Walden's Paths provides a mechanism for overlaying linear paths over World-Wide Web material. The path structure does not alter the pre-existing structure of the original material, but provides a means for organizing and contextualizing the collection of material. This paper first reviews the design of Walden's Paths and our experience with it. It then presents a new mechanism that enables the provision of ephemeral "lightweight" paths, which may be generated by a computer process.
Walden's Paths enables the provision of directed paths over the World-Wide Web. In other words, it permits the specification of a meta-structure over the Web--a structure that brings together materials potentially found at widely-separated locations on the Web, forming a single, cohesive unit while at the same time leaving the pages' original structure and content untouched and accessible. Although the function provided by Walden's Paths is somewhat similar to that provided by bookmark lists, the path implementation provides the reader a stronger dynamic sense of context than do static lists. Navigational context is provided through addition of user controls that enable traversal through the path while representing position within the whole. Additional contextualization can be provided by a path's author through the addition of annotations that can be used to help the reader understand the page's material in the context of their own environment and experience.
The usefulness of paths (particularly linear paths) in organizing and contextualizing hypertextual material has been long-recognized [Bush 1945], and stand-alone implementations have existed since the 1980s [Trigg 1988, Zellweger 1988 and 1989]. Even so, implementing paths over the World-Wide Web has required addressing novel issues, particularly issues related to heterogeneous ownership of and decentralized administrative control over Web materials.
Our initial application of Walden's Paths, beginning in the Fall of 1995, has been supporting K-12 classroom use in collecting, contextualizing, and communicating Web materials. We have reported on these experiences in earlier papers [Shipman et al. 1996, Shipman et al. 1997, Furuta et al. 1997, Shipman et al. 1998, Shipman et al. 1999]. Another project with similar educational goals has been described by Nichol [Nichol 1996].
These activities, in support of teacher-student, student-teacher, and student-student communication, require long-lived artifacts. Consequently, the publicly-available version of Walden's Paths supports paths that are temporally persistent--authoring and annotation of paths is supported explicitly by a separate interface; paths, once "published", appear in a centralized site directory; and issues of sharing and of path maintenance are being addressed by our current research activities.
Paths, however, have additional uses beyond that of supporting interpersonal communication. They can be useful to an individual as an organizing vehicle--as a means of arranging discovered resources for subsequent inspection. They can provide navigational assistance that can aid a reader in making sure that all selected resources have been inspected. Such applications require paths that can be reconfigured dynamically--paths that perhaps are generated initially by a computer process and then are manipulated and pruned as the reader inspects the resources contained.
To increase our understanding of these applications, we have extended Walden's Paths to incorporate ephemeral paths--paths that provide a service upon creation but that are expected to be regenerated subsequently, rather than retained. We consider this to be a "lightweight" path mechanism--lightweight not from the standpoint of implementation but from the standpoint of user expectation. Indeed, the implementation costs of lightweight paths are approximately equal to that of the earlier-mentioned persistent paths. However, since they are not catalogued, no specific user-selectable "handle" is provided that encourages their general access. Consequently, such paths are intended to be a personal, rather than public, artifact.
The next section of this paper contains a discussion of the relevant portions of the design, implementation, and use of the publicly-available version of Walden's Paths. Following this, we return our attention to lightweight paths in section 4, presenting examples of their use in section 5. Section 6's discussion and conclusions complete the paper.
Figure 1 shows the Walden's Paths architecture, illustrating the major components (the WWW Browser and the HTTP Server included in the diagram are external to the Walden's Paths system). The two components that are visible to users are the Path Authoring Tool and the Path Server. This paper will focus on modifications to the Path Server, so the other components will not be described here; see our earlier papers or our Web site for details on them.
The Path Server, implemented in C and invoked as a CGI program, takes the path information and produces the frame set displaying the control-flow buttons in the upper-left, the annotation in the upper-right, and the source Web page from the original site in the lower frame (see Figure 2). The control-flow frame provides the path reader with a sense of where they are in relation to the overall path, access to later pages in the path, and a pointer to where the source page comes from so the page can be viewed directly, without the Walden's Paths frames on top, if so desired.
Links in the source page are active and, besides changing the contents of the bottom frame, change the upper-left frame to a "back to path" button so the reader can quickly get back to where he/she left the path. Figure 3 shows an example of this when browsing off of the page. Here, the reader has selected the "FACILITIES" button in the lower window shown in Figure 2.
The Path Server (as well as the other programs that make up the Walden's Paths system) is maintained in versions for Sun Solaris and for Windows NT. The paths displayed by the path server is not browser-specific or platform-specific, although the current version requires the provision of frames. In practice, current versions of Netscape and Internet Explorer are both being used to access paths.
