Association and argument: Hypertext in and around the writing process

Abstract

While hypertext is often claimed to be a tool that especially aids associative thinking, intellectual “work” involves more than association. So, questions arise about the usefulness of hypertext tools in the more disciplined aspects of scholarly and argumentative writing. Examining the phases of scholarly writing reveals that different hypertext tools can aid different phases of intellectual work in ways other than associative thinking. Spatial hypertext is relevant at all phases, while page-and-link hypertext is more appropriate to some phases than others.

Notes

In popular discussions, hypertext has come to be associated with the notion of associative thought.

Some people are appalled by the lack of linear logic in hypertext, and the way this makes traditional rhetorical argument very difficult, but is that type of argument a relic of the past? Was it better suited to a society that was characterized by different needs and interests and a different understanding of the world? (Anlitz 1998)

For other examples, see Lepers (1993), Brown (1999), and Gosse et al. (2002).

In his vision of the Xanadu “system for the supply and presentation of material” Nelson proposed “two basic relationships: what we would call the link, which is an unchanging connection between objects, or parts which are different, and the transclusion, which is a maintained connection between parts which are the same” (Whitehead and Nelson 2002). Transclusion places a reference to a source into a text's code, which is displayed as if the transcluded item were part of the document, when in fact it is called from the source. Nelson envisioned this as, among other things, a micropayment scheme for dealing with intellectual property issues. Current page-and-link systems do not support transclusion, though see the gIBIS-descended Compendium tool as a link-mapping example of transclusion at both node and link-map levels.

Stephen Robertson discusses the relevance of associative thought to scholarly method in history:

This website has been designed to present history hypertextually. … Traditionally this is seen as facilitating “associative thinking”. … Therefore, we find many attempts to produce interconnected, nonlinear narratives to facilitate this thought process. It is the suggestion here that the “associative thought” model may not translate well to historical inquiry. It is worth noting that history, to the extent that it can be defined as a discipline at all, must be defined as a way of thinking, as a methodology for analysis. Certainly there is much variety in historical modes of thinking, and its precise terms are almost impossible to define, yet it would appear to be a distinct possibility that “associative thinking” could mark a clear deviation from the strict processes of analytical thought that would appear to constitute historical thought. (Robertson 2002)

Emphasizing how the affordances of tools can change processes, Engelbart asked what would have happened if our ability to record and manipulate symbols had depended on a writing instrument the size and weight of a brick:

The effort in doing calculations and writing down extensive and carefully reasoned argument would dampen individual experimentation with sophisticated new concepts, to lower the rate of learning and the rate of useful output, and perhaps to discourage a good many people from even working at extending understanding. (Engelbart 1962)

The NLS/Augment system was deployed at a variety of government bureaus and corporations, but his full vision, like Nelson's, has been frustrated by corporate thoughtlessness, confusions about “office automation”, and a lack of funding. At the ACM Hypertext Conference 2004, Engelbart protested that hypertext research was still not properly pursuing his ideals for the augmentation of intellectual skills. His complaint seems to me only partly justified, for while there has been no centralized and hierarchically arranged suite of tools such as he envisioned, there has been gradual but continuing development of tools that augment various cognitive tasks for scholars, scientists, and poets. The resulting fragmentation and variety is a key difference between our situation and that envisioned by all three pioneers.

Parallel to the development of general hypertext tools, there have been efforts to develop specialized systems tailored to represent argument structure, such as Aquanet (see Marshall et al. 1991), gIBIS, Compendium, Notes (see Neuwirth et al. 1987), ClaiMaker, Araucaria and DebateMapper. Their theory and history are presented in Buckingham Shum (2003).

Hypertext can signal explicit connections by colored links, spatial relations, or special symbols. Texts also possess implicit connections suggested by the meanings of words and syntactic relations. These multiply beyond the ability of any set of explicit links. Joyce's Ulysses is richer in linear prose than a hypertext version would be, since the hypertext would have to make explicit and overemphasize a selection from the many implicit connections suggested in the text.

Zellweger et al. (1998) demonstrate the purest form of stretchtext, where text flows aside to allow an expansion of a phrase or sentence to appear. Guide, described in Brown (1987), uses buttons that are replaced by new material (see also DeBra et al. 1999, concerning MetaDoc). HTML can perform stretchtext either by replacing one page with a longer one, or by using CSS to hide and reveal portions of a page.

A kind of spatial mapping appeared in Engelbart's NLS/Augment system where text objects could be repositioned on the screen and linked by lines. Mapping was prominent in NoteCards (Halasz 1988) and gIBIS (Conklin and Begeman 1988; the most recent version of gIBIS is available as Compendium). Link mapping was important in Intermedia (Garrett et al. 1986, Kahn et al. 1995, Landow 1997) and became central to Storyspace, whose sibling Tinderbox combines link mapping with spatial hypertext features. There has been active research toward developing tools that can add link mapping to the web. See, for example, Chen (1997), Mukherjea and Hara (1997), and Durand and Kahn (1998), and Toyoda and Kitsuregawa (2001).

See Marshall and Shipman (1997), Shipman and Marshall (1999), and Shipman et al. (2001). The most elaborated spatial hypertext system is VKB (Visual Knowledge Builder); see Shipman et al. (2002).

