2 Computing and metaphor

2.1 Metaphor

2.1.1 Introduction

The word 'metaphor, comes from the Greek metaphora (metaphora), which literally means to 'transfer' or 'convey'. This literal meaning remains in modern Greek but even in ancient Greece, 'metaphor' had a second meaning: "Metaphor is the transport to one thing of a name which designates another" (Aristotle, Poetics). It is this meaning, itself originally metaphorical, which has been adopted by English and other European languages.

Metaphor is an example of a trope, the rather inadequate dictionary definition of which is a 'figure of speech'. The next chapter will look at tropes more closely. However, most linguists agree that tropes are more than parts of speech, following Richards' definition:

Richards also introduced what is now a standard terminology for the components of a metaphor:

• Tenor: the original concept
• Vehicle: the second concept 'transported' to modify or transform the tenor
• Ground: the set of features common to the tenor and the vehicle
• Tension: the effort demanded to span the gap between the tenor and the vehicle>

Confirmation that metaphor deals with thoughts rather than simply words comes from its role in the development of new concepts in science. (Leatherdale 1974). Eileen Cornell Way gives some examples of the importance of metaphor to science:

The power of metaphor in science is not always beneficial. For example, Huygens view of light as continuous waves, which he likened to those caused by a stone dropping into water (Eisenberg 1992, p.144), led to confusion and argument when other experiments appeared to show light acting in a particulate manner (Feynman 1990, p.15). It appears that the metaphors that help early understanding can sometimes hinder the further development of that understanding.

2.1.2 Metaphor in computing

Although metaphor is a thought-process that can be expressed through language, is it equally valid to express metaphor through computation? Certainly the correspondence between computation and language at a low level is very strong:

Following this definition, Brady also provides a mathematical proof (Brady 1977, p.88-90). Alternative proofs may be found in Cooke & Bez (1984, p.265-73) or many other Computer Science textbooks. These demonstrate that each of Chomsky's grammars generate a language that is computable by a specific class of abstract machine, from Finite State Machines upwards. The highest level of language, generated by a Chomsky type-0 grammar, is computable by a Turing Machine (TM) and also approximately corresponds to the class of natural human languages.

Both Chomsky's grammars and TMs are idealised forms. In particular, a TM demands unbounded, though not infinite, storage capacity and unlimited time. In other words, it is not always possible to predict in advance how much storage a program will require, or how long it will run, except by running it on a TM. By contrast, computers have limited storage capacity and typically include control programs that will halt a program that recurses more than a certain number of times. These limitations can be seen as directly comparable to the limits of human memory and our inability to handle more than a limited amount of clause-nesting in a sentence. In practice, computer languages are comparable to natural languages, as we normally use them.

Computer programming is also, generally, a verbal activity in that it is based on words with a linear syntax, its semantics depending on the ordering of these words. In this it is close to most human languages, although a natural language does not have to be verbal or have a linear syntax. For example, natural sign languages have a spatial syntax quite different from the linear syntax of verbal language (Sachs 1989, p.76).

Thus, there is a mathematical concordance between computers and language at a basic level. The comparable syntactic structures imply that similar semantic structures could exist on top of these. It does not confirm that they do.

One reason to suggest that computing will involve metaphor is that the basic principle of programmable machines is very close to that underlying metaphor. Even before the existence of computers, Turing showed that a Universal Turing Machine (UTM) could be programmed to behave exactly as any other Turing Machine. In other words, one abstract machine may be 'transported' to another in a similar manner to the principle of metaphor 'transporting' a concept from one context to another. Alan Kay goes further:

A simple practical example of this comes when one type of computer is used to emulate a different computer, forming a virtual machine. Similarly, software can be used to create virtual input and output devices or virtual discs. Conversely, this power means that computing and information systems can themselves become powerful metaphor vehicles in areas such as management (Jackson 1995).

Machine code and assembly language keep to the step-by-step instructions of the Turing Machine but higher level languages can involve the introduction of structures, such as 'objects'. These are unarguably metaphors, introduced to assist the programmer, and have no existence in the low level machine code which is generated. Programming is a very specialised form of human-computer interaction and most computer users do not use full programming languages, but metaphor is also prominent in the interfaces used by ordinary computer users, particularly graphical user interfaces (GUIs).

It might be argued that, in contrast to computer languages, a GUI is not a full language. At the extreme, consider the very limiting definition of language given by Weinrich:

According to this definition, non-programmable GUIs are definitely not true languages nor, even, are computer languages. However, they do classify as what Weinrich refers to above as 'semiotic mechanisms' or 'systems [of] sign vehicles'. Semiotics may be seen as a superset of linguistics, dealing with all semiotic systems: both language and other 'systems of sign vehicles'. The next chapter will look further at semiotics; for now it is sufficient to point out that semioticians regard any semiotic system as capable of carrying a metaphor.

2.1.3 Metaphor and other tropes

Metaphor is an example of a trope: a non-literal method of description. Appendix A lists the common tropes, together with other potentially relevant rhetorical devices. Other tropes work in a similar manner to metaphor, most notably simile and analogy. The difference between simile and metaphor is whether the drawing together of the two concepts is implicit or explicit, respectively:

Metaphor: "The Macintosh interface is a desktop"

Simile: "The Macintosh interface is like a desktop"

Analogy differs from these in drawing parallels between extended processes or narratives. For example, the political endeavours of John Iselin's family in Richard Condon's book, 'The Manchurian Candidate' (Condon 1973) are often described as analogous to those of the Kennedys and were certainly intended to be so by the author. However, many people also saw a post hoc analogy in the death of John Iselin in the book, though it was written before John or Robert Kennedy was killed. Analogy and metaphor, unlike simile, do not have to be intended by their creators.

