Human computer interaction, or HCI, is concerned with the way in which people communicate with computers, and the goals of HCI are to produce usable, functional, reliable, and safe systems. When you choose between two alternative brands of software (e.g., WordPerfect and Word for Windows), you often make your choice on the basis of how easy it is to use the software, rather than its power or efficiency. The way in which we conduct a dialog with a computer is every bit as important as the hardware itself.

Much of the glamour in computing has been grabbed by advances in computer hardware and software. Hardware has been getting faster, cheaper, and more powerful year by year. Scarcely a month goes by without the release of a major upgrade to an operating system, word processor, spread sheet, database, or compiler. HCI is the Cinderella of computing and often has had to wait at home while the hardware and software go to the ball. Today, people are paying more attention to human computer interaction because it determines how efficiently and how reliably we can use a computer. In this chapter we are going to introduce some of the fundamental concepts of HCI and describe the ingredients of an effective interface.

Few of the elements of computer science involve a single discipline; for example, the operating system is the point at which hardware and software meet. HCI is probably the most multidisciplinary element of computer science and is the point at which hardware, software engineering and human psychology all meet. It’s impossible to design an effective interface between a human being and a computer without taking account of the human’s characteristics. For example, there’s little point in displaying text on a screen faster than you can read it. If an image is animated, the motion must be swift enough to avoid boredom and slow enough to avoid jerkiness. Similarly, if the message from the computer is too complex and detailed, the human operator will neither be able to understand it nor remember it.

Why should we be so concerned about human computer interaction? Computers, or any other machines, do exactly what we tell them. It is therefore vital that the communication between the human being and the machine be both efficient and unambiguous. I can think of few interfaces worse than that between the human being and a typical video recorder, VCR. When I go on vacation it takes me ages to program my VCR—I have to step through the channels, days, hours and minutes (for both program on times and program off times) for each of the eight programs I wish to record. Why can’t I enter the name of the program I want to watch from a keyboard? Hasn’t the VCR learned by now what programs I like?

Consider another example of a poorly designed human-machine interface. On 5 July 1970 a DC-8 aircraft was approaching Toronto International Airport with over 100 passengers and crew on board. At 60 feet above the runway the co-pilot moved the spoiler control lever in readiness for the landing. Spoilers are large flat metal plates hinged to the upper surface of an aircraft’s wings. When spoilers are deployed immediately after landing, they swing up into the airstream and destroy the wings’ lift. Spoilers ensure that the aircraft‘s transition from flying to taxiing is clean. The spoiler control in the DC-8 has two active positions—if you lift the lever the spoilers are armed and deployed automatically after the aircraft touches down; if you pull the lever the spoilers are deployed immediately.

At 60 feet above the runway the spoiler control lever in the DC-8 was pulled. The spoilers were deployed and the wings immediately lost their lift. The passengers and crew lost their lives. The initial response to the incident was to suggest that a placard be placed alongside the spoiler lever with the caption Deployment in Flight Prohibited. Perhaps they should have written Please do not crash this aircraft. It took two more tragedies before the spoiler control was modified to prevent inadvertent operation in flight.

Although this dramatic example doesn’t involve computers, it demonstrates what can happen when the designer creates an interface that doesn’t take account of how a real user might operate it under all circumstances.

Poor interface design doesn’t necessarily cause errors as such. But what price frustration? I once used a desktop publishing package to produce a book. This package could easily modify the font of all the text in a whole paragraph, but it couldn’t easily change the font of a single word in a paragraph. Because I was using a special font for any assembly language words embedded in a paragraph, I had to go through three pull-down menus to make each and every font change—and there were thousands. A later version of this software has solved this problem. Jacob Nielsen (Iterative User-Interface Design, Computer, November 1993) describes the construction of a home banking system that yielded a 242% improvement when the user interface was tested and improved over three versions.

A heartfelt comment on the user interface is made by Baecker and Buxton in their introduction to Readings in Human-Computer Interaction:

"Yet despite its importance, the user interface is one of the most poorly understood aspects of any system. Its success or failure is determined by a complex range of poorly understood and subtly interrelated issues, including whether the system is congenial or hostile, easy or difficult to learn, easy or difficult to use, responsive or sluggish, forgiving or intolerant of human error".

We are going to look at both the physical and the logical interface. The physical interface encompasses the hardware used in the dialog between the human and the computer. The logical interface describes the way in which a user employs the physical interface to engage in a dialog with the machine; for example, a programmer might use a mouse and a display (the physical interface) to select items from a menu on the screen (the logical interface). A study of the way in which computers and humans interact helps user interface designers to create systems that make best use of the available technology, enhance the user’s productivity, and reduce the number of errors they make. In short, the goal of HCI is to increase the usability of a computer and its software.

Humans communicate with each other, principally, by auditory and visual stimuli; that is, we speak, gesticulate, write to each other, and use pictures. You would therefore expect humans and computers to communicate in a similar way. Computers are fairly good at communicating with people; they can generate sophisticated images, although they are rather less good at synthesizing natural-sounding speech. Unfortunately, computers cannot yet receive visual or sound input directly from people. Hardware and software capable of reliably understanding speech or recognizing visual input does not yet exist—there are systems that can handle speech input and systems that can recognize handwriting, but the error rate is still too large for general-purpose use. Consequently, people communicate with computers in a different way than they communicate with other people. We communicate with computers largely by means of the keyboard and pointing devices like the mouse.

The keyboard is still the most commonly used means of getting data into a computer. Sometimes, the keyboard is an efficient interface—especially when you’re banging in text with two or more fingers. Sometimes the keyboard is a positively horrible interface—like when you are trying to use it to control a flight simulator. In this section we are going to look at the characteristics of the keyboard as a computer input device.

Consider the layout of the ubiquitous QWERTY keyboard. The term QWERTY isn’t an acronym but the sequence of letters on the back row of characters on a keyboard. When the first mechanical typewriters were constructed, the sequence of letters were chosen to reduce the probability of letters jamming. For example, if t and h were next to each other, typing the would sometimes cause the letters t and h to collide and jam. As you can imagine, the anti-jamming property of the QWERTY keyboard is optimum only for the English language.

