A Pragmatics of Links

Susana Pajares Tosca
Departamento de Filologia III, Facultad Ciencias de la Información,
Universidad Complutense de Madrid, Avda. Complutense s/ n, 28040 Madrid, Spain
Email: spajares@eucmos.sim.ucm.es

This paper was first presented at ACM Hypertext 2000 in June in San Antonio, TX, where it won the Ted Nelson Newcomer Award, an award cosponsored by JoDI. We are pleased to be able to reproduce the paper here. ACM notice

Abstract

This paper applies the linguistic theory of relevance to the study of the way links work, insisting on the lyrical quality of the link-interpreting activity. It is argued that such a pragmatic approach can help us understand hypertext readers` behavior, and thus be useful for authors and tool-builders alike.

1 Introduction

I want to start by taking up the challenge that I posed a year ago in "The Lyrical Quality of Links" (23), a short paper arguing that hypertext might be a lyric¹ rather than a narrative form. It proposed the "close examination of explicit links" as the starting point for a study of hyperfiction rhetoric. This claim was based on a comparison of links and poetry, the conclusion being that "the ability of hyperfiction to combine mental jumps with physical ones conveys a new poetical dimension that should be studied in relation to choice as another way of exploring context". Not very sure of how to look at these movements of meaning, I suggested examining the relationship between links and the classical figures of speech.

Curiously enough, this is exactly what Wendy Morgan did about the same time in her paper for the Writers Workshop at HT´99: "Heterotopics, towards a grammar of hyperlinks". (22) She offered two possible categories of links based on two different linguistic approaches. The first (called "Conjunctive functionality") was that of systemic functional grammar, accounting for links "in terms of their cohesiveness", an ultimately unsatisfactory choice unable to accommodate the linguistic "incoherence" of hypertext.

The second (" Disjunctive dys/ functionality") was based on classical rhetorical analysis, and she gave a list of figures of speech (anacoluthon, aposiopesis, catachresis, etc.) without actual linking examples. As Morgan herself acknowledges quoting Jean Clément, her description is based in the assumption that normal communication is essentially different from literary communication, here comparable to hypertext: "hypertexts produce - at the level of narrative syntax - the same 'upheaval' as poems produce at the level of phrasal syntax". And this is probably why she jokingly proclaims her model to be "a perverse, structuralist attempt to tidy into categorical boxes what of its nature evades such neatness." (22) The dys/ functionality approach would require a previous "functionality" study that told us how a text is supposed to be working properly, so as to identify the incoherences - she says: "as interruptions (links) need a sense of structure to work against". In order to recognize the abnormal, there has to be a norm, and I don't think that such a study is possible. Pragmatics has proved that everyday language uses the same mechanisms as the so-called literary language, often with the same cognitive results, see Pratt (27).

As Morgan suspects, applying a set of a posteriori categories doesn't really tell us how links work, how people choose them or what to expect when writing a hypertext. Equating hypertext with a presumed abnormality of literary discourse (its ability to accommodate incoherences) takes it to the realms of deviation, and very much resembles a certain formalist position: the "literariness" of a text is in the text itself, coded under the guise of figures of speech and other rhetorical devices. Literary criticism's mission would be to find that code, to interpret it and to classify it. But literary criticism has long ago done away with such claims for immanence², see for example Fish (10) or Eagleton (8); and even post-estructuralist rhetoricians (1) recognize that deciding if a text is literary depends on factors external to the text in question. This doesn't mean that figures of speech are all of a sudden useless. Quite the contrary, current linguistic approaches find the same kind of semantic "play" in literature as in normal speech, see again Pratt (27), the difference being more a question of degree or attitude than a real separation between the two kinds of discourse.

One such approach is pragmatics, a linguistic discipline that has flourished enormously in the last twenty years - for a complete introduction see Levinson (20). There are different "schools" within this approach: Austin's early speech-act theory (2), Grice's conversational maxims (12), etc., although all could be said to share an interest in the situation in which speech is produced, going beyond both syntantic and semantic analysis to include the context as an essential part of the communication process. I think the most useful of all the points of view within pragmatics is the relevance theory, developed by Dan Sperber and Deirdre Wilson in the 1980s, because it accounts for phenomena that the other approaches are unable to explain, (see 30 for their objections to the code model of verbal communication, the gricean maxims and other reduced inferential models of communication unable to explain phenomena such as irony, seeing them as deviations from the rules).

