Although eminent writers such as Kay and Nelson (see Chapter 2) have attacked the use of metaphors in human-computer interfaces, this thesis has demonstrated that they are impossible to avoid. Indeed, some studies reported in Chapter 2 suggest that metaphor is more than a central feature of communication and may be fundamental to human thought. Attacks on the use of metaphor appear to be based on particular examples of bad metaphors or employ a limited view of what a metaphor is:
Here is the problem with metaphors: you want to be able to design things that are not like physical objects, and the details of whose behaviour may float free, not being tied to the details of some introductory model. Metaphors are like WYSIWYG: useful in limited contexts, but ultimately a drag, a dead anchor. (Nelson 1990, p.237)
Metaphors do not have to be based on physical objects, as shown by examples such as 'demons' and 'wizards', but even where they are, their functionality can be transformed through 'magic'. There is also the question of what to use instead of a metaphor; as I showed in Chapter 2, Nelson's suggested replacements for current interface metaphors are themselves metaphors, though perhaps better ones. Earlier literature, particularly the work of Carroll described in Chapter 2, showed that metaphor can act as a powerful mechanism for learning, but metaphors can also interfere with the learning process leading to problems for the user. The aim of this thesis has therefore been to accept that interface metaphors exist, try to understand their role in HCI, and find ways in which to assess the suitability of an interface metaphor in a given situation. If metaphor is central to thought then the limitations of metaphors are ultimately the limits of human thought.
In a discussion of metaphor in film, Whittock (1990) suggests that the study of metaphor could proceed in two directions, towards cognitive psychology or towards "rhetoric and strategies of communication." Previous studies of metaphor in the human-computer interface have concentrated on the first of these approaches; I decided to examine the second. The rhetorical approach is descriptive, classifying the devices used in communication, with particular concentration on non-literal devices, such as metaphor, which are known as 'tropes.' This descriptive approach was applied in a trope analysis, described in Chapter 4, which showed the widespread use of metaphor and metonymy in MS-DOS, even though this is not generally seen as a metaphor-based interface.
Modern studies of metaphor have developed beyond rhetorical classification towards more complex models of communication, a field known as semiotics. Like rhetoric, semiotics sees metaphor as one of many tropes available for our communication, although most writers recognise it as one of two master tropes, the other being metonymy. Metaphor, in particular, is seen as central to the development of language and that of comparable semiotic systems. Following this direction, it becomes impossible to study metaphor as an element in its own right; it must be considered as one part of the semiotic process. Conversely, any principles or techniques which help in understanding the semiotic process will be of particular help in understanding metaphor.
A semiotic model was developed in Chapter 3, based on studies by de Saussure and Barthes, together with Eco's Theory of Signification. Other researchers have used Eco's Theory of Sign Production to examine the way in which a designer expresses concepts through the interface. This thesis concentrates on the perception of the interface by the user, a process known in semiotics as signification. Signification is a recursive process leading to many layers of meaning inherent in even the simplest interface element. These layers of signification (LoS) are not alternative meanings; they co-exist in the user's mind, affecting the way in which the user will approach an interface or a command and the way it is used.
Having proposed the LoS model as a semiotic approach to metaphor, a number of questions and assertions were put forward. It was important to compare the semiotic approach with that of cognitive psychology. One previous experiment had used an approach based on mental models to compare three alternative interface metaphors with very positive results. However, the original experiment was carried out with computing postgraduates and compared three systems with extremely limited functionality. As described in Sections 4.4-4.8, the experiment was extended to a more realistic manufacturing scenario, with the users needing to carry out three distinct functions to complete the scenario. This experiment showed none of the effects shown in the first experiment, with users failing to form accurate or consistent mental models. Even if the mental model approach was correct, it was too simplistic to account fully for user behaviour.
Three metaphors were used in the experiment, chosen to represent three distinct types of metaphor - spatial, activity-based and interactional - which had previously been identified as distinct categories. An analysis of the subjects' responses showed that they did describe the experiment within the terms of the chosen metaphor category. More importantly, inspection of their descriptions showed that they described the system in terms of different layers of signification; the spatial system in terms of its interface, the activity-based system in terms of its functionality, and the interactional system in terms of the tasks the system could support.