Layout of the QWERTY keyboard

Now that keyboards are electronic and have no moving parts except the keys themselves, there is no longer a need for a QWERTY layout. A better layout would make it easier to type English by reducing the distance a typist’s fingers have to move on average. The Dvorak keyboard was developed in the 1920’s to make it easier to type English—it is also biased towards right-handed typists. Studies demonstrate that a typist can achieve a 10 to 15% improvement when using a Dvorak keyboard. Unfortunately, so many typists and programmers have been trained on the QWERTY keyboard, that it would be very time-consuming to retrain them to use a new layout. The Dvorak keyboard has therefore failed to topple the QWERTY standard.

Some systems designed for infrequent computer users and non-typists have a simple ABCDE keyboard in which the keys are laid out in alphabetic order—this keyboard makes it easy for users to locate keys, but prevents experienced users entering data rapidly (because they will have been trained on a QWERTY layout). As you can see, there is sometimes a trade-off between the needs of the experienced user and the inexperienced user.

A radically different form of keyboard is the chord keyboard that has only a few (typically 4 or 5) keys. You enter a letter by hitting a subgroup of keys simultaneously; it’s rather like using Morse code or Braille. The chord keyboard is very small indeed and can be used with one hand. Chord keyboards have found a niche market for people who operate in cramped spaces or for those who want a pocket-sized device that they can use to make notes as they move about.

In order to provide the total number of keys necessary for efficient computer operation a keyboard would have to be gigantic. In practice, most keys have a multiple function; that is, the meaning of a given key can be modified by pressing another key at the same time. The shift key selects lower case characters as the default mode, and upper case characters when it’s pressed at the same time as a letter. The shift key also selects between pairs of symbols that share a key (e.g., : and ;, @ and ‘, + and =, etc) and between numbers and symbols (e.g., 4 and $, 5 and %, 8 and *, etc).

Although the layout of the letters and numbers on a QWERTY keyboard is standard throughout the English-speaking world, the layout of other keys (e.g., symbols) is not—in particular, there is a difference between keyboards designed for use in the USA and those designed for use in the UK. Consequently, software has to be configured for the specific version of the keyboard currently in use.

Modern computer keyboards also include a control, Ctrl, key that behaves like a shift key. The control key gives a key a different meaning when control is pressed at the same time. Computer books indicate the act of pressing the control key and, say, the letter D at the same time by the notation CTRL-D.

Why do we need all these special keys? When we communicate with a computer we need to provide it with two types of entries. One is the information or data the computer is going to process (e.g., the text entered into a word processor, or a booking entered into an airline’s database). The other type of information entered into a computer is the commands that you want it to execute. Suppose that you are entering text into a word processor and wish to save the file. You can’t simply type Save file because the computer cannot distinguish between the command you want to carry out and the words you are entering into the document. By typing, for example, CTRL-S, you unambiguously are telling the computer that you are entering the command to save a file.

PCs go one step further and provide an alt (alternative) key to give yet another set of meanings to the keys. Consequently, you can enter a key unshifted, with shift, control, alternate, or any combination of the three function-modifier keys. Computer programs that make extensive use of function control keys to enter commands often provide the user with a plastic template that is attached to the keyboard to remind him or her the meaning of the various control character combinations.

In addition to the use of the shift, control, and alternative keys, the PC keyboard contains 12 special-purpose function keys labeled F1 to F12 that can be used to perform special functions. Moreover, these functions are modified by the shift, control, and alternate keys. Finally, keyboards have several dedicated keys like home, end, PgDn, PgUp, Del, Ins, and so on.

The use of the shift, control, alternate, and function keys, makes it much easier to communicate with a computer. All you have to do is to remember the special codes; there can’t be more than a few hundred of them...... Moreover, the way in which keys are assigned to control functions differs from one application to another. When applications writers adhere to a common control key usage (e.g., function key F1 is normally used to invoke a program’s Help function), the user finds it easy to switch between applications from different manufacturers. When an applications programmer assigns control keys in an idiosyncratic way, it becomes very difficult to switch applications because you keep hitting the wrong key.

Computer displays invariably have a cursor—a marker on the screen indicating the currently active position; that is, if you enter a character, it will appear at the position indicated by the cursor. Cursors can be vertical or horizontal lines, small blocks, highlighted text, or even reversed text (i.e., white-on-black). Modern applications frequently make use of several different types of cursor; for example, a solid line indicates where text can be entered, an arrow points at a command, and a cross indicates the edge of a picture or a table. Nearly all cursors blink because human vision can more easily detect a change in a static picture). Computer keyboards also contain four cursor control keys. These keys move the cursor on the screen up, down, left, or right, by one unit—either a character position horizontally or a line position vertically. These keys can also be used as a crude type of joystick or mouse for systems that require a cursor to be moved anywhere within a certain area.

There are many ways of designing a keyboard and several technologies can be used to detect a keystroke (e.g., mechanical, magnetic, capacitive, etc). The difference between keyboards is often a matter of cost and personal preference—some typists prefer to hear a satisfying click when they depress a key, others don’t. Important keys like enter, shift, control, and space are often made larger than other keys to make it easy to hit them. If you are a real sadist, keyboard design is just for you— you can guarantee a maximum level of user misery by locating a key that has a potentially destructive function (e.g., delete text) next to a normal key such as the space bar. Good practice would ensure that it is difficult to enter a potentially fatal command by accident. Consider the following two examples of safe operation: you cannot start a VCR recording without pressing two buttons simultaneously, and the master engine switches in some aircraft are under a bar to ensure that you cannot switch an engine off accidentally.

Although the keyboard is an excellent device for inputting text, it cannot be used efficiently as a pointing device, to select an arbitrary point on the screen. The three most popular pointing devices are the joystick, the mouse, and the trackball.