It might be argued that, for example, the Macintosh's desktop metaphor's explicit nature makes it a simile. Others might argue that its extension to so many sub-metaphors, such as folders and documents, means it is more truly an analogy. However, the underlying thought processes, that of transporting a concept from one context to another, is the same in all these cases. Analysts such as Lodge and Jakobson have identified metaphor as one of two master tropes, with analogy and simile as sub-classes of metaphor (Jakobson 1956; Lodge 1990). The second master trope is metonymy, with synecdoche as a sub-class, in which a part or an attribute stands for the whole, or the whole stands for a part. For example, we might say, "Netscape announced a new attack on Microsoft today", rather than "A spokesperson from Netscape...." or "Sculley produced too many low-end machines", rather than "Apple factories, when the company was headed by Sculley, produced...". Lodge sees the two master tropes as central to discourse:

Eco (1985, p.251) agrees with the primacy of "metaphoric mechanisms and metonymic mechanisms; to these one can probably ascribe the entire range of tropes, figures of speech, and figures of thought." There is also some argument about the relationship between metonymy and metaphor. Jakobson (1956) sees metonymy as a different principle of organisation to that of metaphor, as do Wellek and Warren (Wellek 1976). Lodge (1977, p.79-80) repeats Jakobson's argument, using Jakobson's evidence of different forms of aphasia (Jakobson 1956, p.58). In one form, patients have problems in talking of anything not present, generating apparent metaphors; in the second form, patients make 'metonymic' mistakes such as substituting 'knife' for 'fork' or 'smoke' for 'pipe'. However, neither Jakobson nor Lodge presents any evidence that these patients are using the same processes of metaphor and metonymy as those used in normal speech or literature. Indeed, Lodge has some difficulty with the extensive use of metonymy in literature, which he sees as essentially metaphoric. He explains this by saying that even though metonymy is used, "The literary text is always metaphoric in the sense that when we interpret it...we make it into a total metaphor: the text is the vehicle, the world is its tenor." (Lodge 1977, p.109). This does not appear to fully resolve the contradiction.

Others see one trope as a form of the other: Whittock (1990) sees metonymy as a type of self-referential metaphor, whereas Eco (1985) argues that metaphor depends on metonymy in that it abstracts a feature or features which the two domains have in common, i.e. the ground. Whichever of these arguments is accepted, it appears likely that the role of metaphor in user interfaces cannot be examined without some attention to the parallel role of metonymy.

The primacy of the two master tropes is widely recognised as being not only central to discourse but also to the development of language, particularly in the case of metaphor:

Some see the two processes as central to more than just language. As Lodge explains, "Metaphor and metonymy are in fact manipulations of two processes that are basic to language, and perhaps to all perception and representation - selection and combination." (Lodge 1990, p.150). Lakoff and Johnson take, if anything, a stronger view:

There is also a third, minor trope separate from metaphor and metonymy - irony. Irony is rare in human-computer interface design, although the titles of 'yacc' (yet another compiler compiler) or 'Yahoo!' (Yet another hierarchically organised object) could be seen as examples, as could the use of the 'Jack-in-a-Box' icon for the far from playful ResEdit program on the Macintosh.

Language evolves and develops new terms through metaphor and metonymy, as the references from Lodge and Eco make clear. This is also why these two tropes are so important to computing. Less than fifty years ago there were no computer languages; today there are probably many hundreds, if not thousands, of computer languages and dialects, in addition to many different user interfaces. Both programming languages and user interfaces require descriptive systems for the new concepts computers introduced, such as programs, files and disc drives. As with other semiotic systems, it is to be expected that the two main routes to describing these new concepts will be metaphor and metonymy.

Although this thesis is more concerned with metaphor, metonymy is also used in the development of computing terms. Unfortunately, the features which initially distinguish a concept or object may not be the most useful in the long term. For example, when floppy disc drives were first used in personal computers, the term 'disc' was a useful distinction from the cassette tape then in use, echoing the use of the same terms for music cassettes and 12" discs; the term 'floppy' helped distinguish the new type of disc from the removable 'hard' discs then used on mainframe computers. Now, 31/2" discs have both their flexibility and disc-shape hidden in a hard, square case.

2.1.4 Dead metaphors

The processes of metaphor and metonymy are usually seen as the first stage in the formation of a new term. After a time, the origins of a word or expression tend to be forgotten and it becomes an accepted part of the language, a process known as assimilation. In the case of a metaphor, we describe the metaphor as 'dead'. The death of a metaphor depends on many factors and is personal; computer metaphors that have died for those working in the field will be seen afresh as metaphors by new users.