As I will argue using relevance theory by way of critical tool, hypertext can be said to be lyrical because it exploits certain cognitive conventions that we use to interpret the texts we call literary, specially poetry, and not because it is written using a particular kind of language full of figures of speech. Moreover, this model serves equally to explain how links work in hyperfiction and in non-fictional hypertexts.

2 Linguistics and hypertext

Linguistic approaches to hypertext have usually been formal, concerned with finding a "hypertextual grammar" in order to facilitate the automatic generation of links in digitized text. A good synthesis of this point of view is the book Techniques avancées pour l'hypertexte (3), by Jean-Pierre Balpe et al., of the University Paris VIII. This tendency is also well represented in the hypertext conferences organized by this University, see for example the last H²PTM conference proceedings (4). The ACM Hypertext conferences also have examples of formal linguistic approaches, like the paper "Structural Properties of Hypertext" by Seongbin Park (24). More recently, Kaindl discusses the automatic generation of glossary links and examines the previous work on automatic link generation presented at ACM conferences in "Semiautomatic Generation of Glossary Links: A Practical Solution" (14) , and Mehler analyzes text-cohesion from a formalist point of view in "Aspects of Text Semantics in Hypertext" (21).

Other authors have concerned themselves with the restructuring of the idea of text motivated by hypertext. Renear, Mylonas and Durand have investigated the structural nature of text in "Refining our Notion of What Text Really Is: The Problem of Overlapping Hierarchies" (28), for a better understanding of text processing and text encoding. George Landow, (18) Jay David Bolter (6) and other early theoreticians of hypertext have addressed this problem, seemingly more interested in the nature of linguistic discourse in general, attempting a sort of "textual ontology" with their discussions about authority, linearity, hierarchies, etc. Their reference authors are Roland Barthes and Jacques Derrida, and they try to integrate hypertext in the postmodern philosophical tendency to problematize the status of all linguistic discourse.

Apart from these tendencies, linguistics hasn't had a great impact on the attempts to build a rhetoric of hypertext, which strikes me as strange in a field so concerned with communication between author, (hyper) text and reader.

3 Relevance theory and literature

Dan Sperber and Deirdre Wilson explain the principles that sustain relevance theory in the second edition of their classic Relevance: Communication and Cognition:

1. (Cognitive) "Human cognition tends to be geared to the maximisation of relevance."
2. (Communicative) "Every act of ostensive communication communicates a presumption of its own optimal relevance." (30)

The second principle is the one directly behind the more concrete applications of the theory to particular examples. According to the theory, people speak because they think (or want the addressee to think) that what they are saying is relevant; that is, that the information they are communicating helps the recipient understand a part of the world better. Sperber and Wilson call this useful information, "cognitive effects". Let's suppose that IÃ'´m going out with a friend and she asks me if we will go dancing. I answer: "Linda is coming with us". Literally, I haven't answered her question, so she has to interpret it in a process that could go like this:

• Linda is an elderly Philosophy Professor that only enjoys classical music.
• Linda wouldn't want to go to a disco.
• We won't go dancing tonight.

In fact, my sentence has given her more useful information than if I only had said, "no", because then she wouldn't know why I refused. Now she can think of something else to do that evening, knowing what would be acceptable and what wouldn't. Speakers will always try to ensure that their utterances have the maximum cognitive effects. This is not limited to new information, to information directly related to the conversation topic or to what the speaker knows to be the hearer's interest; all information that significantly changes the hearer's "set of assumptions" (what some authors call "encyclopedia") is relevant³.

Another important point when considering relevance is the processing effort required to interpret an utterance. All other things being equal, the smaller the processing effort, the greater the relevance. If somebody asks me how much I earn a month and I answer: "195,600 cents", it might be true, but it would be much more relevant to say "2,000 dollars". It's easier to understand and it gives a clearer idea of what the hearer wants to know. So the processing effort would determine the degree of relevance of an utterance, together with its cognitive effects.

It is obvious that people strive to achieve optimal relevance in their everyday exchanges, (unless they want to annoy somebody with pointless utterances), and it is also clear that most exchanges succeed at this. The speaker tries to achieve the greatest cognitive effects with the smallest processing effort possible, whereas the recipient interprets an utterance choosing the context that maximizes its cognitive effects, and stops searching for implicatures (the author's term for this particular kind of implication) when a reasonable level of relevance is achieved. The implicatures of a given sentence can range between the strong and the very weak, and it is the hearer's decision when to stop interpreting. Let's see an example taken from Seiji Uchida (32) (stages numbers 5 and 6 are mine). If Peter asks Mary: "would you drive a Mercedes?", and she answers: "I wouldn't drive any expensive car", Peter's search for implicatures could go like this:

1. A Mercedes is an expensive car.
2. Mary wouldn't drive a Mercedes.
3. A Rolls Royce is an expensive car.
4. Mary wouldn't drive a Rolls Royce.
5. Mary doesn't like expensive things.
6. Mary won't accept the diamond ring I've bought her.

