October 1999
Although metaphors are common in computing, particularly in human-computer interfaces, opinion is divided on their usefulness to users and little evidence is available to help the designer in choosing or implementing them. Effective use of metaphors depends on understanding their role in the computer interface, which in turn means building a model of the metaphor process. This thesis examines some of the approaches which might be taken in constructing such a model before choosing one and testing its applicability to interface design.
Earlier research into interface metaphors used experimental psychology techniques, which proved useful in showing the benefits or drawbacks of specific metaphors, but did not give a general model of the metaphor process. A cognitive approach based on mental models has proved more successful in offering an overall model of the process, although this thesis questions whether the researchers tested it adequately. Other approaches which have examined the metaphor process (though not in the context of human-computer interaction) have come from linguistic fields, most notably semiotics, which extends linguistics to non-verbal communication and thus could cover graphical user interfaces (GUIs).
The main work described in this thesis was the construction of a semiotic model of human-computer interaction. The basic principle of this is that even the simplest element of the user interface will signify many simultaneous meanings to the user. Before building the model, a set of assertions and questions was developed to check the validity of the principles on which the model was based. Each of these was then tested by a technique appropriate to the type of issue raised. Rhetorical analysis was used to establish that metaphor is commonplace in command-line languages, in addition to its more obvious use in GUIs. A simple semiotic analysis, or deconstruction, of the Macintosh user interface was then used to establish the validity of viewing user interfaces as semiotic systems. Finally, an experiment was carried out to test a mental model approach proposed by previous researchers. By extending their original experiment to more realistically complex interfaces and tasks and using a more typical user population, it was shown that users do not always develop mental models of the type proposed in the original research. The experiment also provided evidence to support the existence of multiple layers of signification.
Based on the results of the preliminary studies, a simple means of testing the semiotic model’s relevance to interface design was developed, using an interview technique. The proposed interview technique was then used to question two groups of users about a simple interface element. Two independent researchers then carried out a content analysis of the responses. The mean number of significations in each interview, as categorised by the researchers, was 15. The levels of signification were rapidly revealed, with the mean time for each interview being under two minutes, providing effective evidence that interfaces signify many meanings to users, a substantial number of which are easily retrievable.
It is proposed that the interview technique could provide a practical and valuable tool for systems analysis and interface designers. Finally, areas for further research are proposed, in particular to ascertain how the model and the interview technique could be integrated with other design methods.
Some of the earlier work in this thesis was carried out in collaboration with fellow researchers from RACE Project R2094: MITS (Metaphors for Integrated Telecommunications Services). Where specific work was undertaken by other members of the team this has been acknowledged within the main text. It is often difficult to remember exactly where an idea originated and I would therefore like to offer my thanks to all other members of the project for the work they did and my apologies to them if I have inadvertently overlooked any specific acknowledgements which they deserve. In particular, some of the initial literature searches of metaphor theory and rhetoric which contributed to the introductory material in sections 2.1.1, 3.1.1 and 3.3.1 were originally carried out by Roger Knott of Loughborough University of Technology.
Although all other work in the thesis is entirely my own responsibility, I would have been unable to complete it without the help of my supervisors. In particular I would like to thank Professor Alty at Loughborough where I originally registered, and Professor O'Keefe at Brunel where I completed it. I would also like to thank Hans Panse and Stephan Keuneke of BIBA for their assistance in setting up and running the experiment described in sections 4.4 - 4.8. Finally, I would like to thank Mark Perry from Brunel and my daughter Louisa Condon from the University of Sussex for carrying out the content analysis described in section 5.4.
This work is concerned with the use of metaphors in the human-computer interface and their role in the relationship between the user and the computer, and thus falls into the area known as human-computer interaction (HCI). The target area in which it is hoped the results will prove useful is that of systems analysis and design, with particular regard to the specification and design of the interface.
Some of the earlier work described in this thesis was carried out as part of the MITS project (Metaphors for Integrated Telecommunications Services), funded by the European Commission as part of RACE (Research in Advanced Communications for Europe). Although that project studied problems associated with multimedia communications and computer supported cooperative work (CSCW), this restriction does not apply to this thesis.