Simply because a metaphor has been in existence for a long time does not necessarily mean that it is dead. Lakoff and Johnson argue that certain types of metaphor are fundamental to all language, built up from our common experiences when growing up. One such concept they identify is that of the 'orientational metaphors' (Lakoff 1980, p.14-24). For example, the use of the word 'right' in contrast to 'wrong' probably has metaphoric roots in that right-handed people are seen as doing things the 'right' way. In contrast, left-handed people do things in a manner that is unnatural or 'sinister' - the Latin word for 'left'. This metaphor is probably now dead for most people. For example, a user confronted with the following dialogue would be unlikely to experience confusion or discomfort, apart from its departure from the Apple Human Interface Guidelines:

We usually talk of 'right and wrong' and the dialogue follows our normal verbal ordering, placing the more important term first: it does not matter that the word 'Right' is on the left. However, the following dialogue would also be acceptable:

To treat a dead metaphor as being alive is not likely to damage the user's understanding, but the dangers of ignoring metaphors are much greater, and metaphors can persist for a long time without completely dying. For example, we talk of 'high' temperatures or 'depths' of cold. Consider the fictional graph below, in which hotter temperatures are shown as physically lower:

It is not just the association with the numbering on the temperature scales that dictates this relationship. The graph above uses degrees of frost as the scale, in which a larger number refers to greater cold. People to whom I have informally shown this graph find it difficult to see it in this way and persist in seeing 'hotter' as being 'higher'. Though people may no longer be conscious of the metaphor, it has not completely died.

As a simple guide, a metaphor can be considered dead once the metaphoric meaning is listed in the dictionary as a meaning in its own right. Goatly (1997, p31-35) discusses why some metaphors die but not others, and shows that the situation is more complex, giving examples of metaphors which he sees as 'dead', 'dead and buried', 'sleeping' and 'tired'. "Dictionaries are certainly the cemeteries and the mortuaries, definitely the dormitories, and generally the resting place for the populations of metaphors." (Goatly 1997, 31).

2.2 Other ways of naming new concepts

Languages also evolve in other ways that need to be distinguished from the two main tropes. Although this thesis is concerned with metaphor and, to a lesser extent, metonymy, it is important to be aware of these mechanisms to avoid confusing them with metaphor or metonymy.

2.2.1 Word importation

A common way of naming new concepts is through word importation. A language imports words from other languages as the associated culture imports new objects or concepts, such as 'sauna' or 'Zeitgeist'. At first glance, it might be thought that the adoption of English words in a computer language or interface is an example of importation. Take, for example, the use of the term 'bookmark' to apply to a marker for a frequently accessed page in a Web browser.

It might appear that the term has been imported from English to the interface language. However, the term 'bookmark' was already familiar to the designers and users of the application and was applied to a new, unfamiliar concept, i.e. the process of metaphor. Word importation takes place when new, unfamiliar words are introduced together with new, unfamiliar concepts. Thus, for example, we have recently imported both the word and the concept of 'ombudsman' from Sweden. This was a new concept described by a new word, not a new concept named with a familiar word as was the case with 'bookmark'. In German, the term 'bookmark' has been imported and is seen as a new word (in German) for a new, computer interface concept.

2.2.2 Neologisms

A few methods of creating entirely new words, or neologisms, can be used. Acronyms can be formed, RAM (Random Access Memory) and ROM (Read Only Memory) being examples. Alternatively, new words can be generated from historical languages, such as Norbert Wiener's coining of 'cybernetics' - "We have decided to call the entire field of control and communication theory... by the name Cybernetics which we form from the Greek cubernhthz or steersman." (Wiener 1961, p.11). In this, as in many examples, metaphor is involved, but most people would not have sufficient knowledge of Greek to see this, seeing it simply as a new word. Other new terms can be taken from peoples' names, such as the Bernoulli drive or the Ada language.

2.2.3 Onomatopoeia

Onomatopoeia, in which a word imitates the sound made by the object, is rarely appropriate to computing. Some computer concepts are described by onomatopoeic words, such as 'beep' and 'click', but these words were already current in the language when they were adopted as computing terms.

The principle of onomatopoeia can be extended to words that sound like other words. For example, many words beginning 'sl' have slippery or icy meanings (slip, slime, slide, slant, slope, slalom, sled, sleigh, sledge, sludge, sleet, slop) or are associated with sleeping (sleep, slumber), both obviously onomatopoeic. This has been extended to metaphorically 'slur' people or places (sleazy, sloppy, slippery, slovenly, slouching, slob, slattern, slut, slag, slug, slum). Thus, when inventing a new term, an 'sl' word will automatically have such connotations, such as Lewis Carroll's description of the 'slithy toves' in Jabberwocky (Carroll 1982, p.134). There are a few examples where this process may have taken place in the choice of computing metaphors. As explained below, the term 'bug', originally came from military slang. However, the choice of this particular term, rather than 'insect', 'gremlin' or 'spanner in the works', may well owe something to its similarity to other, less polite 'b-words' which could come to mind when coming across such a fault.

2.2.4 Back-formation

Languages also develop through back-formation, where a new term is formed from another, even though an existing form exists. For example, computer users commonly use the term 'to input' data, a back-formation from the noun 'input', rather than the original verbal form 'to put in'. The two forms have now diverged and 'to input' is almost always restricted to its meaning in computing.

2.2.5 Slang and jargon

Natural languages include many words and expressions which form communal codes, specific to social groups of one sort or another. Terms may be associated with a geographical area (dialect), a profession (register, jargon), or a community such as prisoners (cant, argot), while other words are in general use but not accepted in 'polite society', such as swearing (Andersson 1992, p.67-90). Some terms remain outside mainstream usage for centuries, while others rapidly move from one social group to another, and even into the mainstream 'polite' language. Indeed, the term 'slang' was itself once slang (Andersson 1992, p77).

