This article evaluates how well modern computer-based writing tools support the writing process. The writing tasks that need to be supported by tools are identified and a set of criteria for tools to support writers is developed. Current writing tools are then evaluated against this criteria to determine how well they suppose writers. The article concludes by examining the implications for the design of new software to support writers.
This article evaluates how well modern computer-based writing tools support the writing process. The remainder of this section describes how writing tools can influence what is written and therefore why they need to be designed carefully. Section 2 describes the writing tasks that need to be supported and develops a set of criteria for tools to support writers. Section 3 evaluates current writing tools against these criteria, and section 4 looks at the implications for the design of new software to support writers.
The aim of writing tools is to support the tasks performed by writers. Writing tools can influence and even change what is written and therefore need to be designed carefully. The medium of written communication will now be examined to explain this point. McLuhan (1967) wrote that “the medium is the message.” Kay (1991) understood from this that the most important aspect of any communication medium is that “message receipt is message recovery,” i.e. anyone that wants to receive a message embedded in a medium must first have “internalized the medium” so that it can be “subtracted out” to leave the message behind: you have to become the medium if you are to use it.
The computer is a medium and writing tools are media within a medium. It is necessary to look at the computer as a medium for two fundamental reasons: new media change society and culture, and the design of writing tools can effect the medium of written communication. It is therefore important to consider the impact writing tools have when designing them. The introduction of the printing press will now be used as an example of how a medium can change culture and society.
Today it is easy to see the profound effect writing has had on our culture because there is detailed knowledge of many cultures that do not have writing, and we have seen the impact writing and printing has had on such cultures (McLuhan 1970). Before the development of printing, books and other written material were produced by hand by highly skilled literary craftsman. Texts were often embellished with multi-coloured drop caps, for example, and took a long time to produce. Written text was therefore vary rare and available to few people.
The printing press changed the production of written material, the availability of such material, and the societies that used them. The total cultural effect of the invention of printing cannot be overestimated (Johnson 1960). Printing enabled new ideas to be disseminated easily so thousands of readers could benefit from them rather than just a few dozen. For the first time people were able to learn about historical events right back to the time of Aristotle and Plato. The availability of printed history books put the world in context. Information previously confined to monasteries and a few schools was released and available to all. More people were able to be taught to read and educated as more books were published. The Dark Ages rapidly came to an end and “probably more than any other single thing, the invention of printing ushered in the modern era” (Johnson 1960).
Although the literature does not agree exactly when the printing press was invented, it does concur that Johann Gensfleisch zum Gutenberg (around 1400 to 1468) made the most important contribution: printing with movable type. Although Gutenberg did not invent the printing of books—the Chinese had long before mastered the art of printing books with wooden blocks and engraved metal plates—his fundamental contributions were those that make the modern word processor so useful: the ability to edit and correct a text that is identical in every copy, i.e. the uniform edition preceded by critical proof reading.
The printing press accelerated the growth in literacy and changed the message. As Kay notes, “the main point is that the [printing] press didn’t do it just by making books available, it did it by changing the thought patterns of those that learned to read” (Kay 1991). These changing thought patterns changed the message. The message changed because the language and writers changed. When written text was produced by hand, writers had to think very carefully about what they were going to write because mistakes would be very difficult if not impossible to correct. The structure of sentences and their content were considered carefully and each word was chosen with great care. The style, content, and structure of printed text was very different to that produced by hand simply because of the method used to produce it. The ease of correcting mistakes afforded by the printing press meant that less thought had to be given to what was written. A simpler sentence structure developed which inevitably changed the language of every country that used printing.
In contrast, the results of a study performed by Haas (1989) show that composing text with modern word processors might have the opposite effect. Haas performed a study to ask two questions. First, does hardware and software change writing efficiency? Specifically, do longer and/or more quickly produced documents result from the use of advanced writing technology? Second, what is the effect of different text-editing technologies on the quality of written products? The fifteen subjects were experienced writers—faculty, administrators, and system designers from Carnegie Mellon University. They were asked to write letters on various subjects to known and unknown audiences using longhand and word processing. The subjects that used a word processor took longer to compose and used more words than those using pen and paper.
Word processing users focus more on local changes to the text such as spelling, grammar, and punctuation, than when writing in longhand (Hatty 1993). If hand-written manuscripts are compared with those produced on a printing press and longhand text is compared with word-processed text, just as the printing press changed sentence structure, so might a word processor. It is the editing facilities of the word processor that make it an extension of the printing press with moveable type rather than the typewriter.
