Children who can converse well and have the ability to write still have difficulty composing long stories by themselves. There are several reasons for this difficulty in moving from oral conversation to extended written narrative. The main innovation discussed in this paper is designed to address two of these barriers. These problems are the lack of conversational support in writing and the absence of an immediate audience for one's discourse.
In solitary writing, the child lacks the familiar turn-taking structure of conversation, and the supports offered by the other person or people in the conversation.
When people converse they help each other in numerous, mostly unintentional ways. They provide each other with a continual source of cues - cues to proceed, cues to stop, cues to elaborate, cues to shift topics, and a great variety of cues that stir memory. ... In written composition, all these supports are removed. (Bereiter and Scardamalia 1982)
Fortunately, the distinction between conversational storytelling and story writing is not as severe as the distinction between a discussion and an essay. In conversation, the rare occurrence of a lengthy, mostly uninterrupted section of talk for a single speaker is often a story. Stories usually span several utterances and require that the teller take over the conversation temporarily (Sacks 1970). So children have more of a conversational antecedent for this type of writing than they do for essay writing. This makes story writing a good way to bridge the gap between conversation and other forms of writing that have less connection to oral communication.
However, even in the conversational story there are frequent interruptions to provide backchannel feedback, request clarification, and express interest or distraction. There are many types of conversational support from the listeners when stories are told in conversation (Polanyi 1989). If a story is unclear it can be clarified afterwards, or listeners can ask questions during the telling of the story. There is no such opportunity in writing. When children in grades 2-5 tell stories to adults, they usually do not require the step-by-step prompting that younger children do. Still, they are aided substantially by the adult in telling their story. Even an adult speaker telling a conversational story is helped along and motivated by feedback, cues, comments, and questions. While story writing is a good first step to more extended writing, it is not without challenges. Skill at writing stories involves more than simply the combination of oral storytelling skill and writing ability.
Another difficulty for young story writers is that the audience for whom one is writing is not immediately clear, as it is when speaking to someone. The writer and reader do not share the same physical or textual context, as interlocutors do. So in addition to lacking the interaction of conversation, the writer also lacks the clear idea of whom the communication is directed to.
The linguistic and cognitive demands of writing are much more exacting than oral discourse. Without the supportive prompts of a conversation and its ambient context, the writer has to sustain a coherent discourse, by retrieving and organizing information for an envisaged reader, in an appropriate mode and style. (Martlew 1986).
This change has several repercussions. Questions, replies, and assessment of agreement or disagreement occur in writing only "as echoes of the forms of the spoken language." Assumptions of shared knowledge can be violated when an unanticipated party sees one's writing, a situation which happens much less frequently in spoken conversation. So writers cannot assume as much knowledge on the part of their audiences.
Spoken texts may leave information implicit because the speaker knows what the hearer knows and because he can assess as he speaks whether he has been correct in his assessment. Hence writing tends to be marked by greater explicitness and elaboration than speaking... (Kress 1982)
Even very young children speak differently to different audiences. By age four, numerous studies have shown that children clearly do adapt their speech to the listener (Menig-Petersen, 1975; Sachs and Devin, 1975; Waterson and Snow, 1978; Anderson and Prosser 1993). Children's ability to consider their listener is well-developed by the time they reach grade 2. Not having the necessary cues to distinguish who the audience for writing might be makes for one additional difference between speaking and writing, and creates one more barrier to writing a lengthy composition.
The stories that children write, as opposed to other forms of speech and writing, may depend less on the intended audience and be less inhibited by this problem. At age 11, children do little to distinguish the stories they write for children from those written for adults, only using "once upon a time" much more frequently as a beginning (Martlew 1986). Although children do not write fundamentally different stories, they are clearly aware that distinct reading audiences each require appropriate types of writing. For these reasons, story writing might be a good starting point that can allow children to deal with an absent audience. In grades 2-5, the lack of an immediate audiences is noticeable, but the inability to imagine the audience and adapt to the reader's needs does not inhibit story writing. This could make story writing a good exercise in which the concept of audience can be introduced.