Terms enter the computing and communications community from other social groups. Some have argued that computing is dominated by military slang, such as 'bug' and 'fragging' (Levidov 1989). Others terms come from science fiction and comedy: 'tekkie' (a technical person), formed by analogy to 'Trekkie' - an insulting term for a 'Trekker' or Star Trek fan; 'spamming' (flooding newsgroups with irrelevant, often repeated information), attributed by Crystal (1998, p.108) to the Monty Python 'Spam' sketch.

2.3 Interface metaphors

2.3.1 The desktop metaphor

Most conscious use of interface metaphors has been applied to graphical user interfaces (GUIs). Interface metaphors were undoubtedly used in non-graphical interfaces, as the discussion of the work of Carroll et al below explains, but there is little evidence that this stemmed from a conscious decision by the interface designers.

The best documented examples of the explicit use of metaphor for GUIs are the Xerox Star, the Apple Lisa and the Apple Macintosh. What is now usually known as the desktop metaphor began with the Xerox Alto and was refined on the Xerox Star. The designers chose what they referred to as the "physical-office metaphor" because the Star was intended as an office information system, so reflecting the familiar world of the potential users (Smith 1982, p.246). The designers saw their metaphor as providing a 'physical' environment rather than a language of interaction:

The concept was further developed in the design of the Apple Lisa and Apple Macintosh, where it is usually known as the 'desktop metaphor' rather than the 'physical office metaphor'. Apple also provided design guidelines to ensure the consistency of the interface across third party applications (Apple 1987). The desktop metaphor has proved very successful, being adapted to front-end DOS with MS-Windows and UNIX with Open Look and Motif, but there is evidence that it is reaching the limits of its usefulness as new applications for computing arrive. Even when the Star was designed the designers were aware that "It is probably not possible to represent everything in terms of a single model" (Smith 1982, p.247) and were forced to look for a different metaphor for the records processing facility.

The most interesting of the incarnations of the desktop metaphor in the context of this thesis was its implementation on the Apple Lisa. Apart from the use of formal grammars in the analysis of programming languages, this was probably the first time that a semiotic approach to the design of the human-computer interface was used in the construction of a graphical interface. Nadin (1988) shows the way in which the Lisa interface was given a clear semantics as well as a regular post-fix syntax. For example, one semantic convention was that icons should represent nouns and menu-items represent verbs.

Nadin's work on the Lisa interface seems to have been largely neglected since. An examination of the Lisa's successor, the Macintosh, shows a confusion of prefix ('Duplicate') and postfix ('Find') commands. The Macintosh has also dropped the Lisa's convention that menu-items should all be verbs, mixing verbs ('Find') with nouns ('Information'), adverbial phrases ('By Size') and even a menu of colours.

Nadin referred to his approach as 'semiotic' rather than ''linguistic. Linguistics deals only with languages, particularly those based on words, whereas semiotics looks at all the ways in which any symbol or sign can carry meaning to a person. As such, it appears to have the potential to help in the design of both text-based and graphical user interfaces. I will explore this potential further in the next chapter.

2.3.2 New interface metaphors

The suitability of the desktop metaphor is now being challenged by new areas of computing. These include CSCW (Computer Supported Cooperative Work), hypertext systems, multimedia, the Internet, VR (Virtual Reality) and portable computing. Physical desks are generally used for office tasks by individuals and it is difficult to extend this metaphor to cover group working or other types of task, such as editing of video or audio.

Various metaphors have been proposed for CSCW, the most common probably being the room. A number of researchers have independently explored the use of the room metaphor. Xerox PARC (Henderson 1986) developed a room concept to be used by one user at a time on a single machine, while the concept was extended to multi-user groupware by Bellcore (Root 1988) and in my own work (Condon 1990). We also explored the combination of the room metaphor for informal, real-time work with a form-based metaphor for formal, non-real-time work (Hämmäinen 1991). Other have expanded beyond the immediate room to include balconies, doors and corridors (Pemberton 1993).

As its name suggests, hypertext was initially developed from a book or document metaphor, with links taking the user from one page to another. Apple recognised the limitations of the desktop metaphor when dealing with hypertext and used a 'card index' metaphor for HyperCard. This conflicted with their existing interface guidelines based on the desktop (Apple 1987), to which they responded by providing a new set of guidelines for HyperCard (Apple 1989). The hypertext principle has now expanded to hypermedia and hyperspace. Some have expanded the book metaphor to cope with these more expansive demands (Rauch 1997) or extended it to libraries (Pejtersen 1988), while others have looked towards various extended physical spaces or communities, listed below in the context of the Internet.

Multimedia presents obvious problems, in that media such as sound and video are not usually handled in an office context. Apple effectively abandoned the desktop in finding suitable ways of presenting the QuickTime multimedia facilities, adapting the standard control panel of a video recorder (Apple 1991). The wisdom of this is, perhaps, arguable, in that the poor usability of video recorders has also been condemned (Thimbleby 1991). Various multimedia preparation programs have used other metaphors drawn from the existing media industry, such as films (the storyboard in Macromind Director), newspapers/magazines (page layout programs), studio equipment (mixers for sound manipulation programs).

The combination of hypertext and multimedia on the Internet has led to a series of communications or link-based metaphors, such as the World Wide Web, the Information Superhighway, or simply, the Net. Many of the suggested interfaces for future systems are based on VR and a number of metaphors have been suggested for managing these virtual spaces. Many are based on extended spaces and landscapes or on various types of community. These include fields, villages, rivers and highways (Florin 1990), farms, including information fields and swamps (Bernstein 1993), information forests (Rifas 1994), or urban metaphors such as the city (Dieberger 1994b; GeoCities 1998). Others have even suggested non-human communities such as the ant colony (Bilchev 1993).