It could be said that the typewriter is a halfway house between hand writing and word processing because it is (nearly) equivalent in quality to the printing press and is a tool for writing. The typewriter does provide a better presentation than handwriting and, as McLuhan (1970) observed, “as a writing machine, the typewriter is far more efficient than the hand. Your fingers type automatically and simultaneously what you’re thinking. When you handwrite, you write carefully and slowly, and usually less colloquially.” However, standard mechanical typewriters do not provide editing facilities and are certainly not as forgiving as a pencil, paper, and eraser. Electric typewriters now offer limited editing facilities, such as at the sentence level.
Desktop publishing is a contemporary example of the impact that a change in the technology of producing written materials can have. It has been said that the freedom of the press is available to anyone who can afford a press. In 1985, when Apple introduced the Laser Writer laser printer for their Macintosh personal computer, (almost) anyone could afford a press. Typeset quality text and graphics became available to the masses for producing written material ranging from books and newsletters to simple fliers.
The next section describes the tasks writers perform when they write and develops several criteria for support.
This section describes the tasks writers perform and the writing strategies they use, and lists several criteria for tools to support writers.
Flower and Hayes (1981) developed a model that describes the tasks writers perform when writing. The three main tasks are planning, translating, and reviewing. The planning task is divided into generating, organizing, and goal setting sub-tasks. Translating then converts the plans into text. Finally, the text is reviewed by reading and editing sub-tasks. These tasks are set in the context of the task environment which contains the writing assignment and the text produced so far, and the entire writing process is co-ordinated by continuous monitoring.
The Flower and Hayes model is a cognitive model. Cognitive models are necessary for understanding the user but task models are necessary to develop software. Text editors are one of the oldest applications and have been the subject of much of the task modelling research using the Goals, Operations, Methods, and Selection rules (GOMS) approach (Card et al. 1983) to study line editors such as vi. The focus of this task modelling research is on editing operations such as insertion and deletion. However, editing is only a part of writing, so the research on line editors makes only a small contribution to task modelling for writing software.
One of the problems faced by the designers of writing tools is that different writers have different writing strategies, such as the Beethovian and Motzartian strategies. When a writer follows the Beethovian strategy, all of the ideas are written down and then edited. When the Motzartian strategy is followed, all composition and editing of ideas is done in the writer’s head and then written down.
Sharples (1991) has developed an external representation model, shown in the table below, that defines various stages of the planning and text produced so far categories of the Flower and Hayes model. The end product of the writing process is box six (instantiated linear organization). The writer may start from any box and choose any path from box one to box six, for example one ⇒ three ⇒ six. This model can cope with three of the above strategies: plan ⇒ draft ⇒ revise, outline ⇒ draft, and draft ⇒ revise.
Techniques: Note-taking (verbatim), Collecting quotes
Techniques: Following a thread, Writing as dialect
Techniques: Organising notes, Filing
Techniques: Linear planning
Techniques: Drafting text, Revising text, Copying text
The following table shows the properties of various media that can be used during a writers path through the boxes in the external representation model (key: Yes–has property; No–lacks property; ?–debatable).
|Supports Browsing of Large Documents||Supports Reordering||Supports Non-Linear Organisation||Reusable End Product||Supports Annotations||Supports Indexing|
|Sheet of Paper||Yes||No||?||No||Yes||Yes|
|Back of Hand||No||No||No||No||No||No|
There are many different contexts in which writing software can be used and different users have different needs and write in different situations. There are two broad classes of users: those that need day-to-day writing tools or writing support tools, and those that need teaching or writing training tools. It is possible to provide a different solution for each of these classes as shown in figure 3.
|Example Writers||Aims of Writer||Writing Tools|
|Academic, editor, publisher
Writing teacher, small firm manager
|Improving own text
Improving someone else's text
Write to a specification
Learn how to write
Teach someone to write
|Augmented word processing tools
Style sheets and checkers
Teaching and assessment tools
Writing support tools should be interactive and immediately available but operate in the background so that the user is always in control. They should run in familiar word processing environments and for those who require it, perform corrections or provide correction suggestions. Tools in a support role should not offer intrusive comments or detailed explanations because users do not need them. Tools should operate as part of the natural writing process and be available when users need them. Feedback should only be given when it is required rather than continual comments on everything that is incorrect because style, for example, is rarely incorrect but inappropriate in the current situation.
Writing training tools should cover all aspects of the writing process from pre-writing and composing to post-writing and be sensitive to the writing context. Such tools should be well supported with tutors because humans cannot be replaced with computers. In contrast to writing support tools, a tutor should not be allowed to make corrections because the aim is to learn by making the correct change. Explanations and critiques of what is wrong with a piece of text should be given in different ways and with different levels of detail for different classes of learner. Structured feedback on writing rather than just what is required for the given task should be provided and a full report should be given in a printed commentary and batch output even though the tool is interactive. The self esteem of poor writers should be preserved by reducing the distinction between errors and inappropriateness, perhaps avoiding the concept of error all together.