A spatial hypertext system may also provide for creating links among spatially distant items, but this is not its major virtue. Page-and-link hypertext can approximate the flexibility and tentativeness of spatial hypertext by using and revising multiple types of links. But the result is visually clumsy, and a precise link labeled as “tentative” is not the same as the “maybe it's sort of related maybe not” of a spatial positioning. Spatial hypertext can be added to page-and-link and link mapping hypertext tools. Tinderbox has perhaps the most complete integration: a net of links appears in a manipulable link map, but there are also non-link spatial colorings and areas that can be employed to overlay other kinds of structure and relation onto the link map.

The interplay of text and visual space creates challenges for those working to make hypertext useful for the visually disadvantaged.

Systems aimed at writing often provide multiple different views of the developing linear or hypertxtual documents, such as SEPIA (Structure Elicitation and Processing of Ideas for Authoring; see Streitz et al. 1992), KMS (Knowledge Management System; see Akscyn et al. 1988), Storyspace, Tinderbox, and the ART suite of tools. For a review of writing support systems, see Smith et al. (1987), Sharples (1992), Buckingham Shum and Hammond (1994), and Sharples and van der Geest (1996).

See Marshall and Shipman (1997) for a discussion of information triage.

There are tools such as Devon Agent that try to structure web search engine results. On the Web, see the Web search engine Clusty and the blogging indices that try to keep keywords and linked families in view (Blogdex, Daypop, Technorati, and others).

The spatial parser in VKB helps refine the user's practices, as does the related WARP system, described in Francisco-Revilla and Shipman (2000). Tinderbox agents can perform user specified (but not automatically adaptive) actions that affect both link structure and spatial presentation of a hypertext.

The growth in the number and popularity of simple tools that can hide the more detailed levels of outlines shows that writers value graded access to complex document structure. Outliners provide what in hypertext terms are composite nodes representing whole sections of a document, but hypertext programs also provide richer interrelations than hierarchical outlines.

Amazon.com now provides for some books a list of distinctive terms and phrases automatically extracted from the book's text and listed as “statistically improbable phrases” that are presumed to reflect unique concepts presented in the book, a common approach to document differentiation in information retrieval research. However, one suspects that this approach would not be effective in differentiating scholarly articles within a given field, which make different points while manipulating the same vocabulary as related articles. An approach to adding layers of human-encoded metadata is required at present to extract and formalize argumentation (e.g. Buckingham Shum et al. 2005).

Buckingham Shum (2003) points out that researchers seeking to represent on paper concepts and arguments have from the beginning used box-and-arrow notations.

See Kolb (1997a, b) for examples.

Mancini (2000) and Mancini and Buckingham Shum (2001) explore other visual ways of expressing specific argumentative and rhetorical connections.

“The [links] whose establishment and use within the files [Bush] describes at some length provide a beautiful example of a new capability in symbol structuring that derives from new artifact-process capability, and that provides new ways to develop and portray concept structures” (Engelbart 1962). Engelbart's classic filmed demonstration shows how text objects can be manipulated spatially to express relations and implications. Engelbart seems to use “concept structure” to designate both sets of classifications and sets of propositions; this can conflate two kinds of linguistic entities and their different kinds of relations, but their spatial representations might well be similar.

See, for instance, the program Microcosm, discussed in Davis et al. (1992) and Carr et al. (1996).

It would be useful to study how hypertext tools compare to outliners and to programs for organizing clippings and snippets of text. Some hypertext tools, such as Storyspace and Tinderbox, can function as outliners, but this is a subset of their abilities. Goranson (2005) continues a detailed discussion of the features and uses of outlining programs. There are also clipping organizing programs such as Devon Think, Hog Bay Notebook, NoteLens, and the clipping facilities added to OmniOutliner and Tinderbox. See Neuberg (2005) for a discussion of such tools. There are also programs, such as the editor Ulysses, that manipulate buffers and multiple files, using concepts from Emacs and programming editors but presenting them in ways more clear to creative writers.

See Nakakoji et al. (1998), Yamamoto et al. (1999, 2002), and Yamamoto (2001).

If a writer wants to recall a past state of a document, in order to understand its present state, or to examine alternative developments, in most systems the writer will need to have saved past versions of the file under different names. VKB's “timeline slider” offers an internal, automatic, and elegant method of reviewing the development of a document.

Continuous zooming has been implemented in single-purpose tools from Windsor Interfaces and from the University of Maryland Human–Computer Interaction Lab, and the persistent value of stepwise zooming in link-mapping mode is consistent with its provision in hypertext tools from the early NoteCards and Storyspace systems, through to the more recent VKB, WARP, Compendium, and Tinderbox.

I have discussed elsewhere how hypertext might be of service to argumentative analysis, critique, and writing. See Kolb (1996, 1997a, b, 2000a, 2002). For other discussions, see Carter (2000), Dalgaard (2001), and Buckingham Shum (2003).

“Our approach is to make computer systems ‘invisible’; users must be able to feel that they are interacting with ‘representations’ not with ‘computers’ in using such systems” (Nakakoji et al. 2000, 1). This may be an argument for simpler cooperative single purpose tools. On the other hand:

Both Adobe Photoshop and the Emacs text editor are tools that require significant effort to learn effectively. And yet, those people that do put in that effort often love those tools, sometimes fanatically, and use them in countless ways for years. … Part of the reason … is that once an expert has mastered them, they can focus on the task at hand without interruption. This is due in my opinion to their general efficiency, powerful macro capabilities that allow automation of repetitive tasks, and to the robustness and predictability of the software. (Bederson 2004, 3).