A separate, very popular category of metaphor is that of the interface agent or guide to show the user around (Oren 1990; Laurel 1990; Isbister 1995; Rich 1996; Lieberman 1997). This concept has now been widely adopted commercially, particularly for help facilities, in applications such as WordPerfect 6.1 (the Coach) and Office '97 (the Office Assistant).

It is finally worth noting that the size of the interface also affects which metaphors might be suitable. There is a tendency to choose metaphors for smaller devices which correspond to their size. Portable computers are often based on a book metaphor, such as Alan Kay's Dynabook (Kay 1990) or the Apple PowerBook. Palm-held computers have adopted metaphors such as the pen and notepad metaphors chosen for GO's PenPoint operating system (Carr 1991) or Apple's Newton MessagePad.

2.4 Mixing metaphors

2.4.1 Mixed metaphors in language

Metaphor can exist at many levels in language. I have already used the example of the central family in 'The Manchurian Candidate' as being analogous to the Kennedys but the same book also includes descriptive metaphors entirely unrelated to this. Consider the sentence, "He felt like dropping the phone, the call, and the whole soggy, masochistic, suicidal thing in the wastebasket." (Condon 1973, p.10). The overall sentence is a metaphor - dropping the whole affair into the wastebasket - for abandoning a particular course of action. Within this metaphor, the action itself is described by a number of metaphors - 'masochistic', 'suicidal' and 'soggy' - to put over the character's feelings about the position he is in.

Most of these metaphors are not related to one another, nor are they related to the higher level political metaphor embedded in the full narrative. Although it may seem like a conflict to describe something as masochistic and suicidal on the one hand and soggy on the other, most of us will have no problem with a sentence like this, when reading it in context, as the metaphors relate to different levels of the narrative. 'Masochistic' and 'suicidal' refer to the possible outcomes of the action, whereas 'soggy' describes the character's ambivalence about acting. Similarly, unrelated metaphors associated with different aspects of computing can be successfully presented to the user simultaneously, provided they apply to different levels of the presentation. This is quite different from combining aspects of the same overall metaphor, such as 'documents' and 'folders' within a single level.

2.4.2 Mixed metaphors in computing

Foley et al. talk of the different levels at which a user interface can be viewed in linguistic terms - semantic, syntactic and lexical levels, which they equate to functional design, sequencing design and binding design, respectively (Foley et al. 1990, p.394-95). They place metaphor within a higher level than these three, which they call 'conceptual design'. However, the authors go on to contradict themselves, describing metaphor-based design concepts, such as 'direct manipulation' or 'windows' (Foley et al. 1990, p.397,439) which they place at quite different levels of design.

Hutchins (1989) proposes three categories of metaphor in the process of human-computer interaction:

Although Hutchins gives no empirical underpinning to these categories, they provided an initial framework for the work here. However, initial examination of his system and attempts to fit known metaphors into it shows some inherent problems. Where activity and task-domain metaphors apply to computing this is in the sense of where the metaphor is taken from, rather than what it is applied to. For example, he includes activity metaphors such as 'playing a game' or 'writing a paper', and task-domain metaphors that provide a 'structure for understanding a particular task'. By contrast, Hutchins divides mode-of-interaction metaphors into four modes, each a style of human-computer interaction: conversation, declaration, model world, and collaborative manipulation. It appears that his activity and task-domain metaphors are classified by what a metaphor addresses in the user domain (the vehicle), whereas mode-of-interaction metaphors are classified by how they are applied in the computer domain (the tenor).

In examining examples of metaphors in literature or in use and attempting to find suitable categories, I have produced two separate classification systems, one based on the metaphor's tenor, the other on its vehicle. Like Hutchins', they are experiential rather than theoretical in nature, pragmatic rather than empirical, but they help to show the potential for the use of metaphor in computing.

2.4.3 Categorising computing metaphors by tenor

These categories are based on the tenor, i.e. the aspect of computing a metaphor supports, rather than the origin of the vehicle. The list is not exhaustive but provides a way to see how metaphors currently in use can be mixed.

Concept: Computer as theatre, interface as facade.

A conceptual metaphor provides a way of looking at the entire design process. For example, Laurel (1986; 1993) has advanced the idea of treating the computer as theatre. Her concept does not imply that the system should look like a theatre but suggests ways of structuring the interaction to 'maintain mimesis', keeping the user's interest and attention. Hooper (1986, p.13-14) prefers an architectural metaphor, with the screen acting as a facade that should invite the user to enter.

Design: Using metaphor as a 'tool for thought'.

Design metaphors are used as a 'tool for thought' (Smyth 1995a) and are not necessarily embodied in any part of the final implementation. This type of metaphor often comes from 'brainstorming', with designers and users generating as many metaphors as possible to help provide insights into the design process. For example, the 'Quick-cash' option to obtain £50 from a single action at an ATM (Automatic Teller Machine) came from a brainstorming session which raised consideration of the 'Less than 10 items' tills at supermarkets (personal report from a member of the design team at RACE Concertation Meeting 1993).

Development: Work-flow, system life-cycle, object-oriented design.

These are metaphors employed as part of the design methodology to assist developers in the development process which, again, the user will not be directly aware of.