The software developed in this project is based on the notion that planning, writing, and revision of stories are separate processes within the broader writing process. Writing instruction can therefore focus on a particular process. This project takes this three-part categorization of writing as a foundation. The point of this project is not to evaluate the effectiveness of an approach that focuses on planning, writing, and revision. The conversational computer character, not the segmentation of the writing process, is the main feature being analyzed. The conversations that the user and the character have are based on this segmentation of writing process, however, and this framework is essential to the software's design. So it is important to justify the separation into these three processes as appropriate and useful in writing education.
The three processes of writing (along with a fourth process, evaluation) are widely used in the teaching of story writing. Education at all age levels and in all types of writing, in fact, employs this segmentation of the writing process. This allows interventions to be focused on one particular process at a time (Huntley 1986, Daiute 1985a). Gathering ideas about the main story elements usually occurs during the time before writing. Focus on writing without distraction from brainstorming or proofreading is important when the story is actually being set down. After this has been done, additional changes and additions can focus mostly on clarity and style rather than major high-level revision. When a high school writer is engaged in the first process of planning, that writer thinks about broader ideas that form the topics of paragraphs, for instance. During writing it is useful to focus instruction on the sentence level. During the revision process, writers can improve if they think about the individual words and about what may have been left out in between sentences. By creating appropriate educational interventions for each of these processes, instruction can help people improve as story writers.
These processes are not just features of writing pedagogy, however. The writing process as cognitively carried out by all writers is generally divided into these components. Individuals may perform different specific activities in their own planning, writing, and revising processes, but for all writers the cognitive processes involved in story creation fall into these three categories. They can be further characterized by processes within each larger process, which have been identified by researchers studying non-fiction writing. During planning, writers generate ideas, organize these ideas, and set goals for their writing. The writing process itself consists of translating these organized ideas into written language. Finally, revising involves reading and editing, two processes which detect and correct problems in the text (Hayes and Flower 1980, Flower and Hayes 1981). Even writers who are poor at planning and writing and do little revision do indeed appear to plan, write, and revise (Perl 1979). The underlying cognitive processes involved in most writing are what make the pedagogical division of the writing instruction into planning, writing, and revising a good and useful one.
These processes often are carried out in sequence, but they are processes that writers engage in at different time, not clear and distinct sequential stages (Daiute 1985a). Division of the entire writing process into three sequential phases is only a rough approximation which may be of use in educating young writers but is not an ideal representation.
Teachers and parents can assist children in writing stories in many ways. An adult can supply ideas, providing characters or plot outlines which the child can use in writing a story. Adults can also help with the mechanical process of writing by typing a story as children dictate. In evaluating stories, adults can provide guidance about what good story qualities children should build upon and can determine where they may need help in the future. The method of assistance this project focuses upon is not, however, related to the subject matter, to assistance in typing, or to evaluation of a story. The method used here involves help on abstract elements of the writing process, a structural form of assistance sometimes called procedural facilitation (Bereiter and Scardamalia 1982).
This category of assistance as discussed here has included many sorts of educational interventions. These range from oral prompts to mimeographed story planning checklists to computer writing environments with a wide array of interactive features. A procedure for revision, such as reading each sentence and thinking of an evaluative statement to follow each one (Bereiter and Scardamalia 1982), is procedural facilitation. The category also includes planning assistance to help with the development of story elements. The sorts of interventions not in this category include simply asking what happened next or suggesting particular subject matter.
Procedural facilitation can provide clear benefits that may be observed when the child is using the software or other prompting mechanism. Salomon distinguishes this as the cognitive effect with the software. This is related to how good of a helper or tool the software can be. Another effect is the lasting cognitive impression, or "cognitive residue" in Salomon's language, made by a computer program. This is termed the cognitive effect of the software. This is related to the pedagogical value of the tool. An educational intervention that succeeds in having a lasting effect might serve to activate an existing skill that is suitable for writing. It also might help a student internalize some useful operation and add an additional skill to that child's repertoire. Finally, it might help the child think more explicitly about the writing process, and then internalize this metacognitive ability (Salomon 1990).
These two effects - benefit during use and lasting benefit - are not exclusive of each other, but neither are they always complementary. A computer writing tool that helps a user write may lead to dependence on that tool, so that the writer actually becomes worse when deprived of that help. Or a tool may provide help that the writer internalizes as the cognitive process of writing is improved. Then, when that writer later sets out to write without the explicit help of that tool, its benefits will nevertheless be felt. One tool may provide its main benefit only after the writer has discontinued use of that tool. Another tool may be good for some specific task like proofreading but prove impossible for the writer to learn from. Of course, some tools provide neither sort of help, and some are designed well enough to be able to offer both sorts of cognitive benefit.
