ACM Computing Surveys ??(???), December 1999, Copyright © 1999 by the Association for Computing Machinery, Inc. See the permissions statement below.

Functional Link Typing in Hypertext

Richard W. Kopak
University of Toronto
Faculty of Information Studies    Web:
140 St. George Street
Toronto, Ontario, Canada. L6H 5Y1


University of Toronto
Interactive Media Lab    Web:
Department of Mechanical and Industrial Engineering

ABSTRACT:  Though a fundamental component of hypertext, little research has been aimed at developing the link's ability to express specific qualities of the relationship that it establishes between nodes of information. A proposal is made for the development of a taxonomy of link types that describe the function (discursive quality) that characterize the formation of links. Furthermore, it is argued that the taxonomy should be empirically derived and not established a priori. A number of existing link typologies are reviewed, and it is suggested that there is sufficient breadth in these typologies to provide a test-bed of representative types for use in the empirical investigation.

Categories and Subject Descriptors:

H.5.4 Information Systems, [Information Interfaces and Presentation]: Hypertext/Hypermedia. I.7.2 Computing Methodologies, [Document and Text Processing]: Document Preparation, Hypertext/Hypermedia, Experimentation

General Terms: Design, Documentation, Human Factors, Theory.

Additional Key Words and Phrases: Electronic Publishing, Link Typing, Link Taxonomies.


By any standard definition, the associative model of hypertext is minimally described as a set of nodes containing informational content connected by links that allow traversal between these nodes. Such a rudimentary definition does, of course, belie the complexities involved in creating useful hypertext. The volume of accumulated literature on this topic is a testament to this complexity, and demonstrates the rich potential offered by such a seemingly simple conception.

Much of this literature has viewed the information point (node) as primary, relegating the other fundamental characteristic of hypertext, the link, to the status of some kind of neutral connective. Burbules [Burbules 1998], p.104 notes ironically, that links' "ease of use makes them seem merely shortcuts, and subservient to the important things: the information sources that they make available. Their speed in taking a user from one point to another makes the moment of transition too fleeting to merit reflection; the link-event becomes invisible." Landow [Landow 1997], p. 11, is of much the same opinion, further noting the importance of the link, and stating that it is nothing less than "the element that hypertext adds to writing and reading."

If, as these comments suggest, the link is important but underutilized, the question arises of what latent capacities exist in links that would allow them to be better used? One possible answer to this question involves the ability of the link to indicate or convey information about aspects of the relationship between the nodes in advance of the traversal of the link. In short, a form of link typing. Landow [Landow 1997] suggests this as advantageous and important to hypertext in that link typing can usefully limit a link to a specific kind of relationship (e.g. exemplifies, or contradicts) that indicates connectedness within the structure of the document on semantic or rhetorical grounds. When clearly labeled, link types provide a kind of preview that aids in navigation.

It should be emphasized that it is not the link's ability to describe what will be shown when the link is traversed that is of primary concern here, but what the link can tell us about the relationship between the source and destination nodes that the link bridges. The former falls more properly into the category of node typing than of link typing, and is directly concerned with issues of genre [Haas 1998]. Rather, what the link should be doing is telling us how the content of the destination  node (modifying information) is intended to alter our understanding or interpretation of the source node (the object information). In other words, it should tell us what discursive function the modifying information is serving in relation to the object information [1].

The challenge is to discover what kinds of functional relationships the links might best represent. The author is currently involved in developing an empirically derived taxonomy of functional link types that addresses this question. The form of the research is largely based on the methodological approach used by Chaffin and Hermann [Chaffin 1987], [Chaffin 1984] in the testing of their "Relation Element" theory. Like Chaffin and Hermann, the taxonomy is grounded on the ability of research subjects to group relations on the basis of similarity. Unlike Chaffin and Hermann, the relations being investigated are not those between words at the semantic level, but between "relationships" at the functional level [2]. The goal of the research is to provide a reasonably equipped ontological representation of this relationship set.

Existing Link Typologies and Approaches

An empirical approach of this sort requires a minimal set of link types to begin with. Though the published literature proposing specific link typologies of the sort being considered here is not extensive, it does offer some variety in regard to the kinds of relationships that should be represented. Trigg's dissertation [Trigg 1983] is one of the earliest reports of research using any form of typed links. Although the larger concern of the dissertation was in providing a system for distributed access to scientific information, and the tools required to accomplish this, Trigg realized the significance of link typing in its ability to "make explicit the relationship between two nodes."