Hardware: Notebook, notepad, pen, organiser.

The physical packaging of the computer can embody metaphors which might influence the presentation metaphor (see below) but do not necessarily do so. For example, many 'notebook' computers still use 'desktop' interfaces.

System: Directories, menus.

System metaphors describe the internal software structures introduced to assist the user and to structure basic interaction with the system. They are not dependent on the hardware and are often independent of the metaphors for presentation and interaction.

Presentation: Documents, filing cabinets, rooms.

Based on the lower level of Hutchins' classification system, three styles of presentation metaphor can be identified - interactional, spatial and activity-based. These will be looked at more closely later in this thesis.

Interaction: Direct manipulation, command, conversation.

This is independent of the presentation metaphor. For example, the concept of 'moving' a file from one directory or folder to another can be the typed command, 'move', or the dragging of an icon in direct manipulation. It is also independent of the system metaphors - a menu can be directly manipulated as with pop-up or pull-down menus or it can be a menu of commands as part of a command-based system.

Support: Interface agents, speech bubbles.

Metaphors can be used to help the user to understand or use the system. Examples include the 'speech balloons' in the standard Macintosh Help facility and interface agents such as the talking paper clip in Microsoft Office or the 'Coach' in WordPerfect. Additional support metaphors may also be introduced in supporting documentation or third-party manuals, although these are not part of the computer system and will therefore not be considered further.

Different metaphors can co-exist, separately, at each of these levels. Although some metaphors could be difficult to reconcile, metaphors in each category often come from very different sources. Take, for example, an Apple PowerBook running Macintosh System 8:

Table 2.1: Metaphors in the Apple PowerBook.

Metaphor Category	Implementation
Concept	Computer as appliance (the original Mac concept)
Design	not known
Development	Objects, classes, inheritance
Hardware	Notebook
System	Windows, menus
Presentation	Desktop
Interaction	Direct manipulation
Support	Balloon help

Other writers have used the term 'interface metaphor' to refer to any metaphor involved in the user's interaction with the computer software. Although most closely corresponding to the 'presentation metaphor', this could apply to most of the above categories apart from design and development. As much of the literature uses this terminology, I will also use it where distinctions between metaphor categories are not immediately important. I will now examine types of interface metaphors and the categories of metaphor vehicle which they employ.

2.4.4 Categorising interface metaphors by vehicle

The above categories apply to the aspect of computing the metaphor supports (the tenor), but metaphors can also be categorised according to their origin (the vehicle). As with the previous categorisation, Hutchins' (1989) categories do not correspond to the metaphors found in current use. Examination of the metaphors discussed in the two sections above shows that they fall into one of five categories:

1 Spatial metaphors (room, landscape, space)
2 'Communications link' metaphors (net, web, highway)
3 Book or document metaphors
4 Guides or agents
5 Tools (video recorder, pen)

Looking a little deeper, it is apparent that the spatial metaphors depend heavily on structuring information. Both communications and book metaphors concentrate on the linking and interaction between units of information. Despite their immediate differences, it therefore makes sense to group these two together as 'interactional metaphors'. Agents and tools can also be grouped: they do not structure the information or provide links between information units, but instead provide an intermediary to allow the user to carry out specific activities. The five categories can therefore be reduced to three basic forms:

1 Spatial metaphors
2 Interactional metaphors
3 Activity-based metaphors

Aspects of each property can be reflected in a single interface metaphor but stressing one aspect of the metaphor will tend to decrease the extent to which the other properties apply. We can therefore consider any metaphor-based interface as a point somewhere within a triangular area:

Of the three, spatial metaphors have probably been given the greatest attention. Some have argued that they are inherently better than other types of metaphor:

It should be noted that this classification deals with the metaphor vehicle, not the medium in which the interface is presented. For example, adventure games and MUDs (Multi User Dungeons) often use text to create spatial environments, as in, "You are in a room. There are exits to the North, South and West. In front of you is a staircase." Text-based spatial metaphors have also been used to develop more 'serious' interfaces, such as Dieberger's Information City (Dieberger 1994a; 1994b). Some have even developed auditory environments based on spatial metaphors (Lumbreras 1993; Mynatt 1994).

Rather than the medium, therefore, the three categories of metaphor depend on how they support the user. Spatial metaphors structure people and information according to where they are, activity-based metaphors by what can be done in relation to the information or the people, and interactional metaphors by how the units of information, or people, link to one another.

2.5 Similar interface concepts

There are some concepts used in the design of computer interfaces which are superficially similar to metaphor but depend on different processes. Although separate from metaphor, many of them will feed into the development of metaphors.

2.5.1 Puns

Whereas metaphor is a matter of taking a concept from one domain into another, a pun depends purely on word play. Puns depend on the specific language they are expressed in and can be distinguished from metaphors by changing the language. For example, the program 'MicroPhone' is used to connect a micro (computer) to a (tele) phone and uses a microphone as its icon. Although the French language uses the same term microphone, this does not obviously relate to connecting an ordinateur and a téléphone. A true metaphor, such as 'move', will still work when expressed as remuer.

2.5.2 Anthropomorphism and animism

Anthropomorphism is used to describe the 'humanising' of non-humans, whether animals or objects and is reflected in product names such as 'Mr' Sheen, the Sony 'Walkman' or 'Gameboy'. An example of anthropomorphism in computing is the 'smiling Mac' that appears at start-up on the Macintosh, or the 'sad Mac' when the computer cannot start.