Early versions of Walden's Paths have been used by teachers taking part in summer teaching workshops and by teachers and students in eight high-school art and music classrooms. It is these experiences that have determined subsequent development, including the lightweight path mechanism.
In both workshop and actual classroom situations, one of the most striking outcomes of Walden's Paths use is how quickly the path technology fades into the background. Teachers and students find the path metaphor very natural and, after a short period of learning about the system, quickly focus on the content within paths or technological limitations of the Web in general. This is a positive outcome as it indicates the ability for the technology to slip into the background in the classroom, allowing teachers and students to focus on what they are teaching/learning [Furuta et al. 1997, Shipman et al. 1999]
One observation from our experiences is that while teachers find paths easy to integrate in the classroom, they find it difficult to find time to author paths from scratch. Many of the paths used in the classroom were authored during training sessions rather than during the semester when they are spending most of their time in the classroom. It is this observation that has driven the need for greater authoring support for teachers and the more general notion of lightweight paths.
The persistent paths described in the previous section suggest entry specification, refinement, and annotation by a human author. The investment of time and effort is reflected in an expectation that such paths will be long-lived--that they will be catalogued, reused, shared, and maintained.
Generation of lightweight paths assumes the involvement of a computer process. They are intended to enable the application of Walden's Paths representations to computationally-generated collections of resources--results from queries, subsets or supersets of previously-developed collections, and applications customized for an individual or environment. Since lightweight paths are ephemeral, it seems unwise to support their fine-tuned tailoring or other large-scale investment of human energy in their specification.
Our lightweight path implementation fits into the general architecture of Figure 1. Lightweight paths use the same internal description format as persistent paths, but their entries are flagged as ephemeral and include information about their creation and retention times. To allow the registration of lightweight paths, an additional path insertion interface has been added to the Path Database, replacing the Path Authoring Tool functionality. This interface adds a temporary path to the database, but does not register the name in the path directory (i.e., does not publish the path). The interface then returns a handle to its caller, in the form of a URL, that can be used to access the path. A CGI program creating a lightweight path, can, in the expected case, generate an HTML page containing a redirection to this URL (i.e., the ONLOAD attribute of the BODY element), or it can carry out further processing to use the lightweight path in whatever way it wishes.
In this section, we will examine some scenarios of lightweight paths in use.
A group in our Center is developing a resource for humanities scholars studying the life and works of Miguel de Cervantes Saavendra (1547-1616), the author of Don Quijote de la Mancha. Cervantes, a figure of immense significance in Spanish culture, is credited as the inventor of the modern novel. His works and his use of language receive careful scrutiny by modern-day researchers.
This project maintains a set of Web pages at http://www.csdl.tamu.edu/cervantes/. One of the important components of that site is an on-line, searchable, edition of Cervantes' complete works. A scholar seeking, for example, the uses of the term "fortuna" in the first book of the Don Quijote would pose a query as shown in Figure 4. The traditional result of such a query is a list of responses, as shown as Figure 5. It is likely in this case that the scholar will next examine the use context for each occurrence of the term. This requires selection of an element on the list, examination of the page that is then displayed (the page on which the match has occurred), selection of the browser's back button, and finally selection of the next element on the list. Each new selection requires reorientation within the list, to determine first which element had just been selected and then to choose the next in the list. When the list is long, the selection of each element in turn becomes an error-prone operation, as it is quite easy to re-select an element already visited or to skip an element accidentally.
Figure 6 shows the result of the query, implemented as a lightweight Walden's Path. In this implementation, each search result is shown as a separate stop in the path. In the Figure, we are at the first stop. A brief report summarizing the match is shown as the path annotation and the complete text of the page on which the match occurred is shown as the path stop's content (the matching words are highlighted in both places). Selecting the "next" arrow will take the scholar to the next query match. In addition, a link is provided in the matched page display that permits display of the full text of the Don Quijote edition. Selecting this link results in an off-the-path traversal, as shown in Figure 7. Consequently, return to the path can then be accomplished with a single mouse click on the button in the navigation window; this is the standard Walden's Paths mechanism.
The previous scenario has assumed that all pages selected by the search are relevant to the questioner. What if irrelevant material has been selected, a likely outcome of an automatic selection process, and the researcher wants to pick and choose the items worthy of closer inspection? Path elision permits the selection for display of a subset of the path.