The LoS model was used to successfully carry out a simple semiotic analysis of the Macintosh user interface, described in Section 4.3. Having confirmed its applicability to the task, the 'What for?' technique was developed as a method for designers. A study was carried out with users of two existing systems, one group using a generic application, the other a bespoke system. The technique provided substantial information in a short time, showing its practicality for use by system designers. Content analysis of the results showed that users identified between nine and 27 distinct layers of signification. The potential usefulness of these will be discussed below, as will the potential value of the technique in choosing appropriate metaphors.
Examination of semiotics has shown that the role of metaphor can be seen as part of a wide range of signification of the user interface, in which the interface signifies many different things to the user simultaneously. To reduce the study of human-computer interaction to a single level is likely to over-simplify this very complex process. Other researchers have also suggested more complex structures to describe human-computer interaction, though not in terms of signification. It is therefore necessary to explain what the concept of signification is, and how it differs from other approaches which might seem similar on the surface.
The most important factor in looking at signification is that all the factors concerned are in the user's mind. The signifier is not the physical sign but the observer's most basic interpretation of the sign. There are many signs with multiple meanings, any of which could provide a different signifier. For example, in this thesis, the initials 'PC' would be taken to refer to a personal computer; in other circumstances they could refer to a police constable or to something being 'politically correct'. These not only lead to different significations but actually represent different signifiers. In this thesis, the full term 'personal computer' could have been used to generate a similar signifier to the term 'PC'. However, the exact nature of the sign will always influence the signifier. Although the signifiers for 'PC' and 'personal computer' are similar, they are not identical: use of the term 'PC' might be taken by the user to refer only to IBM PC compatible machines, whereas the same user might see 'personal computer' as a more general term, including other types of microcomputer. The signifier is thus a product of the specific sign and its context.
Signification is the operation which takes place once this initial observation has been made. For example, to some people 'politically correct' might be a term of praise, whereas others might see it as a term of contempt, depending on their personal political views. Occasionally, deliberate puns or accidental confusion might also present the user with two or more signifiers at the same time. Many graphical examples can be found within books on the psychology of perception, such as the widely reproduced figure below, which can be interpreted as a vase (white area) or two faces in profile (shaded area).
Figure 7.1: Example of dual signifiers from a single sign.
Obviously ambiguous signs with multiple signifiers might sometimes occur in user interfaces, particularly with simple icons. However, these do not need any further consideration as the ambiguity will rapidly become apparent when examining the higher layers of signification they lead to. What is more important is the fact that a single signifier can be subject to many different acts of signification, leading to many potential signifieds, whether to different users or as multiple signification for an individual user. To use the alternative terminology mentioned in Chapter 3, an ambiguous sign might be seen as having two or more denotations, all leading to different connotations.
Some discussion, particularly in section 6.2.1, has considered how 'high' a level of signification is. This observation has been based on the use of the term 'high' in general conversation, such as reference to 'higher level motives', and consideration of how closely the signification accords with the immediate use of the interface element (a close accordance being seen as 'lower' than a more general observation). The observations made were subjective and should not be controversial. For example it would be difficult to argue that moving to a different screen represents a higher level of signification than consideration of a company's legal duties.
Care should be taken not to extend this concept too far. Although there is a tendency for users to begin with obviously low level significations and move on to higher ones, there were many examples of backtracking and occasional jumps to alternative significations at apparently very different levels. The process might be seen as analogous to someone starting at a particular tree at the centre of a wood and then wandering around, exploring the other trees. Obviously the perimeter of the wood will take some time to reach and will not be among the first places explored. It is also unlikely that the person will move in a straight line from the centre to the edge. Nearer spots might be overlooked initially and returned to later on; interesting areas might be re-visited on purpose. Also, it is extremely unlikely that the person will visit every single tree unless the wood is deliberately surveyed and mapped out.
Presenting this graphically, we might see a pair of interviews in the following manner:
Figure 7.2: Exploration of conceptual space.
Figure 7.2 shows some of the potential significations for two users and the way in which their interviews might uncover some of them. LoS interviews are undirected and allow the user to 'wander around' their conception of the interface element, moving from one signification to another. Of course, some of the significations will be perceived by the user as more important than others and it is likely that the user will feel drawn towards these, but there is no guarantee that the user will necessarily mention all of them. Repetition of the method with larger numbers of users will obviously help to ensure that lower level significations (those close to starting point) will be uncovered but the method can never be exhaustive, as there is no absolute boundary on the potential conceptual spaces different users might have.