Although the innovations provided by a computer character have clear educational motivation and potential, some teachers question whether elementary school writers should use computers at all. Using a word processor for writing, as opposed to writing on paper, may or may not hold great benefits for students who are beginning to develop their writing abilities. A comparison of 32 studies, for instance, seems to indicate that while students who use word processors write better-quality compositions, they do not, as hoped, have better attitudes about writing (Bangert-Drowns 1993). Part of the problem is that in analyses like this one and often in discussion in general, the question considered is usually framed in terms such as "Do computers help children become better writers?" That question is so broad that it has no one answer.
Children's cognitions are not affected by "Television" or "the Computer;" they are affected by specific kinds of programs with which they carry out specific kinds of activities, under specific kinds of external and internal conditions for specific kinds of goals. ... the question is not whether "the Computer" affects minds, but whether the combination of particular kinds of programs that entail particular qualities under specific conditions of activity, goal, and cognitive involvement can have (or can be designed to have) some lasting cognitive effects on children. (Salomon 1990)
A study in a Massachusetts elementary school determined that first-grade students who used word processors did significantly better, based on overall measures of writing quality and development, than their counterparts in the same class who hand-wrote stories (Keetley 1995). One study of more than 50 Italian fourth grade students, lasting several months, suggests that word processing can help to motivate students and result in more writing (D'Odorico and Zammunner 1993). At early grade levels it can often be easier for children to type than to write by hand, so longer stories can be written on the computer. Word processing in the elementary school is not always effective, but it can offer benefits in many situations, as the above studies show. While computers are not always an appropriate classroom technology for every educational goal - no technology is - software that uses the computer as a platform for writing clearly can succeed in bringing educational benefits to students.
The computer-based educational interventions explored here go beyond word processing, and are designed for children in grades 2-5. This group can particularly benefit from educational interventions to help them write in the absence of conversational prompts (Bereiter and Scardamalia 1982). By the second grade, proficiency with the mechanics of computer writing and use is high enough to make use of the system feasible. Much of the research on techniques to help children move from conversation to writing have focused on grades 3-6 (Bereiter and Scardamalia 1982, Hidi and Klaiman 1984), but children are ready for similar interventions by grade 2. In the recent past, facility with typing on the computer was often not acquired until later. This suggested that computer-based interventions to aid writing development should occur mainly in the second half of elementary school. Children are now beginning their computer use at earlier ages. Designing software for an age range a year younger is quite reasonable, given the increasing computer capabilities of young children, the resources of elementary schools, and the appropriateness of these educational interventions for grades 2-5.
Education researchers at universities and commercial software developers have designed several pieces of writing software for elementary school students, some of which add interesting capabilities that assist writing in ways a word processor does not. The features of existing story writing software do not include a conversational computer character, but those innovative features that do exist should be examined and evaluated before creating any new program. Even software that is not innovative but is well-designed and usable can provide a model for the interface of a new system that incorporates more directed educational interventions. Finally, the failures and successes of current programs, as documented by empirical studies or observed in use, should inform any new effort to develop creative writing software.
The most popular and best-selling writing software for children is, unsurprisingly, not the most innovative. The programs usually offer an augmented word processor that is designed to be easy for children to use. The editing and proofreading functions of the word processor are few, but added to it are tools for illustration. Such multimedia features make better selling points than most useful educational features would. However, some of these systems achieve their goal of being easy to use, fun to work with, and motivating, and thus have some measure of success.
The creativity suite Kid Works Deluxe by Davidson also combines a word processor with a paint program, for use by children ages 4-9. It typifies the commercial "first word processor" category, which adds multimedia capabilities like speech and drawing to a simplified word processor. The menus have graphical depictions of creatures instead of words like "File," "Edit," and "Format." These ambiguous symbols are more confusing than the words that are normally used, and of dubious benefit to children. The children who use the software have, after all, come to the program in order to write and do have the ability to read. The program does have some features well-suited to children. Children can create rebuses and switch the figures back and forth from words to pictures, learning spellings they might be unsure of. The text-to-speech system allows children to enter separate phonetic spellings of words by way of a speech editor, an option on the Cricket menu. This way, unusual words can be both written and pronounced correctly. Finally, there are "story starters," templates to help children begin writing that are common features in children's word processing software. Only one of the story starters supplied actually is designed to help start story writing, the others being templates for poetry, certificates, letters, and other forms of writing. This prompt suggests that the writer tell about the strange picture the artist is drawing. The pictured canvas is left blank so the child can draw in it (Davidson 1995).