Cooper and Matsuhashi (1983) consider writing to be a process involving planning, decision making, and problem solving, and ask “what plans and decisions must a writer make in order to produce a written text?” Holt (1992) states that the development of writing tools has two straightforward objectives: to make writing easier and more effective, and to assist in distributing written communication quickly and more effectively. The Flower and Hayes model suggests that writing tools support planning, translating, and reviewing.
Users have tasks they need to perform and software has tasks that it is designed to support. Matching the tasks of writers with the tasks supported by the software will increase the usability of writing tools. The aim of user centred design is to produce a seamless interaction where all tasks are perfectly supported (Norman 1986). This is difficult in practice because by the time a task has been analyzed and defined well enough to provide a perfect match, the users have changed; their tasks have changed and they have raised their expectations because they have become more sophisticated. Two key questions need to be asked when evaluating software for supporting writers. First, how closely matched are the tasks performed by the writer and the tasks supported by the software? Second, how well are the matching tasks supported?
The next section examines how well current writing tools support writers.
This section surveys the current tools for supporting writers. There are four broad categories of tools that are available for supporting writers, each focussing on a different aspect of writing: text editors; text checkers; document creation, planning, and organization tools; and WYSIWYG displays.
This category contains the standard cut, copy, and paste operations which apply to graphics as well as text. An undo facility helps writers to experiment, especially with formatting which is often difficult and time consuming to reverse. Find and replace operations allow writers to quickly find occurrences of words or phrases and change them if necessary. Current find and replace tools are quite simple and more sophisticated versions that deal with sentence structure, grammar rules, and formatting would support writers better.
This category contains tools such as spelling, grammar, and style checkers that are used after some text has been produced. Spelling checkers flag differences between the spelling of words in the text under analysis and the words in the spelling checkers’ dictionary. Dictionaries exist for different languages and different dialects of the same language, for example UK English and American English.
Grammar checkers comment on the structure rules of the language used in the text. The integrated grammar checker of Microsoft Word 5.1 for Macintosh analyses the following elements of grammar:
A style checker offers advice on how to adhere to a certain style and flags violations of that style. Style checking is the most difficult type of checking to perform. Style is the expression of ideas within the constraints of the grammar of a language. However, the term constraint should be used loosely because one aspect of style is how the rules are broken, in creative writing for example. Style is arbitrary and cultural, and is generally classified into a number of different categories, such as business style, personal style, and scientific style. Each category specifies the layout and language of the text because they are used to communicate to different audiences for different purposes. Aspects of style include wordy expressions, jargon, clichés, and pretentious words. These aspects may be desirable in some cases but not in others, depending on the audience and the purpose for communicating. There are three categories of style checkers: post-writing analyzers, pre- and post-writing analyzers, and integrated writing analyzers. Post-writing analyzers are the most common type of style analyzers. Text that has already been written is submitted for analysis. Three examples of post-writing analyzers are PC Style, Writers Workbench, and RUSKIN.
PC Style takes an ASCII text file and gives advice on the style used in the text. A statistical report is produced which provides the number of words including action, personal, and long words, the number of sentences, words per sentence, syllables per word, and reading grade level. Long words are in relation to the average length of a word in English which is five characters. The terms personal and action words are used in American English and illustrates that style is cultural. The reading grade level is used by American schools to assess the reading ability of children.
Writers Workbench is an automated tool for teaching and assessing writing. This tool originated a set of ad-hoc UNIX scripts in the AT&T Automated Writing Lab. It is difficult to say whether Writers Workbench was successful in its aims as experiments showed improvements in the groups that used Writers Workbench and the groups that did not. The conclusion that can be draw is that all subjects improved because they were made aware of style, something they had not previously considered.
RUSKIN is an experimental AI-based tool for report writing and was developed in the late 1980s using user-centered design. Funded by the Artificial Intelligence in Training Programme and supported by IBM UK and AT&T-Bell, it was developed for use on IBM PCs in SD Prolog. The main feature of RUSKIN is its use of context variables. A context variable is a variable of the situation in which the text was produced and which is likely to affect the nature of the writing. Examples of context variables include audience, subject matter, and author. An audience variable describes the age, number, subject familiarity, educational level, and attitude to the subject of the intended audience. A subject matter variable describes the document type/genre, subject complexity, and the technical complexity of the vocabulary. An author variable describes the author’s skill at writing in the genre and his or her familiarity with the subject.