Some aspects of the LoS model are superficially similar to the GOMS model developed by Card et al (1980a). GOMS has formed the basis for many variations and refinements (John 1996) but these do not change the underlying model. This model is based on Goals, Operations, Methods and Selection rules, and depends on breaking a task down into these components to the lowest level, the operations, which represent simple keyboard activities or mouse actions which can be timed. The overall time taken to complete the task can then be worked out, giving a method for comparing the efficiency of different interfaces.
Examination of the results of the 'What for?' interviews in the previous chapter shows that many responses could be categorised as GOMS entities. For example, examination of CU1's responses reveals goals such as, "to bill one part of COMPANY to another part of COMPANY". This goal is broken down into methods, "That's the first button you press, if you like, before going on to the next fields to input invoices.". The methods are also broken down into a sequence of operations, such as "That's the first button you press". Although CU1 did not express any selection rules, other interviewees did, such as AU1, who described his selection criteria for choosing whether to change the name or the type for a file being saved.
There are two very distinct differences between LoS and GOMS. Firstly, GOMS is a top-down approach, taking as given that a user wishes to complete a particular task and that the task is well enough understood for the user to break it down into the components required for its successful completion. By contrast, LoS is unstructured, allowing users to express their own beliefs about the interface and the task. The second difference between the approaches is the prescriptive nature of GOMS which only deals with specific types of entity. The LoS model depends on the signification to the individual user. If a user thinks that the goal is important then he or she will mention the goal; in other cases the sequence of operations or the selection method might be seen as more important and the goal not mentioned.
The first difference between the approaches can be summed up in a simple diagram:
Figure 7.3: Comparison of LoS and GOMS.
From this it can be seen that the two approaches could be seen as complementary and could be usefully employed in conjunction. An example of the use of GOMS is given by Card et al (1980b) in which it was used to analyse a manuscript editing task. This bears comparison with the 'What for?' interviews with AUG as the users in both experiments were using a text-editing application. As the interviews showed, users were using Microsoft Word for a wide variety of reasons not normally associated with text-editing or word-processing, such as conversion of file types for use by other programs, or construction of personal databases. Use of the LoS model on its own can give no information about the efficiency of a particular interface, though it might indicate possible problems; use of the GOMS model by itself might be seriously flawed if the tasks chosen for analysis are based on a misunderstanding of what the users use the application for or why they want to carry out particular tasks. A designer could begin with 'What for?' interviews to develop a fuller understanding of the purposes to which the system is put and the tasks likely to result from them, followed by a GOMS analysis of those tasks to compare the interfaces.
The second distinction between the two methods is perhaps more important. By choosing pre-arranged types of entity, the GOMS analysis takes no account of what is important to the user. Card et al are clear that their text-editing example looks at a routine cognitive skill (Card 1980b p.33 original italics). Many users carry out routine computing tasks, such as CUG who regularly input invoice data on the same system. Many other users, such as AUG, use computer applications for their own purposes. In general, the move from mainframe computing towards personal computing has been a move away from routine use of computers towards adaptive use of computers to meet personal goals. In these circumstances, it is essential to develop models such as LoS which allow the users to express both the differences and the similarities in use amongst themselves in their own terms. The LoS model and the 'What for?' method represent a very efficient approach to gaining a substantial part of this information and may even uncover levels of signification not normally considered in GOMS which could help to provide a wider context for the user's behaviour and motivation. However, the LoS model can never provide a substitute for GOMS; although goals, operations, etc., might be uncovered, LoS is non-exhaustive and important GOMS entities might be entirely overlooked.
Although uncovering layers of signification can lead to the expression of goals or operations, they do not represent a chain of causality. A signified, x, might signify y to a person because of signification z, but this is not necessarily the signification the person will be conscious of. Nor does the LoS model represent a chain of processes (as one might uncover in forming a process flow chart), nor a logical chain. Each separate layer of signification might be formed in a different way, one signification representing a goal, the next layer a causal chain, the next an operational sequence. We are not usually aware of why signification takes place and might never know. For example, people with phobias may have no knowledge of the reason that a particular thing signifies fear to them.