The Amazing Writing Machine by Brøderbund is a similar suite. This program offers specialized sets of writing tools for creating stories, letters, journals, essays, and poems. Children can further pick one of nine writing environments genres within which to write, each corresponding to a genre such as fantasy, horror, and romance. Creative writing assistance is provided with Bright Ideas, a feature like Kid Works Deluxe's story starters. Using this, students can customize a set of existing stories. They can choose words from a list or type in their own words (Brøderbund 1994). This allows them to create their own story before they are ready to do all the writing from start to finish. This program is well-designed and has won accolades and positive evaluations from several publications (ASCD 1997).
A more innovative approach to writing for a slightly younger age group (grades K-4), also by Brøderbund, is Orly's Draw-a-Story. This product is similarly a suite of writing and drawing tools, with four main stories that can be modified by children. However, in Orly's Draw-a-Story the Jamaican girl Orly serves as a central character with whom the user can interact in simple ways. Instead of offering help with story elements, Orly talks about Jamaican culture. She and the characters in the stories and provide story ideas and encouragement, though not in textual conversation (Brøderbund 1997).
Storybook Weaver Deluxe is a story-building suite of writing and illustration tools (MECC 1994). Reportedly included in the 1994 version of the software are "several story starters that ask students who will be in the story, where the story will take place, when it will happen, what will happen first in the story and how will the story end" (ASCD 1997). Yet such a feature is absent in the currently available version 1.1 of the program, from 1994. The program does include several English and Spanish story templates that function like the subject matter suggestions in The Amazing Writing Machine, but ask no questions to help plan a story. The supplied story starter "Hey, that's my face!" begins with an illustration of two identical girls placed on a shopping mall background. The opening line is provided below the image: "I couldn't believe it. One day at the shopping mall I met a girl who looked exactly like me." Storybook Weaver Deluxe has been highly ranked by several educational publications, earning five stars and straight A's from MultiMedia Schools (Hixson 1996).
Another piece of software to encourage story writing is Once Upon a Time, which allows children to draw a picture and then write a story about it (Compu-Teach 1990). This software comes in three volumes, each with a different theme. Since this program is older, it is not as widely available as the products above. However, it has been used in one study that sought to compare the empirical benefits of a graphical story assistance program to those of text-based software. Learning disabled students used Once Upon a Time for part of the study and used a computerized list of story writing prompts during other times. The study concluded that such approaches may help address particular student needs if tailored to the individual student. This study was small and lacked a control group that used no special writing assistance software. It was not clear that either program resulted in substantial benefits (Bahr, Nelson, and Van Meter 1996).
Many researchers have devoted extensive effort in developing software to aid in writing. Although few writing assistance programs have been developed specifically to offer help to young story writers, many interesting general-purpose writing systems for older students have been designed. Certain other programs employ techniques applicable to the writing process at all grade levels.
Burns developed programs for the US Air Force to aid college writers in planning essays. Although these programs were for a very different age group and type of writing, they address the user directly and interact in a conversational manner (Burns and Culp 1980). These programs date back to 1978 and influenced many other developers of prompting and planning software to assist writers. They were initially created on the VAX 11/780 and ported to some personal computers. Burns based the content of his three programs on different heuristic strategies in rhetoric. The concept grew out of earlier work in computer-assisted prewriting done by Ellen Nold at Stanford University in the early 1970s. Nold's pioneering work, which asked students questions to help then identify a subject, audience, and organization for their essays, was influential to many early developers of prompting software (Wresch 1984).
The most widely discussed of Burns's systems is TOPOI.