Pointing devices

The joystick is so called because it mimics the joystick used to control military aircraft and some light aircraft. This device consists of a stick that can be moved simultaneously in a left-right and front-back direction. The computer reads the position of the stick and uses it to move a cursor on the screen in sympathy. You don’t look at the joystick when moving it; you look at the cursor on the screen. Without this visual feedback between the hand and the eye, people would not be able to use this, or similar, pointing devices. Joysticks contain one or more buttons that can be used to enter commands. The joystick is well suited to computer games, particularly aircraft, because it mimics the behavior of a real control column.

Although the joystick is similar to the mouse and trackball, there is one difference. When the mouse and track ball are not being moved, there is no signal from them and the computer unambiguously interprets this as no input. However, the joystick continually transmits a position, which means that it is very difficult to centralize or neutralize its output. Consequently, joysticks often have a dead zone around their neutral position. Until you move the joystick out of the dead zone, the cursor on the screen doesn’t move.

The mouse is probably the most popular pointing device in this group. A mechanical mouse consists of a housing and a ball that rotates in contact with the surface of a desk—you could say that a mouse is a ball-point pen on a large scale. As the mouse is dragged along the desk, the ball rotates and circuitry in the mouse translates the movement of the ball into a signal that can be read by the computer. An electronic mouse is used on a special pad that has a grid of horizontal and vertical lines. The mouse reflects a light beam off the grid and counts the lines crossed as the mouse moves about the pad.

When the computer gets a signal from the mouse, it is scaled and used to move a cursor within the screen (exactly like the other pointing devices in this group). When the software needed to control a mouse is installed, the user can choose the mouse’s sensitivity; that is, how much the cursor on the screen moves for a given movement of the mouse. The sensitivity chosen is a function of the user’s hand-eye coordination.

A modern mouse is comfortable to hold and can be used to move the cursor rapidly to any point on the screen. Once the mouse is at the correct point, you depress one of two (or possibly three) buttons that fit naturally under your fingers as you move the mouse. Pressing a button activates some pre-defined application-dependent function on the screen. Typical mouse-based systems require you to click the button once to select an application (i.e., highlight it), and twice to launch an application (i.e., run it). Clicking a button twice in this way is called double-clicking and is not always easy to perform because the interval between the clicks must fall within a given range.

A trackball is an upside-down mouse—it remains stationary on the desk and you rotate a 2" to 6" ball to move the cursor on the screen. Unlike the mouse, the trackball requires no desk space and can be fitted on the keyboard of a lap-top portable. Trackballs are often built in to electronic equipment that requires an operator to select a point on a screen (e.g., a target on a radar screen). Some computer users prefer the trackball to the mouse.

The trackball or a tiny joystick is now routinely built into laptop and notebook computers. Some manufacturers go a long way to make the pointing device easy to use. Other manufacturers carefully position the pointing device to ensure that it cannot be operated efficiently by a left-handed user.

Other, less widely used, pointing devices are the touch screen, the light-pen, and the tablet. By either coating a display screen with transparent conductors or by using ultrasonic/infra-red beams, the computer can detect the location of a finger on the surface of the screen. Consequently, the screen can be used as an input device simply by touching the point you want to activate. A typical system displays a menu of commands. Touch-sensitive screens are still relatively expensive and are found only in specialized applications. Moreover, the finger is a rather course pointer and cannot be used as precisely as a mouse or joystick. However, touch screens are useful when the operator has no computer experience whatsoever (e.g., a user-controlled guide in a shopping mall).

The light-pen uses a stylus with a light-sensitive device in its tip. When placed against the computer’s screen, the light-pen sends a signal to the computer when the beam passes under it. The light-pen is just a much more precise form of the touch screen and is cheaper to implement. Sophisticated algorithms can be used to convert the light-pen’s movement over the screen (i.e., handwriting) into an ASCII-encoded text format for internal storage and manipulation. Unfortunately, it’s not easy to convert the output from a light-pen into text, because the way in which one person writes, say, a letter "a" is often different from the way another person writes it.

Some computers do have a primitive form of speech input. You speak into a microphone and the audio signal is sampled and turned into digital form. Voice recognition systems require you to train the system by first recording all the words it is to recognize. During the training process, it builds up a pattern or template for each word it is to recognize. When you later say a word, its digitized version is matched against the patterns stored during the training process and the closest fit selected. Computers cannot yet handle continuous speech reliably in real-time. Moreover, once a computer has been trained, other operators cannot use the system without retraining it, because human speech varies widely depending on the gender, age, and regional accent of the speaker.

The vast majority of general-purpose computers communicate with human beings via a screen, which may be a conventional CRT or a liquid crystal display. As we describe the screen in detail in the chapter on computer graphics, we will not describe its construction here.

The screen is viewed by human beings. Consequently, the way in which human visual perception operates is of interest to those designing the visual interface. For example, we can see some colors better than others; we cannot read text if it is too small nor can we read it rapidly if it is too large. Colors themselves are described in terms of three parameters: hue is determined by the wavelength of the light; saturation is determined by the amount of white light present in the color; and intensity is determined by the brightness of the color. Objects on a screen are viewed against background objects—the luminosity of an object in comparison with its background is called its contrast. All these factors have to be taken into account when designing an effective display.

Not only can we perceive light in terms of its hue (i.e., color), saturation (the depth of the color), and intensity (brightness), we can also perceive size and depth in a two-dimensional image. If you examine, for example, an application running under Windows, you will almost certainly see buttons and sliders that look as if they are three-dimensional. There is no need to make the buttons appear realistic, but the interface looks like the control panel of a TV or a hi-fi system. By making the computer interface resemble something we are already familiar with, we don’t have to be trained in its use.

Having looked at the physical mechanisms we use to tell the computer what we want to do, we are going to look at the logical interface that determines how the human-computer dialog is to be formatted.

Communications, whatever their nature, should be clear and unambiguous. In this section we look as some of the factor affecting the way in which we structure the dialog between the human being and the computer.