While implicatures number 1 and 2 are very strong, and numbers 3 and 4 are more or less unnecessary elaborations, number 5 is a very weak implicature, and number 6 is probably sheer paranoia. A sensible hearer would stop inferring at number 2, unless there was a reason to do otherwise.

For example, if we are talking about Linda coming out tonight and I say: "You know, there is a new critical edition of Sophocles' Electra", my sentence appears not to have any relationship to our conversation whatsoever, even though it might be true and new to the hearer, so that she could be expected to think: "this has nothing to do with the topic", and stop inferring. But what actually happens in every day conversation is that addressees always try to look for a context in which the information received is relevant. She might start wondering why I changed the topic so abruptly, because if I have said it, it must be relevant... Maybe it's Linda's birthday and I'm suggesting a possible present, maybe Electra is being staged at some theatre in town and I want to see it... Other communication theories maintain that context is pre-given and relevance is assessed after it, (in this case my sentence would be irrelevant because it has no relationship to the conversation/ external context), but relevance theory maintains that "it is relevance which is treated as given, and context which is treated as a variable" (30).

The theory is much more flexible than others that would dismiss completely normal exchanges as irrelevant (or incoherent) and wouldn't be able to account for phenomena such as irony. Consider this situation: it's raining ferociously, the water flows along the street and the people clutch their coats trembling in the cold wind; my friend says: "I love having to cross the city in the delightful English spring mornings". Has she lost her mind? Truth-based and coherence-based approaches would probably confirm that she has. Why else would she lie purposefully? But if we assume that her utterance is relevant and process it searching for an appropriate context, the cognitive effects are very rich: not only do we understand that she finds the weather dreadful, but also that she feels like expressing her sense of humour that way. We get an additional "emotional information" about her, something that a sentence like "Terrible weather!" wouldn't give us. But some linguistic approaches have to talk about literary language being a deviation simply because it doesn't fit into their schemas.

Adrian Pilkington elaborates a literary example of Sperber and Wilson to explain the difference between relevance and classical rhetoric approaches. Consider these two sentences:

a. My childhood days are gone.

b. My childhood days are gone, gone.

They express the same proposition, but the stylistic difference is clear. Classical rhetoric would talk about epizeuxis, "a figure used for emphasis". But repetition is not per se emphatic. As Sperber and Wilson show, it can have many other effects (like annoying the recipient). The mere fact of a word being repeated doesn't explain the stylistic effect. "In this case the repetition serves as an encouragement to the addressee to explore further the encyclopedic entries attached to the relevant concepts. The principle of relevance predicts that the extra effort required in doing so will be rewarded by extra contextual effects." The "extra effect" is the access to a wide range of weak implicatures like: my days of innocence are gone, my days of ingenousness are gone, my carefree days are gone, etc. (25) As Sperber and Wilson argue, other figures of speech can also be succesfully explained through relevance theory, notably metaphor and metonymy. (30) We don't need to know a special code to understand literature: it's only that we don't look for the same kind of context reading a business letter as we do when reading a poem. The key here is that there are situations when we are interested in exploring a wide range of implicatures, even if they are weak (for example, literature), and others in which we are not and simply want to disambiguate a sentence (for example, a letter from the bank). We have different objectives and the texts have very different social functions, so we adapt our interpretive strategy to the case. In a rough way of speaking, we stop inferring much earlier with the business letter.

This lengthy explanation is however a very brief summary of relevance theory and its relationship to literature. It was necessary to understand how I use it in regard to hypertext, as we'll see.

4 The relevance of links

Allow me to take the following simplified position for my analysis: hypertexts are communication. Although the circumstances of oral communication differ greatly from those of written (= mediated) communication - there is no "current context" to which to refer to interpret the sentences and there is no possibility of asking the author for feedback - I think it's useful to simplify and say that there is a dialogue between the (hyper) text and the reader. Firstly, because the author builds her hypertext to say something, and secondly, because the reader is active, not only interpreting, but also choosing links and moving through the text. In fact, we could even understand the choosing of a particular link as a response: "I will go this way", an action that we cannot take in print texts. But how does this dialogue work?