Some of the responses obtained from a user could be post hoc justifications for their actions or their perception. The recursive nature of the LoS is such that the false signification could lead to further levels of erroneous signification. For example, the person who is frightened of spiders might justify this by saying that it is because the spider has eight legs. Although this might be false, the person may then attach the signification of fear to another eight-legged creature, such as an octopus which had never previously been seen as an object of fear. Thus, although some significations are 'false' in one sense, they are true to the user and cannot be ignored by the designer. In some cases, the false signification might appear to lie outside the scope of the interface designer but there will often be something that can be done to deal with the problem. For example, CU5's incorrect justification for the accounting procedures might prompt the designer to change the name of the screen or the system to 'OFTEL Accounting' to make its true purpose clear. This might lead to a better understanding on the user's part but obviously needs to be considered against other constraints.
One of the most notable features of the LoS model and the results of the 'What for?' interviews is the fact that it includes very high levels of signification well outside the factors usually considered by a system or interface designer. A designer might therefore dismiss the approach as irrelevant to the practical issues involved in design. However, to deliberately confine the scope of the model to those factors which the designer knows are important would undoubtedly lead to the model missing out some factors which could be of importance for a particular interface. For example, an interface to a one-off standalone system might require consideration of a manual system it will replace; another might not replace a manual system but could form part of an existing suite of programs with which its interface must be integrated. An effective model must be capable of taking any factors into account, including factors which are currently unforeseen, even if this means the inclusion of irrelevant factors which can then be discarded.
Although not constrained in this manner, the LoS model does have constraints in that it is limited to factors which matter to the user. Although the designer might dismiss certain factors as beyond the scope of the interface design, the fact that the user raises a point is enough to suggest that the designer should at least consider it. The converse does not apply, in that many important factors will not be raised by the users. The LoS model should therefore always be used as one of a number of tools to be applied by the analyst or the designer, as with the example of the relationship between LoS and GOMS outlined above.
As previously stated, it is intended that the approach, particularly use of the 'What for?' technique, should be of practical use to interface designers. In practice, this will mean the simplification of the method to provide 'quick-and-dirty' versions. However, there are limits to how 'quick-and-dirty' any method can be whilst still giving valuable results. The degree of concordance between the interviewees in the two user groups shows that a sample of five users is sufficient to yield useful information. However, it is unlikely that the sample could be much smaller, as each group included one interviewee who ended the interview at a comparatively low level of signification.
One factor which could obviously be excluded from practical use of the technique is the content analysis, which took much longer than the interviews themselves. An analyst or designer could carry out a rough analysis of the interviews, particularly if experienced in the task, but most of the important information should be immediately evident from a simple examination of the users' responses. The most extreme reduction in the method would be for the designer to carry out an examination of the signification to him or herself in the manner of the analysis of the Macintosh interface described in Section 4.3. Such an analysis would obviously be highly biased by the fact that the designer would know the purpose of the interface but could provide a 'first pass' to design a prototype to be used in the user interviews.
This thesis began by considering metaphor in the human interface but may have appeared to move far from this root. Looking at the role of metaphor in general led to consideration of tropes and the semiotic model of communication. Metaphor is a central feature of language and other semiotic systems, as was explained in Chapter 2, and any examination of metaphor depends on a model of semiotic processes such as the LoS. It is only now that the model has been developed and assessed in relation to computer interfaces that its role in explaining the power of interface metaphors can be summarised.
Placing interface metaphors within the context of the LoS model, it is immediately apparent that both the tenor (the interface element) and the vehicle (the concept used to form the metaphor) will have their own layers of signification. Consider some of the problems with interface metaphors discussed in Chapter 2, such as the findings of Carroll and Mack (1984) that the misuse of a text editor was due to the users' adoption of a typewriter metaphor, or the examples of 'male' and 'rape' metaphors from Grundy (1996). The typewriter/text editor problems could be reduced to a simple functional mis-match; the examples from Grundy could not possibly be explained in this way. Both examples, however, make immediate sense when viewed in the light of the LoS model.
With the typewriter/text editor example, the layers of signification for the user are likely to be very similar, although exhibiting a mis-match at some of the lower, functional levels of signification as shown in Figure 7.4:
Figure 7.4: LoS mismatch, typewriter and text editor.