Once upon a time, when propeller-driven aircraft were more common, an aircraft was taxiing from the stand to the runway. The captain had noticed that the flight engineer was looking rather miserable, so he turned to the engineer and said, "Cheer up, Charlie." Charlie, the flight engineer, looked a little puzzled but knew that the captain’s word was law. A few seconds later, the aircraft sank onto its belly and there was a grinding sound as both propellers ripped into the pavement. "Gear up, Captain," said Charlie.

Although a computer user can’t damage an aircraft, he or she can still cause havoc. I once had a bright idea. When you use a desk-top publishing package to prepare a book, you spend a lot of time designing the lay-out of a chapter and writing macros or styles for all its elements (i.e., headings, footers etc.). When you start a new chapter, you can either copy all these design items across, or you can be really smart. You open the chapter you’ve just written, delete all the text, and then save it with a new chapter name. In this one simple operation, you’ve created the framework for a new chapter that’s exactly like the previous chapter. This process is repeated as you write each new chapter.

Creating an entire book using this technique is extremely efficient. One day, at the point you’ve opened a previous chapter and deleted all the text, the phone rings and someone asks a question. "No problem," you say, "I’ll just close this file, and fetch the file you want." Approximately one quarter of a second after you’ve hit the key that saves your file, you remember that you didn’t rename it. The empty file you’ve just saved has replaced the chapter you spent months typing. Fortunately, you did remember to take a back up yesterday—didn’t you?

Charlie the flight engineer didn’t query the captain’s orders because of the hierarchical command structure that once existed on a flight deck. Should the computer have queried my actions when I replaced a file with an empty file? Some users might regard an interface that lets you do anything as positively dangerous because it provides no safety net. On the other hand, an interface that asks you to confirm each operation becomes very tedious to use. I once used a system that provided the dialog below. Lower case text represents my input and upper case text represents the computer’s response.

delete Clements_1

ARE YOU CERTAIN YOU WANT TO DELETE CLEMENTS_1? [Yes/No]

yes

DO YOU WANT TO KEEP CLEMENTS_1? [Yes/No]

no

When I told the system to delete a file, it asked me to confirm the operation with the positive response "yes". Having confirmed that I did wish to delete the file, it asked the same question again, but expressed in the negative. In order to delete the file, the opposite response, "No", had to be given. The system demands a different response to each question to prevent the operator giving automatic responses to questions (e.g., typing "yes" the moment you see a question).

Automatically querying important operations is not foolproof. Some might argue that no safety net is better than a partial safety net. Once you feel that the computer is protecting you, you lessen your vigilance. An excessive reliance on the computer leading to a relaxation of attentiveness is thought to have been the underlying reason for more than one aircraft crash.

There’s a common expression—one man’s meat is another man’s poison. We could have said "One user’s obvious is another user’s obscure." Those who have grown up with the computer soon learn its characteristics and capabilities and they find it relatively easy to adapt to new software and interfaces. Indeed, they are often oblivious to the problems a system poses to the non-expert. Consider the mouse: you drag the mouse down the desk to pull the cursor down the screen in response. What happens when the mouse gets to the bottom of the desk? You pick it up and move it back to the top of the desk, because the cursor on the screen moves only when the mouse is in physical contact with the desk. A user who does not know instinctively how the mouse works, stops at the end of the desk and asks someone what they should do next.

The behavior of the mouse is not instinctive to all; if you turn it sideways, the cursor will move horizontally as you move the mouse vertically on the desk. Pressing the mouse button to select an icon (i.e., a small symbol that represents an object or a function such as: ???????????) on the screen in easy enough, but double-clicking to select and launch an application probably takes as much effort as changing gear in a car before the synchromesh gear box was invented (the gear change required a complex operation involving the clutch and gas pedal called double-declutching).

The meaning of the icons found on a typical Windows screen are conceptually obvious. Well, they are if you have read the trade press and computer literature for years and are familiar with the icons for Microsoft or WordPerfect, or for databases and communication packages. Even the meaning of common icons like the trash can is not always obvious to the novice computer user. We are now going to look in more detail at the characteristics of the computer user—the human.

Computers are designed and programmed by humans for humans, and, therefore, any problems associated with the computer-human interface are minimal. Run that by me again..... Computers are designed and programmed by some humans for use by other humans. That’s better. We sometimes forget that we live in a particular stratum of a particular culture, and that it strongly influences the way in which we view the world. I once saw a discussion of music that posed the question, "Is our response to a certain type of music inherent or is it culturally determined? We were given a fragment of the soundtrack from a typical Hollywood film and we all decided that the music was typical of the romance scene. Then we were played a fragment of a music from a romantic scene in an oriental film. In this case, the music was nothing like the Hollywood schmaltz and sounded discordant rather than romantic to the Western ear. Things we take for granted are, in fact, learned responses to stimuli.

The effect of culture on a computer interface is particularly important because, unlike the case of the music, the division is not into two groups such as East and West; it is much more multifaceted. Some of the differences between computer users are:

Age A young person educated in a technological society is likely to have a feel or empathy for computer systems because some of the underlying themes are familiar from the VCR timer, the pocket calculator, and the games machine. An older person might have had little prior contact with computer-like machines.

Cultural background The way we write, from left to right is part of the Western culture. Other societies write from right to left. A computer user in Europe and a computer user in Asia might scan a screen for information in a very different way. These cultural differences might influence the position of important warning and error messages.

Linguistic background Even when the computer designer and the computer user speak the same language, problems can still arise because not all words and phrases are used in the same way. In certain parts of the North of England the word "while" means "until" (e.g., "I am staying here while Friday" means "I’m staying here until Friday). Similarly, a North American might speak of tabling the plan, which would convey the opposite to someone from England (in the USA to table something means to omit it, whereas in England it means to include it). I’ve always felt that British tourists to the USA should have the following message stamped in their passports: "In the USA the device at the end of a pencil used to make typographic corrections is called an ERASER..."). Moreover, the spelling of words is sometimes different in British English and American English, and the presentation of dates is different in the USA and Europe. Some software packages permit the user to select a particular subset of English (e.g., UK, USA, AUS).