Let me consider a hypertext node and its links as the "speaker's utterance", that the reader has to interpret. Apart from the usual presumption of relevance common to all texts, I would add a new relevance principle:

• Every link communicates a presumption of its own optimal relevance⁴.

That is, if a word (or picture) is hightlighted, the reader has to understand that it points to a relevant development of the text. Links don't interrupt the flow of meaning; on the contrary, they enliven it. In John Tolva's words, links open "a space for the reader's mind to construct the extra dimension needed to rationalize the act of 'traveling' a link" (31). In my approach, the link itself would have a sort of "suspended meaning" that wouldn't be confirmed until we have seen where it takes us. In this sense, it can be said that the link has no fixed definite meaning; it is a mere indicator: "there is meaning here: explore the context". This doesn't mean that links are only "meaningful retrospectively", as Morgan proposed (coherently with her rhetoric approach (22)). From the point of view of pragmatics, links force us to make meaning before and after travelling them.

The question of the content of links has already been raised by Jim Rosenberg and, as he quotes, by Kathryn Cramer (29). Although he doesn't give a definite answer to what's inside a link, he seems to imply that the link's meaning constitutes itself in the next level of the reader's activity (following a link is an acteme, multiple actemes are combined into an episode, where we can already distinguish meaning). For Rosenberg, "an episode is simply whatever group of actemes cohere in the reader's mind as a tangible entity" (29). Rosenberg's framework is extremely useful, because it describes the hypertext reader's activity as a search for structure and meaning, as a gathering activity, in line with what Licia Calvi would describe a bit later (7). But like Mark Bernstein's much quoted "Patterns of Hypertext" (5), Rosenberg's description of hypertext activity concerns itself with the perception of meaning at a higher level (typically, a group of nodes). While this is vital to perceive and describe the structure of a hypertext - the understanding of a hypertext's structure being a form of closure - I think there is a lower level of activity: a cognitive closure that happens at the level of individual links. If Rosenberg's and Bernstein's closure could be called "narrative", as the perception of a relatively ample "plot", my link-closure would be lyrical, as it works at a basic level (a word, a group of works) and behaves in a way comparable to what Pilkington described for poetry (26). Both closures are of course complementary.

Let us consider the following text-space in the hypertext "A Dream with Demons", by Edward Falco (9). The node is called: "Joined":

Falco's work is composed of a linear novel and a hypertext at the end of each chapter. Both are written by the same imaginary author: Preston Morris. The novel is a ficticious story, while the hypertext explores his own life, thoughts and fears, that end up by being closely related to the linear novel's themes. This node is from the hypertext, and (not being the first), the reader already knows that all the lines in this hypertext (prose visually organized as poetry), are links to other nodes, as she can check by pressing the Ctrl key and making linked words appear surrounded by a red line, (a feature of Storyspace the author has decided to exploit to make his links explicit). As we are reading about Morris' sad life, his experiences with women, his loss of love and his solitude, we come across "joined". Because of the briefness of the text space and the literary nature of this hypertext, it is safe to assume that the reader reads the whole space before deciding to click on any of the sentences, and then chooses the most evocative to her mind. This means that an extra level of inference making is added to the process of reading normal poetry. If, of the four lines, the word/ link "Away" draws my attention, my search for implicatures (my search for relevance within the context of what I know to be this particular work) might go like this:

1. The love's gone away.
2. Away from this sadness.
3. Away from my sad life.

This context I've created (meaningful in relation to the rest of the work) would make the link highly relevant and worth following. It is now evocative enough for me to click it, maybe more so than the others. (Of course, other readers might not agree with these implicatures or might prefer other lines, but my point is that the process they go through would be the same). If I click on "Away", I go to a node called "I think", where I read:

"I think, maybe, if I can get inside the telling of the story If I can get down inside the foundation If I can break the conventions If I can break the patterns of my storytelling I can break the conventions and patterns of my life"

It's not very far from the implicatures that I imagined the word "away" could yield, but this text adds new meaning to my travelling the link, new cognitive effects that force me to readjust the set of implicatures:

1. Away with the sad stories I tell
2. Away from whatever forces me to do it
3. Away from my sad life
4. I am not happy repeating the same pattern over and over
5. My life and my art are very closely related⁵

In a normal line from a poem (like the sentence quoted above "my childhood days are gone, gone"), we only go through the process of producing implicatures (of interpreting) once. In a hypertext we do this twice: the first time to evaluate our choice, according relevance to the link in a sort of expansive movement of meaning, and the second to contrast our implicatures with the actual text in an opposite movement.