By contrast, the match between the layers of signification for 'abort' and the actual action (designated here as 'abandon') is only successful at the very lowest level, with higher levels being radically different, as shown in Figure 7.5. What these higher levels are will obviously depend on the individual, but for 'abort' these might include moral and religious issues, concerns about women's rights, men's roles, the definition of a human life, and more, none of which would have any relevance to the computing command. In this context, Grundy's extreme reaction to 'rape metaphors' makes perfect sense.
Figure 7.5: LoS mis-match, abort and abandon.
When considering a potential interface metaphor vehicle, many of these layers of signification would be readily apparent to the designer. A simple self-directed semiotic analysis of the type carried out with the Macintosh interface in Chapter 4 would readily demonstrate that the term 'abort' might not be the most appropriate metaphor to choose. However, the designer might come from a background in which the term is more likely to apply to the abandonment of a missile launch than a pregnancy. A more balanced assessment would therefore depend on uncovering the signification of the potential metaphor vehicle through 'What for?' interviews with a range of potential users.
The examples examined in the previous section could also be interpreted in more traditional terms of ground and tension. It could be argued that the mis-match in the 'abort' example comes from the tenor and vehicle having too small a ground and too great a tension, whereas the typewriter and text editor share a greater ground leading to less tension. However, this does not correspond to the traditional definition of the ground as the features common to the tenor and the vehicle. The literal meanings of abort and abandon are extremely similar. Chambers dictionary (Schwarz 1988) gives the definition of abort, when applied to a mission, as "to check or call off at an early stage" whereas that of abandon is "to give up". When applied to abandoning or aborting a computer command, either meaning is literally true - there is not even a metaphor, let alone any tension. However, this only applies to the literal meaning, or denotation, of the term. The mis-match in using 'abort' becomes apparent when examining the connotation of the term.
As stated in chapters 2 and elsewhere, most researchers have seen the value of metaphor in computing as introducing a familiar concept to aid the user in learning a new concept. Obviously this is only possible when the vehicle and tenor have a considerable ground in common. This can be extended further towards what might be termed the 'connotational ground'. The 'What for?' interviews provide a potential method which can be used to compare the signification of a potential metaphor vehicle to the user with the signification to the designer of the intended function or interface component. Judgement of what constitutes the connotational ground will obviously be subjective but significant mis-match will be easily apparent.
Although the designer should be aware of any mis-match between the significations of the tenor and the vehicle, this does not necessarily invalidate the metaphor. Chapter 2 looked at the concept of magic when implementing metaphor-based features. As Kay (1990, p.9) put it, "it is the magic - understandable magic - that really counts... that must be most strongly attended to in the user interface design." When considering this in relation to the LoS model, the critical term is 'understandable'. Understanding can come from explicit training or explanation to the user but should ideally be rooted in the signification.
Consider, for example, the magical features of the Macintosh folder listed by Dieberger (1994a, p.60):
If questioned through a 'What for?' interview, it is likely that most users would mention 'putting things in' a folder as one level of signification. The first two magical features listed by Dieberger extend from this signification and do not conflict with it in the way that the paper feed/ line feed significations conflicted. The contents of physical folders are often sorted into order, for example according to date and it is quite likely that this signification will also occur to users. Extending this signification to automatic sorting is simple and unlikely to cause problems. The fourth feature, searching the contents of a closed folder, does appear to represent a more important mis-match. However, this feature is not actually part of the 'folder' metaphor; it forms part of the 'Find file' command. Options in the 'Find file' dialogue include 'version', 'lock attribute' and other technical terms which owe nothing to the folder metaphor. The mis-match here is not between the signification of the tenor and the vehicle but between the desktop metaphor and the strictly functional 'Find file' command.
In summary, successful 'magic' is likely to come from the designer examining and extending the signification of the metaphor vehicle to the user rather than explicitly denying or ignoring it. For example, it might have been useful to take the chapters of this thesis, each of which is a separate document, place them into a folder and then turn that folder into a single document. This would undoubtedly conflict with the signification of a folder (and that of a document) to most people: in Kay's terms it would be magic, but not 'understandable magic'. If, however, 'chapter' and 'book' metaphors were used instead of 'document' and 'folder', then the operation becomes quite understandable.