Experience Computer users can be roughly divided into four groups: the novice or non-computer expert who has little or no background knowledge of computing; the occasional or intermittent user who uses the system in bursts and forgets the details of the commands between sessions; the experienced user who uses the system frequently, and the expert user who has an in-depth understanding and will often accept a terse user interface if it will provide short cuts and aid productivity. Computer users change as their experience grows. An interface that was optimum during the training phase, might not be optimum when the user becomes more experienced.

Gender To a certain extent, boys and girls are brought up in different cultural environments and receive different messages from society. These environments may affect the way in which men and women relate to the computer. Ben Schneiderman in Designing the User Interface points out that common computer commands like KILL and ABORT may have a different effect on male and female computer users.

Disability Some computer users may suffer from visual impairments such as retinitis pigmentosa. In such cases, the actual design of the display might have a profound effect on someone suffering from impaired vision. Moreover, quite a large percentage of males are color blind. A computer user might have a limited ability to use their hands and fingers to manipulate a keyboard or a mouse. An interface that requires few keystrokes or mouse movements might suit someone with a mobility impairment more than an interface that requires large and frequent movements of the mouse.

Handedness The majority of people are right-handed. However, some physical interfaces are not well suited to left-handed users (e.g., lap-top portables with a built-in trackball on the right side of the keyboard).

Although it happens subconsciously, we build models of the world to enable us to deal with new situations; for example, someone who has learned to drive one model of car can normally drive a new model with virtually no practice. This is because the way in which all cars operate conforms to a set of basic rules. The mental database associated with a car is very complex: we can calculate a car’s dynamics and road handling performance, or handle virtually all its controls from acceleration to turning, to indicating.

When we use a computer interface, we also build a model of how the system works. This model covers both the syntactic and the semantic aspects of the interface. Syntax describes the grammar or the rules by which the interface operates; for example, the syntax of a command to copy file A to location B might be Copy A,B. In a graphical or windows environment, the syntax of the interface might be expressed in terms of the mouse movement required to access a particular menu item. You might say, that a knowledge of an interface’s syntax is similar to a knowledge of the way in which the controls of a car are operated.

Semantics is concerned with meaning, and covers, for example, an understanding of the effects of controls. In terms of an automobile, a driver with only a syntactic knowledge of the stick shift would be able to change gear but not know why a gear change was necessary. If the computer user understands an interface’s semantics but not its syntax, he or she becomes frustrated because they know what they want to do but not how to do it.

Any computer interface should be designed to have a consistent syntax—indeed, a lack of consistency is one of the most annoying aspects of almost any interface (computer or otherwise). If the syntax of the copy command is Copy A,B and has the effect of copying the contents of file A into file B, you would expect that the syntax of the rename command, Rename A,B, to be take file A and rename it as file B. If the syntax were take file B and rename it file A, the syntax would be inconsistent with the copy command. A consistent syntax has two advantages; first it reduces the user’s learning burden; second it reduces the danger of error caused by the incorrect use of a command with an inconsistent syntax. On the other hand, someone described Kai Tak airport in Hong Kong as one of the world’s safest because it is so difficult to land there; that is, if an operation is difficult to perform, you take greater care.

User syntax becomes most confusing when different commands perform similar functions at different stages in a session. Probably the most extreme case of this is the termination or completion command. If you are using a menu driven interface, the carriage return often terminates a command. If you are entering a text string, the operation might be terminated by a special key such as escape (ESC) or control D (CTRL-D). If the system accepts command words, the appropriate command might be Q (quit), or QU, or the full QUIT, or it may be KILL, or EXIT... Anyone who has used a new software package or operating system will be accustomed to entering ever more obscure mnemonics in order to try to execute a forgotten command.

Some software can be made difficult to use by providing the user with too many ways of achieving the same goal. An airline pilot who flew advanced aircraft with computerized systems told me that he could often perform an action in one of several ways. However, he said that this degree of freedom was sometimes counterproductive because he had to decide which technique he was going to use.

In this section we are going to look at some of the widgets that have been designed to facilitate human-computer interaction. A widget is a group of screen structures (e.g., menus, buttons, scroll bars etc.) provided in a graphical interface.

The nature of the interface presented by an application can vary widely. Although at the start of this chapter we said that the human-computer interface was the Cinderella of the computer world, there has been a steady improvement in the design of interfaces over the last few years. Typical approaches to interface design are template, menu, command language, and natural language.

The template is the simplest of computer interfaces because it is analogous to the type of forms you have to complete in daily life. The template or form fillin provides the user with a structure containing headings and blank spaces. The user is invited to fill in the blanks on the form. This interface can be used by almost anyone—the expert need enter only the information actually required, and the novice can often relate to the form because it is a familiar metaphor. However, some novices might experience frustration if the interface asks for Date of birth, and will accept only, say, 27/09/48, and reject 27.09.48, or 27-09-48, or 27/09/1948, and so on.

In recent years, the template approach to interface design has become more sophisticated with the advent of languages like Visual Basic that enable you to design very sophisticated interfaces, and the interfaces are, in turn, composed of ever more sophisticated widgets. The following table demonstrates the extensive template provided by Visual Basic to implement an option button in a Windows environment. As you can see, this template provides complete customization of the widget (i.e., option button).

The use of the template in Visual Basic to specify a widget’s characteristics

In a menu-driven system, the interface provides the user with a list of alternatives, one of which is selected either by a mouse or by appropriate cursor movement keys. A menu puts less pressure on the user to remember syntax; for example, if you select the copy option from a menu, the interface might then provide supplementary boxes labeled source and destination.

The following illustrates a typical menu-based application in PageMaker 5. The Type menu has been selected to display a pull-down menu. On this secondary menu the Alignment option has been selected and you are invited to choose one of the options from this submenu.