In this program, Aristotle's twenty-eight enthymeme topics for generating a persuasive thesis (see his Rhetoric) are reformulated into thirty-seven questions or thought prompts which challenge the student's understanding of his intended thesis in terms of definition, causes and effects, opposites and associations, consequences, and matters of fact and opinion. If the student's initial topic was "computers in education," for example, the computer would prompt, "What is the definition of 'computers in education'?" Or, "Who supports 'computers in education'?" The student's responses produce one of two results: a gradually expanding base of explicitly expressed relationships and perceptions, or a gathering realization that he does not have the information or insight to proceed with his thesis as formulated. The former result provides a transcribed self-interrogation from which a tightly developed thesis may be drawn. The latter informs the student early on that he needs to redirect his attention or research. (Kemp 1987)
TOPOI and Burns's other programs make little attempt to computationally analyze or understand the student's responses. TOPOI does recognize the types of questions the students might be asking themselves by looking for words like "why" or "what." It also notices if the response has fewer than ten words. It attempts no comprehensive assessment of input, however, seeking instead to help improve the students' assessment of their own thinking. In the context of then-popular intelligent tutoring systems, which used artificial intelligence to evaluate students and tailor training toward student needs, this was somewhat unusual. Yet Burns's software proved very effective as brainstorming and planning aids. Although developed to aid college essay writers, these programs employ techniques applicable to elementary school story writers as well (Burns 1984).
Burns's software addressed the user by name and in an informal way. For example, after the user typed that the topic of the essay would be "PROTEST AGAINST MATERIALISM," the program replied "HEY, THAT'S NEAT, WALLY! WE'LL HAVE A GOOD TIME THINKING ABOUT PROTEST AGAINST MATERIALISM." The voice is upbeat but it had few other traits. There is no clear motivation explaining why the voice of the software is offering help or questioning the user. The voice does suggest a person behind it with, for instance, a profession. It does not suggest an age, gender, nation or locality of residence or origin, or other typical identifying traits that a character has. It is unclear whether the voice is that of a peer, or more like that of an instructor. The voice does not reveal any quirks or defects that the simulated speaker might have, the sorts of things that make a character believable. Although TOPOI has a voice, it does not create the impression of a character behind that voice.
Woodruff, Bereiter, and Scardamalia developed computer prompts similar to those created by Burns but for younger children. Like TOPOI, their program provides sets of fixed prompts to help in planning, and does not analyze the text. It offers suggestions if the child does not type for 20 seconds, encouraging the writer to type anything that comes to mind instead of pausing. The program also provides a help menu with information about the structure of essays. A study found that children liked the program and said they found writing easier with it. However, the papers written with the program were not rated any better, and were not any longer, than those composed traditionally (Woodruff, Bereiter, and Scardamalia 1981-1982).
Daiute developed a program at Harvard University called Catch. This software was developed for use by young users, including those in elementary school. Catch extends the spelling, style, and grammar checking concept to providing more writing help. Users can select an option to get comments or questions about what they have written. Some prompts ask things such as "Does this paragraph include details that help the reader see, hear, feel, or smell what you're talking about?" Others suggest that phrases such as "sort of" and "well" might be unnecessary (Daiute 1985a). Children used the program to compose stories in one study, and some revised more after using Catch. The study, which examined how children used Catch and which features they preferred, did not determine whether using Catch resulted in any improvement in children's writing quality (Daiute 1985b). Another four-year study of junior high school students showed that Catch users did improve their revising strategies significantly more than those who did not use the software (Daiute and Kruidenier 1985). Catch exemplifies analytic revising software, a program that examines the text that has been written and offers suggestions based on its algorithmic analysis.
Zellermayer and other researchers developed and tested The Writing Partner, an essay writing prompting program. This software was based on the theory of Bereiter and Scardamalia offered metacognitive prompts in an anonymous voice, such as "Is the reader a novice? Remember that he or she may need some basic facts about the topic." These prompts appeared during the writing phase as well as during planning and revising. Two versions of the system were employed in a study, one providing help without a request from the student and the other offering help only when such assistance was solicited. The found that the unsolicited help software improved students' writing both during and after the study. The improvement was significantly more than that of the control group and the group using the solicited help system (Zellermayer 1991).