Trigg lists seventy-five link types that he developed for his TEXTNET system. At the highest level, these link types are divided into two main categories; normal and commentary links. Of greater interest here are his normal links which served to connect nodes comprising the main flow of information in a work. These normal links are largely rhetorical ones and are used to connect nodes on the basis of argument structure or discourse. Furthermore, since "the goal of a work is to communicate information and/or beliefs to the reader," the specific link types developed were aimed at supporting what were viewed as the major functions of a work: specifying context, posing problems, declaring theories, sustaining arguments, and reporting data.

Baron [Baron 1994], [Baron 1996] identifies two general types of links in her study of the use of a hypertext manual. Organizational links are used to describe the surface structure of documents and comprise elements of the syntactic macro-structure which organizes the presentation of the information. Examples of organizational links range from tables of contents to lower level directional cues such as previous or next page.

The second general link type is content-based and deals more directly with specific relationships between nodes within a text. Of this kind of link, three types are further specified: semantic links, rhetorical links, and pragmatic links. Semantic links describe the relationship or association between words or concepts. Based on the work of Chaffin and Hermann [Chaffin 1984], [Chaffin 1987], Baron proposes three semantic link types to describe the relationships between concepts: similar, contrast, and part/kind of. Rhetorical links are ordinarily used by an author "with the intent of leading a reader through a series of information elements to achieve a learning goal supporting the task" [Baron 1994], p. 39, and include types such as definition, illustration, and summary. Lastly, pragmatic links serve to define relationships that are concerned with practical results (e.g. a warning).

Parunak [Parunak 1991], as do both Trigg and Baron, presents an a priori link typology, but organizes informational relationships according to the characteristic discourse grammar of text. Accordingly, he distinguishes three classes of link types that he states "are useful in hypermedia": association links, aggregation links, and revision links. Association links are the most common of the three classes, and are well suited to illustrate the relational aspect of link usage. Of the association links there are those that relate a single word or phrase to a larger proposition (word-proposition links), and those that relate propositions to each other (proposition-proposition links). In word-proposition links the proposition component seeks to further specify the meaning ordinarily attributed to the word component (e.g. an "Identification" link which defines or otherwise restricts the meaning of a word or phrase). Much more common are proposition-proposition links which describe relationships between nodes containing more complex content. Of the proposition-proposition links four general types of relationships are described, each containing more specific exemplars: orientation links that define the environment for a state or events described in a proposition (e.g. location, circumstance); implication links that describe the logical connection between two propositions (e.g. causation, purpose, warning); paraphrase links that join propositions together that contain similar information (e.g. summary, abstraction), and; illustration links that connect dissimilar propositions that clarify one another (e.g. comparison, contrast).

Cleary and Bareiss [Cleary 1996] use a set of eight "conversational associative categories" that describe link types based on a simple theory of conversation "which argues that at any point in a conversation, there are only a few general categories of follow-up statements that constitute a natural continuation rather than a topic shift." The goal of the link types employed is to provide a structured means of indicating the relationship between nodes that allows users to orient themselves on a local, associative level rather than relying on an explicit hierarchy. The eight categories are represented as binary alternatives under four broader classes: Refocusing Context/Specifics; Comparison Analogies/Alternatives; Causality Causes/Results, and; Advice Opportunities/Warnings.

The Need for a Comprehensive Taxonomy of Functional Link Types

We began by asking the question of what latent capacities exist in hypertext links that might allow them to be better utilized. It has been suggested here that the functional relationship implied in the linking of two nodes of information offers one possible solution. This is especially so in those cases where hypertext is intended as more than a collection of simple associations between somewhat ambiguously related pieces of information. Links, instead, are formed over a distributed information space with the intention of establishing a coherent line of thought or point of view. The author is provided a means of communicating intent, and the user a means of making a more informed decision concerning the efficacy of a link.

The brief review of several existing link typologies (i.e. Trigg, Parunak, and Baron) provides some evidence of agreement that this approach is not without merit. Furthermore, there is a certain level of consistency in the general classes of link types proposed. For example, there are link types that establish context, that sustain arguments, that attribute causality, and that provide detail. A more explicit comparison of the specific link types proposed under each of these more general type classes would reveal even greater similarities.