Menu based systems can be used by all levels of computer user. However, the menu is not very helpful if you cannot see it. I have worked with systems that have proven very irritating because I have forgotten which menu includes one of the commands. For example, if one of the menu items is labeled Edit, you can be sure to find the commands search and replace as options. However, if you wish to change the paper size, is the appropriate paper size command under the File menu, the Edit menu, the View menu, or the Format menu, etc?

The Pull-down Menu

Another important aspect of menus is their ease of rapid access. If a particular operation has to be activated by selecting a submenu of a submenu of a menu, the sequence of keystrokes or mouse movements can be irritatingly slow. As long as such a command is performed very infrequently and all really common operations can be accessed speedily, all is well. If you find that you have to frequently perform some of the tasks thought obscure by the interface designer, the interface can seem leaden. The best menu-based systems permit users to define their own menus in order to tailor the system to individual need. However, this tailorization brings with it its own problems, namely the user’s ability to tailor his or her own environment. A novice probably cannot create as effective an interface as an expert. Moreover, an excessive degree of tailorization can lead to screen clutter—before long the menus take up more space than the applications.

The command language is the oldest and, sometimes, most powerful computer interface. It is syntax driven and you have to know its syntax and the meaning of all its commands before you can use it. If you use MS DOS you can delete a file with the command DEL filename. Similarly, if you are using a text editor and wish to replace the text string chemistry with physics between the strings Monday and Friday in a document, you might simply need to enter the command:

C/chemistry/physics/Monday/Friday.

In the manual that describes the above change command, its syntax might be formally expressed as:

C/<source string>/<destination string>/<first occurrence/<last occurrence>.

In any reasonably complex system, a command language can be used to any degree of proficiency only by an expert or by a frequent computer user. Moreover, errors are easy to make.

The command language can include powerful operators that enable a command to be repeated or its action modified. A typical powerful command language interface is the UNIX operating system. You might devise a command language that allows the command to operate on a data stream; for example, Delete {file1, file2, file3, file4} might indicate that the delete operation might be applied to all files between the { } brackets.

Command languages can even be used to handle conditional operations; that is an action might be carried out only if a condition is met. Conditional commands are useful when you wish to restrict the scope of an action; for example the operation If in column 1 then change/Monday/Tuesday might tell the system to make a text change but only to text in column 1 of a document.

Natural language is the term used to describe the language that humans use to communicate with each other. We will be describing natural language again in the chapter on artificial intelligence. I think it is true to say that one of the principal goals of computer scientists is a natural language interface that will allow you to communicate with a computer just as if it were another person. I must admit, I’ve never entirely understood the desire to design a natural language interface to a computer—in my experience humans don’t seem to do a very good job communicating with each other. However, the perfect natural language interface should enable a computer to interpret the meaning of a sentence like "Please edit the file I created last Friday—the one about the job application."

In practice, even the limited natural languages we have today can often be used only within a certain domain; that is the vocabulary and grammar is a highly restricted subset of a natural language. Real natural languages require a vast database and semantic net to resolve ambiguities. The ambiguities inherent in a natural language can easily be seen in English—just consider the sentence "I saw her with a telescope."

Some believe that the ultimate computer interface will have been created when natural language processing is combined with speech understanding and humans and computers will be able to communicate with each other verbally. However, the following untrue story demonstrates that there might be problems. During the testing and commissioning phase of a computer-speech interface project, the computer is programmed to simulate war games and a five-star general from the Pentagon is invited to take part in the exercise. The battle being simulated is a classic civil war battle and, at some point, the general is told that the enemy is advancing. He than decides to test the computer’s response against what happened historically, and says to the computer, "Well, should I attack, or should I retreat?" The computer replies "Yes." The general, being impatient, says "Yes, WHAT?". The computer, being polite, replies "Yes SIR."

Human computer interaction isn’t always initiated by the human operator. Sometimes, the computer has to tell the operator that something has gone wrong. We are now going to look at how the computer does this.

Computers communicate with people by means of a two-way dialogue; the computer invites you to enter a command or data, and then responds to your input. In the real world, things go wrong. Errors can be categorized into different types—typical errors are: typographic (you mis-spell a command or name), capture (you enter the wrong command and find yourself performing an activity you didn’t intend to), description (the correct action is carried out but with the wrong object or data), sequence (you forget where you are up to in a sequence of operations), and mode (you are not operating in the mode you expect; for example, you are in the operating system and attempt to enter word processing commands).

In addition to user errors, the computer cannot always safely respond to your request. Sometimes the operation is too potentially dangerous to perform without confirmation. Finally, system errors sometimes occur. These are errors put there by the designers of the opperating system or applications package (some companies are very good indeed at including lots of system errors in their software).

When the computer is forced to intervene, it does so by means of a dialogue loosely called an error message. The world of the error message is fraught with problems because three aspects of the human-computer system come together: the structure of the application and all its underlying software, the human interface itself, and human psychology. I think that I can safely say that, with the sole exception of the RS232C printer cable, no aspect of computing causes more anger and frustration than the error message. Let’s look at some of the problems.

Have you ever been in an aircraft and heard the announcement "Ladies and gentlemen, due to technical problems, the flight will be delayed one hour." This is a form of error message. Unfortunately, you are allowed to know nothing whatsoever about the nature of the problem, and, in any case, the message was not delivered until well into the delay. I once had a microprocessor development system that came up with the message "Error 155" when I first switched it on. I looked in the instruction handbook, found the error messages page, and decoded error message 155 as: "Error Message 155—See supplier." So, I phoned the supplier who said, "That’s interesting, we’ve always wondered what that meant ourselves...".

On the other hand, error messages can be too detailed or inappropriate. Imagine the effect of the following in-flight announcement on an aircraft, "Ladies and gentlemen, the flight is delayed as we have had to reduce power to number three engine because the compressor is running hot. It has been determined that 94% of all similar problems present no difficulty, as long as the power is reduced. However, in about 1% of cases, a fan blade shatters and punctures the engine housing." Clearly, too much error information can be as bad as too little—especially if it has no meaning to the user or the user cannot make use of the information.