Many authors have claimed that interface metaphors can be useful to the user, particularly for new users learning how to use a system. Others have identified problems caused when metaphors are taken too literally by users. The preliminary research in this thesis shows that metaphors are common, even in ‘metaphor-free’ interfaces. Other authors have shown that, even where no explicit metaphor has been used in the interface, problems can be caused by metaphors that have been introduced by users. Thus the problem is not whether to use metaphors in the interface but how to introduce metaphors and which metaphors to use.
Designers who have introduced metaphors with the intention of assisting users have done so on an ad hoc basis. A common assumption is that metaphors should be based on familiar aspects of the users’ workplace, such as the materials and tasks found in an office environment. Although this appears to make sense, introducing the unfamiliar in terms of the familiar, there appears to be very little empirical research to test the approach. Much of the argument about the use of interface metaphors, particularly from those opposed to the use of metaphor, has concentrated on graphical user interfaces (GUIs). However, it is often possible to introduce the same metaphor as an icon in one interface and as a word in another. Whether the effectiveness of the metaphor depends on the type of interface remains unanswered.
Ideally, the interface designer needs guidelines on when to use metaphors, which metaphors to use, and how to portray them. Metaphors are rarely introduced in isolation and the designer must also understand how they should be mapped to the functionality of the system and whether they can be mixed with other interface metaphors. Some HCI researchers have offered principles to guide some of these decisions, whilst systems manufacturers often offer style guidelines to ensure consistency across applications. Current guidelines are, however, based on personal experience, common sense and aesthetics. More effective guidance depends on a better understanding of the process of metaphor. It is not possible to directly observe the mental processes through which a user understands and uses an interface metaphor, but it should be possible to build a model of this interaction. A number of disciplines offer possible approaches which could be used to create such a model.
One possible approach might be to look at, and possibly extend, existing HCI methods in software ergonomics and interface design. Many authors offer useful methods for building interfaces and for evaluating their usability, but these are independent of any underlying metaphor. It would certainly be possible to evaluate interfaces based on different metaphors and find which is the most usable. Unfortunately, such data does not show why a particular interface is better. Thus the particular implementation might be more important than the metaphor chosen; it is certainly possible to build bad interfaces with the best of metaphors.
Where metaphor is discussed in the HCI literature, it is generally justified as presenting novel functions in terms of well-understood ones. This is usually interpreted as using metaphors drawn from the environment in which the interface is to be used. However, very little evidence has so far been presented to support this assertion.
The three most important elements involved in this research are the computer, the human and the metaphor. One approach might be to begin with studies of the computer and extend these. Computing and metaphor have always been fundamentally linked. Even the central principles of computing are close to those of metaphor: programming one machine to act as if it is another. Although this indicates how computers might support metaphor, this approach does nothing to indicate which metaphors are likely to be most helpful to users.
Computer science certainly employs many metaphors, such as ‘objects’ but treats them as mathematical constructs. Mathematics is itself largely based on metaphors but having established the original metaphor, mathematics employs abstraction to form a self-contained system. Thus, for example, the mathematical concept of a ‘set’ comes from our common usage of the word in concepts such as a tea-set. While the metaphor may have provided the original inspiration, it is immediately abandoned and replaced by an abstract mathematical definition of the concept. This route is uni-directional: the properties of a mathematical set cannot be used to assess the usability of a tea-set.
Computer science forms a similar type of abstraction which might be useful in creating a formal specification of a metaphor. However, abstraction and formalisation can only occur after the characteristics of a metaphor and the process through which it works have been understood. Computer science offers no techniques for building this initial understanding.
The second element of this study is the human: metaphor is a cognitive process. If we look at the way that metaphors appear to work in the human mind, we find this is easily expressible in terms of mental models. This approach has been widely used in HCI, comparing the designer’s model of the system with the user’s model and comparing both to the physical model. It is usually asserted that mis-matches between the three models can create problems for the user. It is relatively easy to extend this approach to the role of metaphor in the interface by including the user’s mental model of the domain from which the metaphor is drawn.
