The larger part of this thesis is concerned with the design and implementation of a co-authoring and argumentation tool. In this subsection a model of collaborative writing is presented. Though I prefer the term cooperative to the phrase collaborative, I will use the latter as this is an established expression. The model serves as starting-point for a discussion of the design and implementation of The COOPerator, a co-authoring tool that is being developed at the Infolab at Tilburg University. It is derived from the ideas of Galegher and Kraut [1994], Flower and Hayes [1981], Kraut et. al. [1992], Leland et. al. [1988], McCarthey and McMahon [1992], Miles et. al. [1993], Neuwirth et. al. [1990], Posner and Baecker [1991], Rimmershaw [1992], Saunders [1989], and Sharples et. al. [1993].
In a true co-authoring or co-writing process, the peers collaborate on every task as opposed to the situation in co-publishing, co-responding, and helping scenario's which were all identified by Saunders [1989]. In co-publishing, the individuals produce a collaborative text based on individual texts. Interaction in a co-responding environment takes place only during the revision process and in the case of helping, the writers voluntarily help one another during the writing process. In this thesis the focus is on co-authoring, referred to as peer collaboration by McCarthey and McMahon [1992]. In peer collaboration, the group assignment is a truely joint task, all members contribute to the interaction most of the time (i.e., peer collaboration is high on mutuality), and each of the peers has equal control over the text as well as within the interaction (peer collaboration is high on equality). I have identified these criteria as essential for true cooperative work.
Judged by McGrath's typology of tasks, writing is an intellective group task. It is mostly concerned with solving problems with a correct answer. The phrase 'intellective' highlights the cognitive flavour of the task. Especially in scientific writing, the correct answer is based on the consensus of a set of 'judges' who can reasonably be regarded as experts on the specific issue. Mostly, these problems do not have an obvious and intuitively correct answer, which makes writing a difficult task (McGrath [1984]). This point is stressed by Galegher and Kraut [1994]. Among other issues, they describe collaborative writing as involving negotiation about the meaning of facts and as demanding for consensus as to an appropriate solution. In the upcoming paragraph I will show why writing is such a difficult task.
In providing support for collaborative writing a number of difficulties must be faced. Writing is an open-ended design task. In a writing process there is no fixed goal, nor are there formal transitions between states. Also, writing is under-constrained. There are innumerable possible texts that could fit a writer's goals and possible actions that a writer might take at any stage. Moreover, the writing task is recursive; writing brings new ideas which may lead the writer to revise goals and to embark on a new phase of planning and translation (Sharples et. al. [1993]). According to Kraut et. al. [1992], writing involves substantial equivocality. All these points stress the cognitive and creative aspects of writing. They also make one realise that the writing process is hard to support through computerised tools: work on open-ended problems which are characterised by a high degree of ambiguity, requires interactive, expressive communication (Galegher and Kraut [1994], , Kraut et. al. [1992]).
In collaborative writing, the complexity of the task is magnified many-fold. In a collaborative context, writers must not only express their ideas and attitudes toward the document to other members of the group, they must also share and discuss their thoughts in order for the group to establish a solid common ground, a shared understanding of the task on which they can build new ideas to eventually come up with a solution to the given problem (Sharples et. al. [1993]). Trying to impose a structure on the writing task (e.g. structuring the document by using an outline) in order to reach certain group process gains may influence the creativity of the work in a negative way. According to Miles et. al. [1993], an objection to system-enforced document model structuring is that it is too rigid. It simply does not reflect the way people write together.
The famous model of the cognitive processes involved in a writing task of Flower and Hayes [1981] is a good starting point for constructing a model by which to support collaborative writing. It identifies three main phases in the process of which some were already mentioned briefly:
This process is recursive in nature. For example, editing the ideas and goals in the Reviewing phase may lead to a change in plans and a renewal of the drafts, which may again be followed by phase of reviewing, and so forth.
Posner and Baecker [1991] employ a more detailed model which is, in fact, very similar to the model of Flower and Hayes. They identify six primary writing activities: brainstorming, researching, planning, writing, editing, and reviewing. In the remainder of this section, the former model will be used.
According to Galegher and Kraut [1994] and Kraut et. al. [1992], the identified phases differ not only in characteristic activities, but also in modes of communication. In the first phase, planning, group members make extensive use of face-to-face meetings. Planning is typically a joint activity whereas drafting, the second phase, is solitary. Once this phase is reached, communication between group members is likely to shift from lengthy, highly interactive sessions to brief conversations in which group members exchange progress reports or otherwise coordinate their separate activities. When the end is near, the participants enter an integration and reviewing phase during which they may again require extensive interaction to weave together the results of their individually performed work.
Thus a model arises that reveals the temporal sequence in the communication between group members engaged in a collborative writing effort. It implies a differential impact of computer-mediated communication across time. Especially in the planning and revision phases of the work (the most communicatively intense phases of the project) the negative aspects of computer-mediated communication on task performance are expected to be most dramatic (Galegher and Kraut [1994]).
Apart from these task issues and communication issues, designers of co-authoring tools should pay attention to other specific group-related issues (Sharples et. al. [1993], McCarthey and McMahon [1992]). A general notion, found in many articles (e.g. Cole and Nast-Cole [1992] Neuwirth et. al. [1990], Leland et. al. [1988], and Rimmershaw [1992]), is that of role-support. Roles clearly exist within writing groups. They may be general social roles (e.g. mover, opposer, follower, and bystander (Cole and Nast-Cole [1992] or even devil's advocate, a role that counteracts groupthink (Meijer [1994])), or more task-related, such as author, reviewer, and annotator in a co-writing context. The Quilt system (Leland et. al. [1988]) uses role information to determine which actions the group members can perform.
According to Sharples et. al. [1993] role-mechanisms should be applied with care. They warn for the role inertia that is likely to interfere with experienced writers' excursions into other roles, because these may be better suited to the task at hand. For example, if the system were to embody a strict notion of the reviewer role, a reviewer would not be able to add alternative text. In COOPerator's document facility we have dealt with this problem by allowing reviewers to create alternatives to document parts
To conclude this subsection, I present a simplified step by step model by which co-authoring tools can be judged. It takes into account the issues that were previously mentioned. Because of the many possible ways in which co-writing groups can be organised (e.g. it may be a virtual group, ot the members may be co-located, or the interaction may take place in either a synchronous or asynchronous way), the model does not subscribe how and to what extend one of the focal points should be supported by a certain program. In short then, a co-authoring tool should support:
I do not claim this list to be exhaustive, nor is it very detailed. On a conceptual level though, I think this list can be valuable for evaluating co-writing tools. An assessment of The COOPerator on the basis of this model is incorporated in this thesis.
Sjoerd Michels, Tilburg, The Netherlands