We view lightweight paths as the intermediary representation that permits implementation of a set of small "filters"--relatively simple bits of executable code that accept a path (either lightweight or persistent) as input and produce a modified (lightweight) path as output. One such filter permits a degree of path customization--allowing the selection of pages to retain and those to remove from a previously-generated path. As Figure 8 shows, this allows the modification of the previous search-generated lightweight path. The resulting path, also a lightweight path, would then be browsed with the standard Walden's Paths interface.
Such filters are useful when applied to persistent paths, as well. Figure 9 shows a modified directory listing for the persistent paths stored at a particular site. Application of the elision filter allows the reader to customize the path to be viewed. In this example, the reader might wish, for example, to focus on one of the projects of the several that are reported in the tour of our Center's activities.
We are considering the development of additional, more sophisticated, path generators and path filters for applications that are inconvenient to implement with persistent paths. These applications of lightweight paths include those in which the dynamic nature of the underlying content make persistent paths impractical, those in which individually customized versions of a path are appropriate, and those in which the content of the path may vary depending on the specified or identified characteristics of the reader.
Many forms of information are temporally-transient but persistently structured. The contents of a newspaper change on a continuing basis but the general form of the newspaper remains constant (e.g., the sections into which it is divided, the relative importance of the front and inner pages, the significance of placement of materials on a page). Weather satellite pictures provide a consistent set of information, but visible-light images are only useful for areas of the globe that are in daylight. We envision path generators that are able to locate the current versions of a set of resources, annotate them as appropriate to their context, and arrange them into lightweight paths for presentation.
Persistent paths have been used in creating on-line versions of course examinations (see [Shipman et al. 1998]). We envision applications that use lightweight paths to create customized tests from provided question pools using defined criteria. For example, some examinations for types of Federal licenses draw questions from a publically-available pool, with the distribution of the questions' subject areas specified by regulation. Other criteria might include randomization of order so that no two students had the same exact question sequence.
A natural extension to testing mechanisms would be the use of the testing results to identify a set of tutorial material to aid the student in correcting deficiencies. We envision the presentation of such collections of material as being implemented by lightweight paths.
Customization of materials for students of differing skill levels may also be a possible application of lightweight paths. Here we envision a family of paths each with the same backbone structure, and with each drawing materials from a persistently-defined pool. However, the presentation of the specific topics along the path may be varied based on the individual student's age, grade level, interests, or other factors.
Finally, we envision applications that will benefit from randomly-generated paths. We have observed the use of persistent paths as a virtual "tour guide"--presenting an overview of an organization, perhaps. Randomized pages could be used, for example, to show selections from an art gallery's collection, or to show some tips about use of a software package.
Our experience with Walden's Paths has illustrated to us the strength and utility of applying meta-structures over the World-Wide Web. Our existing implementation provides a relatively simple linear path structure. This has had the desirable property of unobtrusively enabling the provision of guidance and context to readers.
While it appears that meta-structures have positive benefit, it is also clear that authoring meta-structure requires effort and attention. This observation is not surprising, as it is a characteristic of any creative and organizational activity. Nevertheless, it does suggest that the benefits gained from authoring should be in proportion to the effort expended.
In this paper, we have reported on a mechanism to allow application of paths in contexts in which the ephemeral nature of the application mandates against large-scale investment of an author's time. We have demonstrated that heuristic approaches to building meta-structure can be effective for some applications in this class. Our assumption is that the paths created using the lightweight path mechanism also will be ephemeral, and consequently will not be catalogued or retained for reuse. In applications suggesting reuse, therefore, the further assumption is that the resources required to rebuild a lightweight path should be similar to the resources required to locate a path in a persistent collection.
Consequently, we have provided mechanisms at two ends of the persistent-ephemeral spectrum. However, the categorization of many documents and structures ranges between fixed and fluid, influenced particularly by the application. Buckland [Buckland 1997] illustrates this clearly in surveying the differing notions of how to define "document". Levy [Levy 1994] considers similar issues in his discussion of literary hypertextual presentations, and Levy and Marshall extend these thoughts into the Digital Library domain [Levy and Marshall 1995].
We see similar influences in the future directions of our project and expect to extend our model of persistent-ephemeral paths as it develops. We are examining a number of such extensions, already. These include extension of the directory mechanism to reflect categorization of paths along multiple dimensions, one of which may be the required temporal persistence of the path. We also see potential for generalizing our model of the authoring process, using lightweight paths to represent intermediate versions of a path while under development.
This material is based, in part, upon work supported by the National Science Foundation under Grant No. IIS-9812040. It also is supported, in part, by an award from Texas A&M University's Interdisciplinary Research Initiatives Program.
The project's Web pages are located at http://www.csdl.tamu.edu/walden/.