One of the problems posed by an error message system is in relating the actual error message to its cause. A user operating at the applications level who is editing a file might receive an error message indicating that the operation cannot be completed. But where did the error massage come from, and what does it indicate?

Any reasonably complicated software has a hierarchical structure. When the application software wishes to open a file, it might call the disk operating systems’ file manager. In turn, the file manager might call the operations required to handle files. Finally, the lowest level software might be called to read a sector on the disk. An error can occur within any of these levels.

Suppose, for example, an error occurs at the lowest level because a word processor is told that a certain file is on a floppy disk, and the disk drive is empty. In this case, it is reasonable (indeed, necessary) for that message to be passed up through the layers of software to the user. The user can understand the massage and act on it. On the other hand, if a low-level error takes place because a sector can’t be read, the user doesn’t want to know that sector track 32, sector is bad. The user wants the system to make several attempts to read the file (because disk errors are sometimes intermittent). If the error persists and is fatal, the user wants advice—has all the data that was created in this session been lost, or should the system save it in a secure place until the underlying fault has been rectified?

A good error message system should achieve two goals. First, the error messages should be appropriate to the skills of the user—a beginner should not be left to figure out the meaning of the message. Second, the errors messages should be helpful and suggest how the error situation can be dealt with. We will soon look at help systems, which are closely related to error message systems.

Associated with the error message is the warning. A warning is a message that is detected when a legal operation is about to occur that may have potentially fatal consequences. The actual operation itself is not illegal. Suppose you have been editing a file called Chapter_5 for a few hours, and decide to take a back up copy of your precious work. If the syntax of the copy operation is Copy source,destination and you type Copy Temp_file,Chapter_5, you are going to lose all your work. A well-designed system might spot that the operation you are carrying out is legal but not sensible (few users ever edit a file for hours and then overwrite it with something old without saving the source). A suitable warning message might be "Do you really want to overwrite Chapter_5 and lose all the changes?".

Warnings are not always effective. You’ve all heard the story about the little boy who cried wolf. If you get a lot of warnings and ignore most of them, there comes a point at which you do not see the warning that should be heeded. This is particularly true of fire alarms—people tend to ignore them and assume that it is a false alarm. Suppose a warning message is followed by the request "Ignore Warning?" inviting you to continue if the warning is not to be acted upon. After a time, you get into the habit of anticipating this command and hitting the ignore key even before the message comes up. Eventually, there comes the moment when your finger is hitting the ignore key just as your brain is thinking "No!" You could solve this problem by requiring the user to enter a more complex response that, possibly, changes from time to time. Unfortunately, the production of spurious warning and error messages can be most frustrating.

The problem of effective warning messages can, to some extent, be dealt with by providing several levels of automatic warnings. An expert user might select a minimum level of computer intervention, whereas a novice might select a level that provides a much greater degree of feedback.

Interestingly enough, tests have demonstrated that alarms accompanied by verbal commands are often acted upon and not ignored. Alarms or warnings that are accompanied by shrill sounds are sometimes counter productive. There is at least one airline disaster during which the crew appeared more concerned with shutting off the alarms than with keeping the aircraft flying.

Some interfaces provide simple non-context specific automatic or background error-correcting mechanisms. When the user makes an error, the system attempts to correct the error, rather than to indicate it. For example, if you type copyy rather than copy, the system can be trained to recognize the error and corrects it automatically. This type of error correction can be enhanced by using context information; for example, if you are using certain specialized words and you mis-spell one of them, the system can substitute the correct work because it knows the vocabulary you are using. However, this system can be tremendously frustrating when it corrects correct information; for example, a very common typing error is "teh" caused by transposing the "e" and "h" when typing "the". However, the name Teh is a common Singapori name and the interface causes much irritation when, say, typing a class list with several Teh’s. Systems with automatic error correction often allow you to limit the scope of the error correction (i.e., turn-off some of the rules used for error correction).

Computers can do more than deal with errors. They can provide information about the application and simulate a User’s Manual. In the next section we look at help systems.

Not very long ago the concept of user friendliness didn’t exist. If you bought a program and wanted to use a feature that you didn’t fully understand, you had either to extract the information from the manual, or find someone who could help you. Today, many applications include a built in or on-line help system that can provide advice while you are running the application.

A good help system should provide help when it is needed and not clutter the screen with information you don’t require. There are several different forms of help; you can provide a tutorial help system to step through examples, a simple dictionary-based help system that looks up the information you need, or a context sensitive help system that gives you the "most" appropriate type of help. Consider a snapshot of a help screen generated by asking about circles in a session with CorelDRAW!

The HELP screen

In a typical system like CorelDRAW! the HELP function is invoked by clicking on the Help command (normally located at the top right-hand side of the screen in Windows-based applications). The HELP command provides its own menu, and you may select the search for help on item that invokes a dialogue box. You can then enter the topic you require help on (or use the scroll bar to step through the available topics), and the HELP system provides a list of related topics. You then click on the topic you want and the HELP screen is displayed.

The HELP dialogue box

This type of HELP system saves you the trouble of having to look up information in a manual. It does suffer from some significant limitations—if you don’t really know what you want, it can be very difficult to find the information. Part of the problem is one of terminology. Suppose you are a beginner using a word processor for the first time, and want to use larger letters in a document. You might be tempted to search for help on "large", "big", or "letter"—you might not know that the appropriate index words are font or character.

Context-sensitive help systems enable you to locate help on a particular topic without spending a lot of time searching. In a typical system, you may use the cursor to select the facility you want (but without clicking and launching the facility), and then press the HELP function key to get the help on the topic you selected.

Another form of help is the on-line tutorial. These are often animated picture shows constructed using multimedia technology (i.e., they may include text, sound, and animated pictures). The tutorial leads you through a presentation or demonstration of the software. I must admit that my own personal experience of the on-line tutorial has been negative. Many force you to move at a snail’s pace through material you already know. With a book, you can always skip through the boring bits.