The term 'animism' is used in two contexts. In developmental psychology the phrase is used to describe a phenomenon in which children credit inanimate objects with self-will. Animism is also used to describe the presence of a human-like spirit inside an object, such as water nymphs or wood nymphs. In anthropomorphism, external human attributes are imposed on an object, such as people's names or faces; in animism the human attributes are internal, 'waiting to get out'. The following quotation makes this distinction clear in the context of computing:

Although Oren et al felt they saw a call from users for personification, other researchers have found serious problems with it. Quintanar et al (1982) examined responses to an anthropomorphic dialogue, comparing it with a mechanistic dialogue and found that users considered the anthropomorphic design to be 'less honest'. Other research has been ambivalent, finding a degree of chattiness to be beneficial but that users are quickly bored if it becomes excessive (Spiliotopoulos 1981).

2.5.3 Personification (or prosopopeia)

This is a trope, rather than a psychological phenomenon, and includes both the animistic and anthropomorphic principles. It can be applied to abstract concepts as well as concrete objects and is commonly used in romantic or detective fiction with expressions such as 'Fortune smiled on her that day' or 'he looked into the eyes of Death'. The scope for personification in computing is small, except in the sub-forms of animism and anthropomorphism described above.

2.5.4 User-friendliness

The concept of user-friendliness has been condemned by Shneiderman (1987, p.73) as a 'vague and misleading notion'. It is a special case of the anthropomorphic and animistic principles, applying the human quality of 'friendliness' to a machine or program. Take, for example, the following message from the Macintosh version of Eudora (an Internet mail program):

The Eudora message is chatty in form, implying an ability to converse in a human manner. As Schneiderman (1987, p.323) points out, "Attributions of intelligence, independent activity, free will, or knowledge to computers can deceive, confuse, and mislead users." This is demonstrated by the error codes '-1' and '{37:353}' in the message above which are completely impenetrable to most users. Had the program been truly 'knowledgeable' and 'intelligent', it would have realised that, in this case, the modem was switched off.

2.5.5 Literal categories

Some people will perceive a concept as a metaphor that another will simply regard as a valid sub-category of the word. Goatly (1997, p.21) points out that phrases such as 'a pike is a kind of fish', a literal categorisation, are very similar to metaphors such as 'a sock is a kind of glove' and that some phrases, such as 'an escalator is a kind of staircase', are ambiguous. An example in computing might be 'directory' which Chambers English Dictionary (1990) describes both as a book and as 'a body of directions'. Although a disc directory is literally 'a body of directions', all pre-computing directories existed on paper and many users would see this as a metaphor.

2.6 Benefits and dangers of computing metaphors

2.6.1 Metaphors and the learning process

Most writers in the past have emphasised the value of metaphor to computing, particularly as part of the learning process. The apparent advantages of interface metaphors are relatively simple to explain:

In an earlier paper, Carroll, Kellog and Mack (1985) had put this in terms of 'active learning', claiming: "Metaphors can facilitate active learning... by providing clues for abductive and adductive inferences through which learners construct procedural knowledge of the computer." Cornell Way agrees with this view: "Metaphor is important because it provides us with a way of moving from known ideas and familiar concepts to new and unknown ones ... Metaphor then is important to learning; it is easier to take parts from other established concepts than to build up new ones from scratch." (Cornell Way 1991, p.8).

In parallel work, looking at the use of the typewriter analogy for text editors, on this occasion by people learning about them, Douglas and Moran came to similar conclusions. They invoke the concept of problem space to explain the process:

2.6.2 Metaphor fit

Some authors have identified problems that metaphors in computing can cause, particularly where the fit, or tension, between the vehicle and the tenor is poor. Where the metaphor is not explicit and the basis of classification is unclear, the scope for semantic confusion in the user becomes greater. Carroll and Thomas (1982) discuss a number of examples, while another study by Carroll and Mack (1984) looks at the particular problems of people learning to use standard word processors. In this study they discovered that many problems were caused by the users' expectation that the word processor would behave like a typewriter. For example, users expected the text to move up the page when they pressed the return key as would happen when pressing the carriage return of a typewriter. In a later paper, Carroll and Mack point to the inevitability of mismatches:

Although mismatches may be inevitable, not all authors see them as necessarily wrong:

Hammond and Allinson (1987) have criticised interfaces that are too heavily dependent on metaphors and Johnson gives some examples of the problems this could cause:

Kay also attacks the too literal implementation of metaphors, calling for greater use of 'magic':

Sometimes 'magic' is explicit, as in games that use 'teleport' devices. Other examples are more mundane, such as the ability to paste an unlimited number of times with the 'cut-and-paste' metaphor. For example, Dieberger explores the many magical features of the Macintosh 'folder' metaphor (Dieberger 1994a, p.60).

Whereas magic adds features, designers also leave out features not relevant to computing activities. For example, gravity is useful for the organisation of a physical desktop but also means that things can fall off onto the floor; on the Macintosh desktop things simply stay wherever they are placed. "In an actual interface design process, the designer has to decide which features of a source domain are to be considered salient and which are not." (Kuhn 1991, p.423). Having made these decisions, it is also critically important that the designer communicates this information to the user:

Pointing out the magic features to the user obviously depends on the designer being aware that they are magic. In the example of Carroll et al's typewriter metaphor, discussed above, there had been no conscious intention to use a metaphor on the part of the designers. The problem was that the metaphor was implicit and thus unstructured. Had the designers explicitly used the typewriter metaphor, the system and the training that the users received could have been designed in such a way as to make this clear and, most importantly, to make clear where the interface diverged from that of a typewriter.