A system called Write Environment was developed by Houlette out of experience with his earlier system, The Writer's Plan. This system for older writers seeks to aid the process of writing by providing add-ons to Microsoft Word for Windows. For each type of writing behavior identified by Flower and Hayes (1981), Write Environment allows the writer to summon a window and get assistance. For instance, the writer can open a planning window at the beginning of the writing process, choose a method of planning, and write ideas in response to the selected style of prompts (Houlette 1991). Write Environment does not simulate conversation as TOPOI did. Instead, it simply lists its prompts all together in paragraph form and asks for a reply. It does, however, offer prompts and help for numerous activities that support writing, not just planning. In the current version of the software, there are windows for defining the central idea, organizing thoughts, thinking about the audience, and doing revision (Houlette 1998).
Another system geared toward supporting the different types of writing processes is MAESTRO, by Rowley and Crevoisier of the USAF Armstrong Laboratory. It is based on R-WISE, software to aid the writing process that was tested in 14 high schools over four years (1992-1996). Students who used R-WISE "outperformed control-group students on overall measures of writing quality and analytical reasoning skills, showing performance improvement of between one and two letter-grades." Some instructional components of MAESTRO are compulsory, and students must demonstrate mastery of these before they are allowed to begin writing. Once in the writing workspace, advice statements appear based on the "emergent writing process of the student." One such advice statement explains how to highlight text, but tabs for planning and revising offer higher-level help. MAESTRO is now being tested in middle schools as well as high schools (Rowley and Crevoisier 1997).
Rosebud is one of the only programs for very young writers that uses such a conversational exchange to encourage story writing, with requests for more if the story is brief. This system was developed by Glos at the MIT Media Laboratory in the Gesture and Narrative Language research group. The main point of the system was linking treasured toys to the stories told about those toys, and providing a way for children to record and exchange those stories. Rosebud simulates conversation with a simple request for a story about the toy presented, and a few additional requests after the story has been written. The software by Glos, like that by Burns, addresses the user by name in an informal way. Rosebud's voice also does not give the impression of a character with a motivation or identifying traits such as a profession, age, or gender. As Glos wrote of the program she designed, "I want Rosebud to have a personality, with likes and dislikes and quirks, but that only comes through weakly" (Glos 1997).
An earlier piece of software that had a conversational exchange with children was Sage, by Bers. Sage, also developed in the MIT Media Laboratory's Gesture and Narrative Language research group, is not a program to aid in creative story writing. Yet it does ask children to discuss problems or type short stories of personal experience. It also places the child in a definite framework of interaction, in the role of someone going to ask for advice from a respected religious or cultural figure. The conversation in Sage is with a character - in one case, a bunny assistant to a rabbi - who is human-like. The rabbit character, created by Bers, does have a motivation: to help users deal with problems through cultural lessons in the form of stories, as consistent with its religious beliefs and job. This character also has traits: it likes carrots and hopes that after helping the user, it would be rewarded with one. Children enjoyed interacting with Sage. The main goal of the project was to enable children to build their own characters. Because of this, there has been no study of how important to interaction the well-developed adult-created characters were (Bers 1997).
Interactive technology has indeed been brought to bear on children's story writing. But most commercial software simply adds multimedia elements and story starters to make a usable but limited sort of extended word processor. Of the conversational programs that do exist, no story writing programs focus the writer on a particular imagined audience. The absence of such a focusing character may not bother college writers, but for elementary school students who have trouble with the distanced nature of writing such an omission is a missed opportunity. None of the existing conversational programs put the writer in a strong dramatic situation which suggests a certain role and particular type of response. Again, while experienced computer users may be able to make use of conversational prompts without such a framework, motivating and scripting the user's action is helpful in general and particularly important for young users. The limited forms of conversation employed by current programs involve moving from one item to the next in a small set of responses. They often do not even provide any opportunity for user initiative in determining the order of questions.
Existing software can prompt writers to think about important issues related to story, encourage them to write more, and ask them to revise. Yet the available software does not aid in the transition from conversation to solitary writing by offering intermediate forms of conversational support or an example audience. There is a clear need to provide a fuller framework for conversation, and build on progress made in earlier research. The development of new software can clearly extend the ideas implemented and tested in earlier work, particularly that done in the Gesture and Narrative Language research group. Character-based improvements can provide better interventions to improve children's story writing. The combination of effective existing techniques and a simulated conversational character constitutes a new approach not seen in previous computer writing assistance software.