RUSKIN is driven by its style rules that relate text to contextual variables. The rules summarize the correlations between text and context that constitute good and bad or effective and ineffective writing. Figure 5 shows two examples of RUSKIN style rules. Although RUSKIN uses expert system techniques it is not a true expert system because it does not combine output from several rules, it does not perform conflict resolution, and it passes only once through the text.
IF the educational level of the audience is low THEN there should be few complex sentences IF the text is long AND the text is complex AND the relationship is formal THEN headings should be used
Integrated style analyzers are usually more convenient to use because the user does not have to leave the software package. For example, Microsoft Word 5.1 for Macintosh is a word processor with an integrated style analyzer, which analyses the following list of style categories. Availability does not mean use, however, and there is a lack of data on the use of such tools.
Style and grammar checkers can also be used for research purposes. For example, they can be used as input to programs such as neural networks that induce style rules. They have been used to produce data for research into the differences between children’s writing in Scotland and Norway.
Many tools are available that take the drudgery out of creating documents such as tools that generate tables of contents and index pages, and outliners that enable writers to view and create the structure of their documents as a set of headings. These tools are integrated into Microsoft Word 5.1 for Macintosh.
There are several methods for planning and tools are available to support each approach. For example, hypertextual organization can be performed with HyperCard, SuperCard and Guide; outlining can be performed with Microsoft Word; and mind mapping can be performed with Inspiration.
Before What You See Is What You Get (WYSIWYG) displays, text formatting and character styles such as bold, italic, and underline had to be explicitly indicated with cryptic codes such as .jo and .nj (to turn justification on and off). Writers were burdened with the language of the typesetter which distracted them from the task of writing. The WYSIWYG display, popularized by the Macintosh, enables writers to quickly and easily control the format and styles used in their documents. Such flexibility and freedom has disadvantages such as ‘fontitis,’ the use of many different fonts in the same document without any clear design pattern. Writers now use the language of the graphic and page layout designer rather than the typesetter.
This section concludes with a look at the implications for the design of software writing tools.
Cognitive and task models have been developed for writers and writing tools (section 2). Cognitive models provide information about the way humans do things, and they help to develop the functionality required of software and form the basis of the functional design. Task models tell us how that functionality can be carried out and provide the basis for the dialog design.
One of the difficulties of producing usable software is that users are highly adaptable and learn to work around problems. For example, the quirks and modes of the vi editor can be overcome to make vi an adequate tool, if not a highly usable one. The aim, however, is to produce the most usable software possible and this requires user evaluation of existing writing tools. The usability problems of writing tools are difficult to identify because users are so adaptable; users work around problems and no longer see them as problems.
Writing tools such as Microsoft Word take advantage of the windowing environment they operate in. One of the benefits of using a windowing system is that a substantial amount of the functionality of an application is already implemented by the environment. The user is able to transfer much of their learning from one application to another and need only learn the specifics of a new application.
The task and cognitive models described in section 2 view writing as proceeding in distinct steps, for example moving from one box to another in the Sharples model. In reality, writing involves rapid multitasking between planning and writing. Writing is like rapid prototyping: some text is produced and reviewed which gives rise to new plans and ideas which produce more text. It is a cycle of planning, writing, and more planning until the text is complete. What is required is a set of integrated tools that can accommodate this approach. Modern tools have a cognitive model but start with the external representation. To improve writing tools, the cognitive model must feed into the external representation, task model, and thus the writing tool.
The needs of users have been identified and cognitive models have been produced (section 2), and tools that aim to meet those needs have also been produced (section 3). The problem is that these tools have not been integrated into the tools writers are using, such as word processors. For example, Inspiration is good for planning writing but it is a stand alone application. Writers plan then write and writing produces new plans; Inspiration should be fully integrated into a word processor to be more useful. Another example is hypertextual outlining in which the end product is in box four or five of the Sharples model. Integration with a word processor would enable the writer to get to box six and produce a complete piece of text.
According to Hatty (1993), “the current role of computers in writing it seems is one of facilitation, reducing the effort involved in the preparation of the document, especially in the editing of a document” and that “the currently available writing tools do not enhance the standard of writing produced on them.” However, Hatty looked at the quality of writing, not the usability of writing tools. Hatty observed that a lack of expertise with a tool diverts attention from applying relevant knowledge for composing. A tool is a hindrance if it does not easily support the move from longhand drafting to the word processor.
The computer is not as effective or easy to use as a pencil and paper for creating a first draft. Current word processors do not enable the user to place a layer on top of the text on which hand written or drawn annotations can be added graphically. The only technology set to change this is Pen-based computing. Even so, Hatty (1993) concludes that even “a pen-based system would add little in the way of functionality to a word processing system. The only advantage being the ease of learning because of the familiar interaction style used.”