2.6.3 Implicit and explicit metaphors

This example suggests that it is impossible for interface designers to avoid metaphor. If designers attempt to build metaphor-free systems, either they will introduce metaphors subconsciously or the users themselves will introduce them. As designers can have no control over metaphors they are unaware of, the effects of the metaphor on the users will be unpredictable, subsequent revisions of the software may not take the metaphors into account and, at worst, they will lead to confusion on the part of the users. Under these circumstances, it is probably better for the designer to explicitly choose appropriate metaphors and keep control of them in the development of the software.

Carroll and Mack's 'typewriter' example was implicit for both the designer and the user. Problems can also occur when the designer's explicit metaphor is not recognised by the user. Anyone who has helped naive users has probably come across examples of this type, particularly where an icon has been mistaken for a picture of something it is not intended to be.

Problems of this type may be seen as an example of cultural mis-match between worlds of the designer and the user, such that the designer is not conscious of the metaphor or, if conscious, does not realise its potential impact on the user. Some extreme examples of cultural mis-matches are given by Grundy (1996, p.85-94). Her main argument is that, "Computing is taught using metaphors, analogies and examples drawn largely from a male environment. Women students have therefore always been required to understand what they are taught through a screen of male values and experience." (Grundy 1996, p.88).

She reserves her heaviest attack for 'rape metaphors' in computing, quoting Francis Bacon's description of scientific investigation, in which he refers to nature as 'her': "Neither ought a man to make scruple entering and penetrating into those holes and corners when the inquisition of truth is his whole object". She argues that "Bacon's rape metaphors helped to shape the methodology of science powerfully in the past... Although the notion of rape itself may have disappeared, the language and imagery of male-dominated sex and violence still persists in computing jargon and must influence girls and women, particularly as they start to learn the subject". She then gives examples of such terms including 'violation', 'degradation', 'chaining', 'abort', 'kill' and 'execute' (Grundy 1996, p.90-92).

The importance of Grundy's argument does not depend on any underlying validity. The fact that she is sincere in her views and that other women may share her viewpoint means that they will feel alienated by metaphors which they see as degrading towards women. The casual use of the term 'abort' in computing is an example that many people might appreciate. Although 'abort' has the wider meaning of abandoning an action, in popular use it almost always refers to the abortion of a pregnancy. A comparable 'male' term might be to replace the error message 'the command could not be completed due to insufficient memory' with 'your command was impotent due to insufficient memory'. This would be technically correct but many men might feel uncomfortable with it.

2.6.4 Opposition to interface metaphors

One of the strongest attacks on interface metaphors comes from Nelson (1990, p.236) who identifies three 'elements of bad design', one of which is what he terms 'metaphorics'. "I would like to venture that this 'metaphor' business has gone too far. Slogans and catchphrases are all very well, and these things have their uses for people who are going to learn software approximating rather than understanding.". He also claims that, "the metaphor becomes a dead weight," and suggests that the "alternative to metaphorics is the construction of well-thought-out unifying ideas, embodied in richer graphic expressions that are not chained to silly comparisons." (Nelson 1990, p.237). His alternative is 'the design of principles', giving VisiCalc as an example and suggesting the use of hypertext as a future alternative (Nelson 1990, p.240-242).

Although metaphor might be a 'dead weight', his examples of better practice do not appear to provide a metaphor-free alternative. VisiCalc may have been a new concept for most of its users but is basically a metaphor, based on spreadsheets of a type already in manual use by accountants. Hypertext is undoubtedly a useful tool, but the references given in Section 2.3.2 show that many developers find it is only part of the answer and are searching for suitable metaphors to help prevent users becoming 'lost in hyperspace'.

Alan Kay (1990) also offers a critique of metaphor, "My main complaint is that metaphor is a poor metaphor for what needs to be done. At PARC we coined the phrase user illusion for what needs to be done." He continues, "it is the magic... that really counts." (Kay 1990, p.199). The importance of magic has been acknowledged above, but it depends on an underlying metaphor to be seen as magic. He offers examples of 'better' solutions taken from interfaces such as Smalltalk and TEX (Kay 1990, p.200; p.203). Although both systems have their adherents, neither has become as universally accepted as the metaphor-based interfaces Kay rejects. Popular acceptance should not, of course, be taken as an automatic guarantee of usability, but requires an explanation if his argument is to be accepted. As a future direction, Kay (1990, p.205-07) offers agents as a more promising possibility, although it can be argued that these are only a particular type of metaphor.

Both of these critiques come from a series of what the editor, Laurel (1990 p.187), introduces as 'sermons' rather than academic studies and offer the authors' personal views rather than experimental results. In general, academic work, such as that by Carroll cited above, has shown that metaphors are unavoidable, being introduced subconsciously by the designer or consciously by the user. However, criticism of the use of metaphor or on the specific choice of interface metaphor, such as those by Grundy and Nelson, suggest that metaphors can raise significant emotions and signify far more than a simple aid to the learning process. I will examine this signification further in the next chapter, looking at whether semiotics, the study of signs, might assist the interface designer to a more complete understanding of human-computer interaction, in particular the role of interface metaphors.

Back to Academic Work