Unlike the CUG responses, the AUG responses showed a multiplicity of signification at similar levels. Users talked of using the 'Save as...' command to make an identical copy of a file to another drive (3.3), to change its name (4.1) or its file type (5.1), or to save changes to its contents (18.1). This was discussed in Chapter 6 in relation to the file metaphor but it also leads to consideration of whether a more appropriate metaphor than 'save' could help the users. In this case a clear candidate might be the 'export' metaphor which is used by some other programs. For example, Adobe Photoshop 2.0 has both a 'Save as...' and an 'Export' command on its 'File' menu.
The 'export' metaphor has quite different origins from the 'save' metaphor. Whereas 'saving' implies protecting from changes, or preserving changes which have been made, 'exporting' implies sending something to another 'territory.' This distinction is exploited by Photoshop 2.0 which uses the 'Export' command when, for example, converting a file to JPEG format. JPEG compression results in the permanent loss of graphical information. Because the image is being exported to the 'JPEG world', the 'Export' command cannot be used to over-write the existing file and lose information. In Photoshop a user can make changes to an image, save it to a different file name with the 'Save as...' command, and then close the file. If, instead, the user makes changes to the image, exports it, and then attempts to close it, Photoshop does not treat the changes as having been preserved. Photoshop will not allow the user to close the file without presenting a 'Save changes before closing?' prompt because the file was exported, rather than saved.
The behaviour of Photoshop, with its separate 'Export' and 'Save as...' commands can be compared with that of Word, which attempts to use the same metaphor in all circumstances. In Word, a user can open a Word document, make changes to the formatting of the file and save it as a text file. Word then treats the changes as being saved, even though the conversion to text has abandoned all the formatting information. Indeed, because there is no separate 'Export' command, in the Macintosh version the user can even overwrite the original Word file and lose not only the format changes but all pre-existing formatting information. Consideration of the signification of the 'save' and 'file' metaphors might have helped to avoid the potential problems this raises.
Chapter 2 introduced seven categories of metaphor which can co-exist within the design process and the interface:
Concept: Computer as theatre, interface as facade.
Design: Using metaphor as a 'tool for thought'.
Development: Work-flow, system life-cycle, object-oriented design.
Hardware: Notebook, notepad, pen, organiser.
System: Directories, menus.
Presentation: Documents, filing cabinets, rooms.
Interaction: Direct manipulation, command, conversation.
Support: Interface agents, speech bubbles.
All of these metaphors signify various concepts to the user and could thus be analysed using the LoS model. The last five categories can all be forms of interface metaphor and thus appropriate material for the model. However, it is difficult to see any immediate value in this approach for the three types at the top of the list. In these cases, the metaphors are used to generate an implementation, but will not necessarily be present in the final implementation. It is at the point of implementation that the LoS model becomes useful and it should certainly be applied to any remnants of these early metaphors which are still present. A metaphor might have been a valuable aid in the design phase, but the same metaphor could be confusing to a user whose relationship with the system is very different to that of the designer.
Although consideration of interface metaphors prompted the work that has led to the LoS model, the development of the theory was not restricted to metaphors or to user interfaces. The LoS model is based on a semiotic approach which applies to our perception of any sign, whether it is a metaphor or not. As has been discussed above, this means that the tenor and vehicle can be independently analysed as signs in their own right and the 'What for?' method can be applied to any potential term to be used in computing, whether a metaphor or not, to examine the connotational ground.
The metaphors discussed so far have applied to specific interface elements. This is not an inherent limitation of the LoS approach. Instead, users could be questioned about the signification of a general metaphor to be used as a basis for the total interface. It is probable that useful information about, for example, the signification of desk-tops could be uncovered but this would depend on very careful phrasing of the initial question. A potential user could be asked, "What is your desk-top for?" or, "What are the objects on your desk-top for?" Obviously, the nature of the 'What for?' method is such that, if the user mentions a specific object, the questioning will continue about that object. This does not help to uncover the signification of the desk-top as a total concept for designing an interface but might help to identify the types of object, and their purposes, that could compose a desk-top interface. Consideration of the desk-top concept as a whole might be better considered by varying the initial question. This does not violate the method, in that subsequent questioning would be of the 'What for?' format. Indeed, when looking for a general metaphor, it might be better to begin with a more free-form disussion to find a concept from the user that can be used in questioning, such as 'my bit of the office'.