Intelligent help systems operate by creating a model of the user’s expertise and then providing help within that framework. For example, if a user has displayed a knowledge of an advanced topic by using a series of commands in an appropriate fashion, an intelligent help system does not provide an elementary level of detail when help is requested. The goal of the intelligent help system is not necessarily to provide more help than other systems but to provide more appropriate help.

In order to implement an intelligent help system you have to be able to create a model of the information provided by the help system, and a means of representing the user’s level of understanding of this information. The intelligent help system is able to make deductions like:

IF the user has drawn a circle AND has selected the fill tool THEN help will be related to filling in a polygon object.

One of the buzz words of the late 1980s was hypertext, a system that promised to revolutionize the storage and manipulation of information. Before the introduction of hypertext systems, the type of information you would find in a book or manual (or even a novel) was stored linearly, page-by-page. You access information in a book by skipping through pages until you find what you want, or by looking up the information in the index. A problem with the book or any other linear form of information is that you have to go through it at the rate and in the order dictated by the author.

In a hypertext system the information is structured in a complex way and held together by large numbers of invisible links. The next figure illustrates the concept underlying hypertext. Imagine that the uppermost plane, marked visible plane, is a conventional text item such as part of a computer manual. This level can be read through page by page just like any other document.

Hypertext

The block dots or nodes in the upper level represent words and phrases that are related to or linked to other items in the hypertext. For example, one of the layers might be a definition layer. Suppose you are browsing through a hypertext document and you wish to know the definition of a term. You simply click on it and up comes a definition (usually in a separate window). The precise mechanism for carrying this out will vary from system to system. For example, words that are linked elsewhere are sometime highlighted or in a different color. The advantage of the hypertext mechanism is that it gathers the information you need—no two readers will require the same sequence of hypertext links.

There are three hidden planes. Information can be linked in many different ways. Suppose the hypertext was concerned with history. The uppermost level might be a conventional narrative, while the lower levels provide the supporting material. One level might cover events going on in other societies at the same time. Another level might give potted biographies of the personalities concerned. Another level might give extracts from associated historical documents, and so on.

You might think that hypertext sounds almost too good to be true. In a way, it is. Although the underlying concept is quite clearly sound, it is strongly implementation dependent. As you can imagine, the construction of a good hypertext system requires a lot of effort and labor. I have seen systems that provide a very poor level of information and are little more than a toy. For example, what is the point of a hypertext system that provides postage-stamp size maps when most homes have much better atlases on their bookshelves. Another problem associated with hypertext (and help and on-line tutorial systems) is that of navigation. (i.e., finding your way around the links). You can sometimes become lost in the mass of links between information and find it difficult to get back to a previous position. Moreover, the author of the hypertext document creates the links and therefore has to anticipate the future users’ needs.

The early 1990s saw the rise of a new technology—multimedia. Essentially, multimedia implies more than one medium (e.g., text, sound, still images, and animated video images). Hypermedia is an extension of hypertext to include objects of any kind as links. In hypermedia, an object can be a fragment of digitized sound that is played when its link is activated. Other objects might be animated diagrams or even video clips. Because both sound and image objects take up a lot of storage, hypermedia systems normally require CD-ROMs to store the necessary amount of data.

The explosive growth of the Internet in the mid 1990s has further increased the interest in hypertext and hypermedia. The Internet is a wide-area network that links countless numbers of computers worldwide. The Internet does not belong to any single government body or organization and there is no controlling body. Internet users can access a global hypermedia system by means of the world-wide web, WWW.

The majority of computers are used by individuals performing isolated tasks. However, several people often work together on related tasks in organizations—this is sometimes known as computer-supported cooperative work, CSCW. Software that supports the cooperation of several users is called groupware. Ultimately, groupware can lead to the so-called paperless office, where electronic media replaces all conventional paperwork.

Groupware is concerned not only with interaction between people and computers but with communication between people themselves. Groups of people require communications facilities based on networking. An office might have a standard memo that can be electronically transmitted to any individual or groups of individuals within the organization. Diaries and appointment books are shared to enable people to coordinate meetings and similar group activities.

Groupware demands a consistent interface to all users to reduce the cost of training and allow people to move easily between terminals. Although the software has a constant interface, all users don’t necessarily share the same interface. Clearly, not all users in an organization will be able to access all the information. Similarly, the interface must cater to both naive users and expert users.

One of the problems associated with groupware is synchronization. Suppose two people decide to edit the same test at the same time. If they both update a particular paragraph simultaneously, which one does the system record?

A very simple example of groupware is provided by Microsoft’s Windows for Workgroups. This is an extension of the popular Windows operating system to a network. Users can access each other’s files—subject to a simple system of permissions and passwords. Two tools are provided to facilitate group operation, E-mail and a scheduler that allows users to view each other’s diaries.

New computer interfaces, both hardware and software, are continually being introduced. Each manufacturer believes that their system is better than that produced by the competitors. Although you can evaluate the performance of various pieces of software on the basis of instruction execution, it is much harder to evaluate a user interface. The user interface "takes the human into the loop" and therefore its performance depends on the characteristics of the human. Unlike computer hardware and software, humans vary widely in terms of their performance. Moreover, they can exhibit sharp learning curves as they learn from experience, and this can decline as they get tired or bored.

Research into the human computer interface is made more difficult because of human psychology. Research was once carried out into a new way of writing programs, and the group being studied demonstrated an increase in productivity that validated the new technique. However, when the old technique was reintroduced, the programmers also demonstrated higher productivity. It turned out that the result of being involved in an experiment had more effect on the programmers than the changes in program design that were being evaluated.

A typical experiment involving the user interface is described by Gould, Lewis and Barnes in ACM transactions on Office Information Systems (Vol. 3, No. 1, Jan 1985 pp22-34). Text editors and word processors operate by moving a cursor about the screen. It is natural to ask whether the speed of the cursor movement affects the user’s performance (and therefore productivity)?