When a new concept or function is to be introduced it can be attached to this existing mental model using the process of metaphor. Over time, a new model will be created, detaching itself from the original and forming a new, separate definition of the new concept. At least one experiment has been based on this model of metaphor. The experiment was based on the idea that we form models of the systems we use and that the accuracy of these models affects how easily we can use the systems. This approach is further explored in this thesis.
The third element to be considered is the metaphor itself. Metaphor has been widely studied in literature and has been examined in a number of areas of linguistics. The question arises as to whether it is valid to apply studies based on language to computer interfaces. Many interface metaphors are presented as images or actions rather than words and it needs to be established whether this affects the nature of the metaphor. A second issue is whether the elements of a computer interface have equivalent syntactical and semantic power to a natural language, or at least sufficient power to support true metaphors.
Although metaphor was originally studied in the context of rhetoric, literature and poetry, more recent approaches have extended this work to other fields such as advertising and film. If these studies are valid then it is obviously possible to extend this approach to computer interfaces. This type of approach has not been widely employed in HCI and appears to show great potential for the study of metaphor and, perhaps, other aspects of human-computer interaction. It therefore forms the main field of study in this thesis.
As explained when considering the research problem above, the most important concern is that there is currently no effective model of the metaphor process. Both cognitive psychology and linguistic approaches are assessed in terms of their potential for building such a model. Either approach depends on a number of assumptions which must be tested before carrying out any study of the full model.
Two approaches could be made when assessing the usefulness of the model. The first is its potential use by future HCI theorists and researchers, which would depend on its internal consistency and whether the model adds to our understanding of human-computer interaction. The second approach is to consider its potential use by interface designers. The research problem outlined above is a practical problem faced by many designers who need guidance. This thesis concentrates on the second of these approaches, looking for practical justification of the model in terms of its usefulness to designers, although it is acknowledged that this will also depend to some degree on establishing a solid underlying theoretic basis.
A number of preliminary studies must be made before a model can be developed. Short studies are used to assess the potential approaches introduced above. In addition to these, a suitable linguistic or grammatical method must be used to test the assumption that the computer interface can be considered in a comparable manner to a linguistic system. Following the results of the preliminary studies, the most promising approach is used to develop a model of the metaphor process.
It is impossible to test whether the model is ‘right’ in the sense of accurately representing the thought processes of the user, in that many relevant thought processes are not conscious. Other ways of testing the model must therefore be considered. The model could be tested as a predictor of user behaviour but this only provides evidence for its descriptive power. Another approach is to look at the particular model, examining its implications for the analysis and design process and testing those implications. As outlined above, it is the second of these approaches that is followed.
A number of research methods are considered and the most suitable chosen. This is a simple interview technique to be carried out in the workplace with users of two types of application. If successful, this can not only provide evidence for the concordance of the model with the users’ understanding but also provide a demonstration that the model could be used in similar circumstances by interface designers to provide practical assistance in the design of the interface.
The contents of this dissertation are given in approximately chronological order. Chapter 2 looks at the background to the work, considering previous research into computing metaphors and the potential approaches which could be used. Previous work, particularly that of Carroll, is then examined, showing some of the problems and benefits of metaphor in computing. The chapter also examines studies of metaphor by Eco and others, including its role in the development of languages. Lakoff’s thesis that metaphor plays a major role in human cognition is then considered. Finally, the chapter looks at some of the most important examples of metaphor in computing and communications.
Chapter 3 then lays out a provisional model for human-computer interaction, together with proposals for the useful application of this model to interface design. A superficially similar approach is considered and shown to be complementary rather than a direct equivalent. Finally, the chapter lays out the assumptions made in the development of the model and some questions which it raises, together with proposals for testing them. Chapter 4 describes the preliminary research used to test these underlying assertions and questions. The short experiments and studies use a variety of techniques from different fields as appropriate to the assertion or question under consideration.
Potential research methods are considered in chapter 5 and the particular study method developed and described. This is the interview based technique referred to above. A series of case studies are then described in chapter 6 in which the technique is used with two different user groups. The results of the interviews demonstrate the potential usefulness of the approach. Finally, chapter 7 draws together results and conclusions from the previous tests and experiments. Suggestions are also made for further research in testing the model and applying it to other areas of human-computer interaction.