Then we would have the interpretative process common to all texts, that can also be pictured as "centrifugal", since the implicatures expand the context searching for the maximum relevance. It could be argued that hypertext provides the reader with an extra level of poetic richness.

This process is also traceable in non-fictional hypertexts. For example, Nancy Kaplan's hypertextual essay, "E-Literacies" (15) includes, among others, the following paragraph in the node called: "E-Locution":

"A significant feature of hypertext environments is their capacity for inclusion: they want to construct vast and necessarily unfinished collages of documents to represent the knowledge (and the agon) of a discipline."

The underlined words are a link to another node. If we have read this screen complete before deciding which link to activate (there are more than this), we know that the essay is about electronic literacy "its social origins and effects". So my first expansion of context regarding this link could go:

1. An agon is a fight in the Greek sense
2. It's about a disagreement in the discipline of electronic literacy, different positions about something
3. There are academics that celebrate electronic literacy and others that revile it
4. It's going to talk about the techno-enthusiasts and those who prophesy the death of the book

Admittedly, I'm a reader who knows something about this question, so that the last implicature goes a bit too far. Another reader might stop at the first, or not even that far, because "agon" is not a word every English speaker knows. The point is that once we have given enough relevance to that link (we imagine what we are going to find and know if it will interest us), we decide to follow it or not. In this case, this takes us to a node called "Speaking for oneself", that talks exactly about the disputes in that discipline that I had anticipated before. (In case the reader wonders how I've done these experiments, I've read hypertexts while taking notes on paper of the implicatures I extracted from individual links. I have then compared them to the new context brought about by the travelling of the link). Reading it, we widen our understanding of the link:

1. An agon implies a continous struggle, and the two positions seem to be competing for public interest
2. Technology is not innocent, and neither are academic positionings about it. The struggle is also socio-economical

The same operation is performed when we face an index with links, such as Landow's Victorian Web entry page, with its list of links to spaces like "Victorianism", "Gender Matters", "Social Context", "Political Context", etc. (19) This kind of linking requires much less effort from the reader. Her search for relevant implicatures is very quick. The knowledge that this is an academic hypertext with information about a literary subject makes it very easy to do the cognitive jump. Consider the implicatures of: "Gender Matters". The link is the headline or title of a section that is going to speak exactly about that, so that the processing effort is minimum. This hypertext is interesting because of the presence of the navigation buttons (a very common feature of the Web). Navigation buttons are links whose destination we already know, and consequently we don't have to process them as thoroughly as the others. The Victorian Web makes uses of these devices to keep its readers oriented, so that it can then sometimes make jumps that go beyond the "headline-link" so common to many academic hypertexts (together with the "definition-link", or the "sources-link"). For example, in the sentence: "encompassed major epidemics of influenza, typhus, typhoid and cholera", clicking on "typhoid" won't take us to a definition of the word or a scientific article about it, but to another text where it is also mentioned in another context, forcing the reader to make new associations between different aspects of the Victorian era.

The cognitive process is the same, but it seems that in hyperfiction (as in print fiction), the processing effort is purposefully increased, and the reward is the exploration of evocative contexts through the search for a wide array of implicatures. This doesn't mean that things couldn't change. As I've said above, I agree with the theorists that think that the decision about what is literature is not inherent to the texts themselves, but it's an external, social decision. Actually, the hypertextual form facilitates this sort of writing inmensely because of the double inference process that links provoke. Hence, its lyrical quality. There are non-fictional hypertexts that can be read this way; not many, it's true: the weight of the old habits is still too heavy. David Kolb's Socrates in the Labyrinth (17) and Diane Greco's Cyborg: Engineering the Body Electric (11), are both exquisite examples of non-fictional hypertexts that allow for a poetically pragmatic reading. Diane Greco, who merges fiction and non-fiction in her quotations and her argument, combines a "classical" hypertext organization (there is an index of themes one gets sent back to once in a while) with associative movement of meaning that calls for a wide exploration of context. For example, her node "splitting" (I've underlined the links):

"Reconstructing the self as cybernetic requires a reconsideration of the boundaries between the body and the machine technologies that invade, colonize, deform, or enhance it. Why?"