What the approach cannot achieve has also become clearer. The 'What for?' method is non-directed and non-exhaustive and could never form the basis of an effective design method on its own. It must be seen as an additional tool, widening the range of considerations for the analyst or designer. The greatest strength of the approach lies in the way in which it models aspects of the user's perception that are not conventionally considered; the greatest weakness of the method lies in the fact that this model remains incomplete. It could be argued that this incompleteness does not matter, in that users will always mention their most important significations and that these must therefore form the most important considerations for the designer. However, some significations will never be mentioned, an obvious example being significations that the user is unable to express verbally. However, there is currently no rival approach which offers more likelihood of uncovering these aspects - indeed no other method can even guarantee uncovering the aspects that LoS does successfully reveal. Rather than abandoning the LoS approach the answer must therefore be to continue research into the approach, seeking further validation, and to widen its scope.
The most obvious extension of this research is to interview and compare a larger number of user groups using the 'What for' technique. There would be little to be gained from larger sample sizes but comparison of more user group/interface combinations would help to isolate the effect on users' signification of different factors. This would provide further validation of the LoS model and the 'What for?' technique, in addition to information about both the interface design and user motivation. As a result of this it might be possible to refine the 'What for?' technique or to develop other tools from the LoS model.
Apart from the technique itself, the analysis of the results could certainly be improved. Rather than generic content analysis methods, analysis techniques could be developed which are more closely linked to the LoS model. These could be as simple as a check list of categories that might be expected from any interview, such as the action taken (clicking on a button, selecting a menu item), the definition of the term (as in a dictionary), the consequent action (moving on to another screen), and so forth. Such a check list could be used to immediately screen out more mundane factors, allowing anomalies to show up more easily. Other tools could be used to analyse the interview structure, identifying loops and branches in the responses, and using graphical presentation to assist the analyst's understanding of the information.
The structure of the 'What for?' interview results showed signification at various levels. Within interviews most responses moved from lower levels to higher levels. However, there is no objective measure of the signification level such that responses from two separate interviews could be ranked in relation to one another. Such a ranking would be extremely useful in comparing different implementations. A measure of this type would simplify the content analysis phase and also allow interviews to be objectively compared.
The relationship between the LoS model and GOMS was discussed above and it was suggested that the methods might be used in combination. Further research should examine ways in which the LoS model might be more tightly integrated, either with GOMS or with other equivalent low level design methods. In addition to this, the relationship of the LoS model with other high level approaches requires further study.
It was suggested in Section 7.2.6 that, where potential users are not immediately available, designers might find some value in questioning themselves through the 'What for?' technique. However, this would only be useful if the designer sees an interface in the same way as its users will. The pilot trial of the technique, reported in Section 5.2.4, included interviews with both users and designers of interfaces which suggested that interfaces carried similar signification for both the designer and the user. The technique could be further used in this way to test whether this is always the case. In particular, it is suggested that a mis-match in signification between the designer and the user might lead to an unsuitable interface. The technique could therefore provide a valuable diagnostic for investigating systems which prove difficult to use. By interviewing both users and designers, signification mis-matches could be identified and corrected.
The experiment with three metaphors reported in Chapter 4 showed that different types of metaphor can direct the user towards different levels of signification. Those results were based on open-ended questions in a questionnaire. Use of the 'What for?' technique could be used to explore this further, particularly if the results could be compared objectively as proposed above.
The LoS model also has great potential in the analysis of metaphors before their implementation. 'What for?' interviews could obviously be used to compare the signification of alternative prototype systems but, unlike many other methods, the technique can also be applied to the underlying concept. By asking users about a potential metaphor vehicle independently of its use in an interface, a 'pure' definition could be developed and used to assess the suitability of the metaphor for its context. For example, the users of the 'Navigate' command in the commercial system could have been interviewed by the designer of the system to establish what the term meant to them. Alternative metaphors which could have been used, such as 'browse' or 'go' could then be compared and the best candidate chosen. If no suitable metaphor match could be found, the structure of the system could have been changed to use, say, an index or map based alternative, before any investment in its construction had taken place.