It seems clear that to carry forward the research agenda efforts must be made to develop a comprehensive taxonomy of link types that has empirical validation. For the most part, the collections of functional link types presented above were derived a priori. One approach to providing such a taxonomy has been to examine and categorize links found in actual use. Haas and Grams [1998], for example, performed a content analysis of existing Web documents to determine what kinds of link types are typically used. The drawback to this particular approach is the relatively restricted repertoire of link types that most individuals bring to the authoring of their Web pages. Furthermore, this approach does not discriminate between what has been described here as functional link types, and those that offer only physical navigation cues, or those that simply provide some ad hoc indication of the contents of destination nodes.

The second approach, and the one the author is currently undertaking, is to transform the elements of existing typologies into a comprehensive taxonomy. The immediate goal is establish a core set of types that are valid representations of the functional relationships typically employed in discursive text. This  includes a specification of the relations between the types in the taxonomy, and the development of a reliable labeling scheme to be used in any future implementation [3]. It is hoped that the taxonomy resulting from this research will provide, to both authors and readers, a means to enhance the innate, but largely under-used, communicative capabilities of hypertext.


[1] In a similar way, Kolb [Kolb 1997] refers to the expression of these functional relationships as "discursive moves".

[2] Relation, as used by Chaffin and Hermann, is understood in the more typical sense of the term as the meaning connection between two words or concepts, and involves the identification of such characteristics as synonymity, class inclusion, case relations, and part-whole distinctions. Relationship, as used here, is perhaps atypical and more concerned with issues of demonstrated relevance, and rhetorical significance. In an analogous use of Guarino's [Guarino 1998] terminology, the universals ("properties and relations"), have become the particulars.

[3]An unambiguous set of link type labels is essential to future implementation and testing in an actual hypertext.  Currently, plans are to instantiate a form of the typing scheme in a corpus employing XML.


[Baron 1994] Lisa Baron. The Effectiveness of Labelled, Typed Links as Cues in Hypertext Systems. Unpublished doctoral dissertation. The University of Western Ontario, 1994.

[Baron 1996] Lisa Baron, Jean Tague-Sutcliffe, Mark T. Kinnucan. " Labeled, Typed Links as Cues when Reading Hypertext Documents" in Journal of the American Association for Information Science, 47 (12), 896-908, 1996.

[Burbules 1998] Nicholas C. Burbules. "Rhetorics of the Web: Hyperreading and critical literacy" in Page to Screen: Taking Literacy into the Electronic Era, I. Snyder (editor), Routledge, London, 1998.

[Chaffin 1984] R. Chaffin and D. J. Hermann. "The Similarity and Diversity of Semantic Relations" in Memory and Cognition, 12 (2), 134-141, 1984.

[Chaffin 1987] R. Chaffin and D. J. Hermann. "Relation Element Theory: A New Account of the Representation and Process of Semantic Relations" in Memory and Learning, D.S. Gorfein and R.R. Hoffman (editors). Lawrence Erlbaum, Hillsdale, NJ, 1987.

[Cleary 1996] Chip Cleary and Ray Bareiss. "Practical methods for automatically generating typed links" in Proceedings of ACM Hypertext '96, Washington DC, March 1996.

[Guarino 1998] Nicola Guarino. "Some Ontological Principles for Designing Upper Level Lexical Resources" in Proceedings of the First International Conference on Lexical Resources and Evaluation, [Online:], 1998.

[Haas 1998] Stephanie W. Haas and Erika S. Grams. "A Link Taxonomy for Web pages" in Proceedings of the American Society for Information Science (ASIS) '98, 485-495, 1998.

[Kolb 1997] David Kolb. "Scholarly Hypertext: Self-Represented Complexity" in Proceedings of ACM Hypertext '97, Southampton, UK, 29-37, September 1997.

[Landow 1997] George P. Landow. Hypertext 2.0: The convergence of contemporary critical theory and technology. Baltimore: Johns Hopkins University Press, 1997.

[Parunak 1991a] H. van Dyke Parunak. "Ordering the information graph" in Hypertext/Hypermedia Handbook, Emily Berk and Joseph Devlin (editors), 299-325 New York: McGraw-Hill Publishing Co., 1991.

[Trigg 1983] Randall H. Trigg. A Networked Approach to Text Handling for the Online Scientific Community, Ph. D. thesis, University of Maryland, [Online: chapter 4 of the thesis -], 1983.

Permission to make digital or hard copies of part or all of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. Copyrights for components of this work owned by others than ACM must be honored. Abstracting with credit is permitted. To copy otherwise, to republish, to post on servers, or to redistribute to lists, requires prior specific permission and/or a fee. Request permissions from Publications Dept, ACM Inc., fax +1 (212) 869-0481, or