After reading it, I am particularly attracted by the first link "reconstructing the self", because my range of implicatures (knowing the cyberpunk/ gender studies general context) is:

1. Changing the human body into a cyborg
2. Enhanced abilities
3. Contempt for the flesh
4. New ethics

I click it, and arrive at a node called: "systems lingua franca", that reads:

"The language one chooses to describe and define experience is implicated as well; the experience of the self as an interactive system communicating between what was formerly known as the body, the mind, and the other (the machine) requires new metaphors."

I find a new relevance to the relationship between the link and this node, and I rearrange my implicatures so:

1. We construct ourselves through discourse
2. The relationship between us and the world has changed
3. Reconstruction is more mental than physical

It was a very pleasant trip, and this also happens with the other links in this hypertext. The reader feels closure, not when her expectations are exactly met, but when her search for relevance is rewarded with valuable contextual effects, as in this case.

I realize that the search for implicatures at the link level that I have been describing is only possible when the links are explicit, because if they aren't, the reader wonÃ'´t be able to imagine what their relationship with the next text space might be. Although it could be argued that even if the reader doesn't know what words are links, like in afternoon (13), she clicks the words that she suspects will yield more satisfying implicatures, so that she goes through the process even if her choices have no actual relationship to the results. In this sense, hypertexts without explicit links go against hypertexts' conventions, but this may be too twisted. Let me instead take a practical point of view and say that most hypertexts have explicit links, especially Web hypertexts, as there is no way to get rid of the arrow changing shape to a pointed hand every time we fly over an active zone of the screen. And even though I'm not saying explicit links are the only desirable kind of linking, I think that their lyrical quality makes them a resource to be exploited to great advantage.

5 Write relevant hypertexts

My pragmatics of links is in line with Robert Kendall's "But I Know What I Like", (16) where he talks about the reader's gratifying feeling of control once she perceives a "clear relationship between source and destination" in the act of following a link. Although I wouldn't speak of "a clear relationship", but of "satisfying relevance" after having searched for a context twice. He points out that "there is a difference, however, between navigational ambiguity that adds depth of meaning and link anchors that are forced on to words that aren't really appropriate for them." Hypertext readers will trust authors that their links are relevant, and when they are not, readers will feel frustrated and might stop reading.

The first thing to consider when planning the links of a hypertext is what kind of interpretive movement we want to provoke. If we want a clear-cut structure where the reader knows where she is and where she can go at all times, we are looking for the following combination:

I stress "informational" because this kind of hypertext (like a newspaper) usually wants the reader to understand one thing, and to find information efficiently, even though the order in which the information can be accessed may vary. The relevance in these hypertexts is determined by the ability to present concrete information. Thus we'll need to:

1. Provide descriptive links. Efficient anchors avoiding ambiguity.
2. Suggest few strong implicatures.
3. Make clear where you are going (definition, bibliography, a related quote...). If there are different kinds of links, a way should be found to visually distinguish them.
4. Provide navigational aids (buttons, maps...): links with known destination.
5. Indexes and other ways of integrating nodes into wider structures.

If we want a structure where the relevance is determined by the cognitive effects of exploring a context made up of a wide range of weak implicatures, we are looking for:

Although the inferring process is the same, our objective (the relevance we look for) as readers is different. We want to take full advantage of hypertext's power of (double) suggestion of implicatures when linking. Consequently, we'll need to:

1. Provide evocative links. Words that are highly meaning-charged in their relationship to the rest of the text.
2. Suggest many weak implicatures.
3. Play with different linking schemes (but ideally, separate the informative links, like bibliography, from the "lyrical links").
4. Let the reader make out the structure of the hypertext, but give her evidence to gather to do so.
5. Play with the reader's expectations when traversing links, reward the exploration of implicatures enriching context.

Both kinds of hypertext should explain their "linking politics" in a node that the reader could access before plunging into the hypertext itself. This would ensure that the reader has the right information on how to use her interpretive abilities. The author is the one who can take the decision of what is relevant in each case. There is nothing that says than non-fictional hypertext should follow the first structure and hyperfictions the second, they can also be mixed and played with. However, as I have suggested above, hypertext lends itself especially well to the second kind of pragmatic communication, that we could call lyrical. Looking at Kolb's and Greco's hypertexts, I would say that the next step in this research is to see if (and how) this lyrical quality affects the nature of academic and other non-fictional discourses. Because I don't think that the first kind of hypertext moves very much away from the implicit textual assumptions of print discourse.