The above discussion of the applicability of the LoS model to various categories of metaphor shows that there is no advantage to the designer in examining metaphors used in the design process. Although this might not help the designer, use of the LoS model could provide a great deal of information for researchers interested in how analysts and designers approach their work. Examining the signification of the system to the designer, particularly in terms of the metaphors used in its construction, would provide a great deal of valuable information which could be used in the development of better design methods. Indeed, once people start using the model, it could be used to analyse the 'What for?' technique and even the LoS model itself.
The studies in this thesis have focused on relatively concrete objects such as interface commands. The use of communications systems involves the user with more diffuse concepts, such as the Web. Unlike the computer systems looked at in this study, communications also involves the interaction of users, each with their own understanding of the technology. The problems of mis-matches between the signification of two users sharing the same communications facility offer a fertile area for further investigation. A particularly difficult issue here is that some of the concepts do not exist until the metaphor comes into being. The Internet existed and a number of sites on the Internet included HTML servers for some time before the concept of the 'Web' arrived; whether the concept existed before the metaphor which named it is more contentious.
The studies in this thesis have all examined particular systems at particular times, ignoring any temporal factors. As was stated in Chapter 2, metaphors gradually die, their signification therefore changing over time. Additionally, the discussion of Carroll's work drew attention to the role of metaphors in the learning process. If this is so, one would expect to see a change in the signification to a user as that user becomes more familiar with the term in its new meaning. Rather than confirming this view, the only relevant finding in my own study was that the most inexperienced user in the CUG was confused by the 'navigate' metaphor, asking whether it meant 'find'. Cornell Way's argument for metaphor as a learning aid, introduced in Chapter 2 was that, "it is easier to take parts from other established concepts than to build up new ones from scratch." (Cornell Way 1991, p.8). The example uncovered in the CUG study was perhaps an exceptional case, in that it is was likely that the term 'navigate' did not form a part of the user's normal vocabulary.
As a semiotic approach, the LoS model is potentially applicable to any part of the communication process. Rather than examining specific metaphors or interface objects, the model could also be applied to other factors which affect the signification of the interface.
One important aspect that should be considered is the presentation of the metaphor. Presentation is particularly important to signification in being the first aspect the user is likely to notice. For example, consider the two heavily reduced screens below.
Figure 7.6: Presentation of metaphors.
It is impossible to read a word on either of these screens or to see what programs may be running but even a cursory glance immediately distinguishes the GUI screen on the left (Macintosh Finder) from the text based interface on the right (MS-DOS). Thus the overall style, or appearance, of the interface creates the initial signification to the user.
Another aspect of the presentation of the metaphor is the degree of realism. In the experiment which examined three alternative metaphors it was found that the spatial aspect of a realistically presented metaphor can be particularly powerful. Sometimes this was taken to absurd extents, such as the requests from two users for a 'remote control' for the 'television' which was always less than two inches mouse movement away. Many metaphor-based GUIs ignore these factors. Consider, for example, the comparative sizes of a Microsoft Word document and a folder on the Macintosh desktop.
Figure 7.7: Document and folder on the Macintosh desktop.
It is obvious that the Word document could not actually fit inside the folder, yet this is what the user is expected to understand. This may not matter in that the MILAN interface used a comparatively realistic perspective view of the objects, whereas the Macintosh desktop uses a flat, two-dimensional view. The LoS model could be used to examine how the levels of signification of metaphors are affected by their presentation, comparing text and graphical presentations and varying degrees of realism.
Some interface metaphors also affect the ordering of the user-system dialogue. Some researchers (Nadin 1988; Foley 1990) have suggested that direct manipulation requires a post-fix syntax, to identify the object then manipulate it, whereas text is better with a pre-fix syntax, corresponding to the typical syntax of the imperative in Indo-European languages, "Do this!" Again, the LoS model could be used to examine the effect that syntax has on the signification of the interface to the user.
Any factor which affects the user's performance will have an impact on the user's efficiency. The effect on performance of the factors discussed above has not been quantified. However, I carried out a simple techno-economic analysis of the potential impact of using the different categories of metaphor looked at in the experiment reported in Chapter 4. Appendix E contains an account of this analysis which was based on a techno-economic model of European industry. The model is not strong enough to be used as a reliable measure of economic impact but it does indicate that the factors considered in this thesis could have an economically significant effect on the take-up of a new technology.