act of going to a central filing cabinet may form the locus for less directed, perhaps social communication.

If these informal channels of communication are recognized by the organisation, they may be formalized and official mechanisms introduced for recording interest and relevant information. However, it has been repeatedly noticed that, however well such a formalisation is done, the informal contacts within an organisation are crucial to its well running. Indeed, the formalisation of such channels may stifle the very informal communications it is meant to capture. With the trend to flatter management structures this factor can only increase in importance.

The equivalent database system loses these informal channels of communication almost entirely. Locking mechanisms exist, but as we have previously noted, these are aimed at preventing the problems of users concurrently accessing data, rather than treating this as an opportunity for cooperation. They do at least give some indication (as did the empty filing cabinet) that someone else is interested. But, whereas with paper records, one might expect the record to be out of the filing cabinet for several hours or even days, the typical database transaction will take minutes or seconds to complete, thus the likelihood of users noticing concurrent access is minimal. Further, multiple readers of files are not usually locked and thus users concurrently referencing the same file will be unaware of one another's interest. See how this is a result of loss of feedthrough. In the paper case the actions of one user upon the file (removing it to read or update) had an effect upon the other users. The electronic database has removed this feedthrough and thus the resulting communication through the artefact.

In "Information processing, context and privacy" (Dix, 1990) I discuss the importance of such annotations in ensuring the correct use of personal information.
Of course, one usually regards the lack of a wanted file as being a problem for the users. Thus database systems attempt to give the illusion that one is the only user. The very success of this illusion is the problem from a cooperative work standpoint. The challenge for the designer of such a system is how to maintain the advantages of the electronic database, but at the same time to provide some replacement for the informal channels of communication focused on the file as an artefact. For instance, one could imagine a system whereby you brought files that you were interested in from the database onto your (electronic) desktop. The file would also remain in the database but with an indication for subsequent users that you were accessing it. As in the group editing environment (4), conversation areas could be associated with individual files. In line with our previous comments, one would hope that such a mechanism would be integrated with the whole communication and cooperation environment.

A similar story could be told about process control systems. In a manually controlled plant, as you go to operate a valve you notice the state of other bits of equipment, perhaps talk to other workers who are nearby. Special circumstances are obvious, such as other workers engaged in maintenance. The electronic control room divorces the worker from this contact. Feedback is supplied by sensors, and indications of maintenance by written reports. Somehow, this lacks the immediacy of physical presence and the paucity of feedback is recognized as a problem. Improved interfaces are suggested as solutions (including aspects of virtual reality) and some plants have included microphones at strategic points to capitalize on the important role of aural feedback.

These attempts to improve feedback for the individual operator may have positive effects on the feedthrough between operators. However, the control system designer should be making explicit provision for communication to replace that lost by the introduction of electronic prosthesis. For example, this may suggest the redundant replication of information for different operators, so that they have indications of one another's activities - even where these are not obviously relevant.

Indeed, there may be opportunity for improving the communication through the artefact compared to the manual situation. Such plants are by their nature complex. Manipulating controls in one part of the plant may have effects in a distant area. Also most of the activity is inside sealed tanks and pipes and one is only indirectly aware of this via dials and sensors. The way to control the pressure in a tank, say, may be via a valve positioned in another area. The operator would need to read the pressure dial on the tank, go to the valve, adjust it and then return to obtain the new reading. The processes is further complicated by the various delays in pipes and stabilising of reactions. Thus the feedthrough in such a plant (and feedback) may already be poor. An electronic control room could help by presenting both physical and logical views of the plant. Representation of copresence (as suggested for the database) could be used to initiate conversation as operators focused on logically connected areas.

In both examples, the database and process control, potential solutions have been posed for the loss of feedthrough introduced by prosthesis. However, the particular solutions are not as important as the general design issue of loss of communication through the artefact, and how we can compensate for its loss.

5.3. Success story - barcodes

So having seen how the introduction of prostheses can cause problems for cooperation, it's time for a more positive slant, in fact, a CSCW success story - barcodes. Now I think it fair to say that barcodes are rarely regarded as an aspect of CSCW, however they form a focus for cooperation between many workers.

If one considers the production, distribution and sales of goods, there are vast numbers of workers involved. Barcodes form an important locus for much of this activity. If we begin at a supermarket chain, goods are scanned on the shelves and as they pass through the tills. This information is used in order to help in the restocking of shelves from store and for the stock control within the local store. Further the resulting information is used centrally by the chain's management to control shipping of stock to different stores and in order to develop suitable purchase policies. Information flows the other way too as barcodes are used to communicate pricing information to the tills.

The level of international cooperation in the production of barcodes is amazing, especially when compared with other computer standardisation. A barcode can identify the country, producer and product line. It can be used then to refer to the product, from its manufacture or packing in a factory right through to the till where it leaves the marketing system.

The level of cooperative activity is perhaps rather diffuse compared to say shared editing, but the sheer scale more than compensates for this. Barcodes should surely be seen as an important case study in CSCW.

Where exactly do barcodes fit into the CSCW framework? The important artefacts of work are physical - the goods and products. However, their usefulness is greatest when many of the soft artefacts - stock control systems, sales and purchase ledgers etc. - are computerized. The barcode is a form of deixis relating the physical artefacts and the logical artefacts which describe them. Arguably, this deixis arc is the most important in the CSCW framework, as it relates the world of work to the world of communication - it allows us to talk about the things with which we are working.

6. Summary

We began by looking at various elements of a general cooperative work situation. This framework has enabled us to examine specific areas for computer support and place existing systems by what aspects of the framework they cover. In particular, we have seen how computer mediated communication should be seen as just an element in a cooperative work situation and as an element which should be integrated into a general 'environment for cooperation'. We have discussed the importance of communication through the artefact and seen how this can be lost by automated prostheses with their loss of feedthrough. Finally, we saw how barcodes could be regarded as an important example of CSCW in that they aid deixis.

The general message is that systems to support cooperative work should take a larger view of cooperation, and that system not specifically designed to support cooperation should consider their effects on cooperative work.

If we look at the term "computer supported cooperative work", we see that work comes at the end. The take home message is that work should be at the centre of CSCW.

7. Acknowledgements

The author is funded by SERC Advanced Fellowship B/89/ITA/220. The author also wishes to thank his colleagues on the 'conferencer' project and all those who have commented on previous draft and oral presentations of this work.