Hypertext as a Thought Construction Kit

Jean-Marc Lepers

DESS Ethnomethologie, Anthropologie et Informatique
Département d'Informatique - Université Paris 8
2, rue de la Liberté, 93526 Saint-Denis Cedex 02, FRANCE


This paper focuses on the ways associative thinking was implemented in computers. It points out the cultural pattern of hierarchical or classified systems, and its presence in the "nodes and links" structure.
It recalls the specific experience and the results of an anthropologist designing his own hypertext generation program.
A mixed practice of reflection on anthropological data and program designing leads to a specific model :
- Texts shall be considered as nodes of concepts - Concepts are links between texts - Texts are links between concepts - No specific object shall be considered as being per se a node or a link
It concludes on the position that any classifying approach is a limit to the use and development of associative thinking.
And, futher, assumes that hypertext system's development is dependent on the acceptance of multiple views on reality.

Hypertext, Anthropology, Rhetoric, Cognition, Information retrieval, Humanities, Symbolic system, Formal model


Our Big Ancestor, Vannevar Bush [1], imagined that "wholly new forms of encyclopaedias will appear, ready-made with a mesh of associative trails running through them, ready to be dropped into the memex* and there amplified". He took an example from a comparison between the Turkish bow, more efficient, and the occidental "long bow" used during Crusades, and another from "the queer ways in which a people resist innovations, even of vital interest". And now, what is up ? The most non-anecdotal thing we get today is the Internet. It gives us a lot of data, mostly organised in hierarchical files. The "associative trails" remain a local, particular work.

* memex is the name given by Vannevar Bush to his imaginatory linking system

1.1. Associative Thinking and Computer Science

So, what happened ? maybe, the rising of computer science. Vannevar Bush imagined, before computers, an appropriateness between his mental associative system and a technical, simple but unusable system. In his own field, Alan Turing did the same thing. But the computer science was entirely designed by very normal people, whose mental patterns belonged to standard hierarchical systems. It was maybe necessary to organize memory and file allocation tables, but not to design structured or object programming. Structured or object programming is very convenient, but a reverse effect is that computer scientists tend to consider that data themselves shall be part of a structure or objects in a hierarchical class system.
Did hypertext systems miss their original aims ?
Maybe not. We may have to assume that the real problem, within hypertext and hypermedia systems, is an internal one. I confess I spent a lot of time trying (unsuccessfully) to convert the rich and moving contents of my own hypertextual database on Anthropology into a kind of expert system. Why ? just because the links I created into this database seemed meaningless to most of my colleagues. It was just as if I had done nothing. On the other side, they were trained to be fascinated by rhetorical demonstrations, inherited from the first Middle-Age Universities traditions. These cultural patterns are present even in the field of computer sciences, as "ordinateur", french name for a computer, is one of the multiple names of God used in ancient french texts ! Some philosophers include hypertexts in their philosophical rhetorical systems That does not make any hypertextual system work; cultured people like to dream about a kind of "New Age of Communication". As far as I know, Ted Nelson , who started to learn computer science in order to write philosophical books, realized in the sixty's that hypertext systems were a new way of apprehending reality. As far as I may know (I never have heard anything about Ted Nelson on T.V.), he never wrote any philosophy book. When Ted Nelson invented the term "hypertext", computers were far away from being the complex structured machines they are today. And the artificial intelligence was not able to propose models of reality.
So, what is the problem ?

1.2. Hypertext and Rhetoric

It could be that the main resistance to the hypertextual project lies in the rhetorical pattern. This resistance does not lay only in the external world, but in our own cultural patterns. As an anthropologist designing hypertextual software, for my own needs at first, I realized gradually that the resistances to the use of real hypertexts (i.e., including a large range of linking abilities) were mostly cultural.
Why did I need to write my own programs ? To start with, because nothing available seemed to correspond to my needs. Anthropology is an hyper-complex discipline, as it attempts to construct general overviews on this hyper-complex object that humankind is. Something not really easy to reduce to an expert system. I realized later that software designers used, unwillingly, the same organisational and rhetorical patterns than the other members of the scientific community. Therefore, a computer being basically a mess in which you may install any order you like, some computer scientists probably get a specific kind of aperture hard to find in other more traditional sciences.
Well, it appears that computers, and specifically hypertext systems, are very disturbing objects. The way we think is strongly dependant on the medium we use for thinking. An oral culture does not think in the same way as a literary culture, and different literary media (from hieroglyphs to modern standardised papers) are different reflects of different ways of perceiving the world. If we assume that, hypertext systems may look like a virtual tremendous shock for our traditional literary and rhetorical culture.
In an hypertextual system, the thought is not a demonstration, but a set of links. In a constructive hypertext, we may always construct new links or new views on facts. Links are the thought. And, what is worse for a Cartesian mind, you may loose the control of those links. I would say, one have to lose this control for making hypertexts a really interesting and productive tool. Otherwise, this is just a strange copy of a classical book, trying to translate in a nodes and links environment the splendid order of rhetorical books.


2.1. Nodes and Links

As a general agreement, hypertext systems are composed of sets of nodes, anchors and links. A very popular example of this approach is "Hypercard". Most of hypertext design systems follow the same basic rules and require for the main hypertext conception to be hand-made. They assume that there is a fundamental difference between the object Node and the object Link. In some sophisticated, object-oriented versions, we may get typed nodes or typed links [5] getting specific functions or methods. In the hypertext community, the differentiation between nodes and links may look like natural, like a general pattern of human thought. It does not seem much, but this primary differentiation is strong enough to impose a specific kind of a representation to an hypertext. Most of the time, an hypertext looks like a network of nodes and links; some have a visual representation of the network (and if you get a lot of nodes and links, it looks like an awful mess), and some present to the user the only available anchors or browsers on one node, but the network is the underlying structure.
The problem is : do we really need to radically separate nodes and links, just like the old tradition in computer science distinguishes the data and the allocation maps ?
Computer science provides us with a large range of procedures that are capable of linking, in different ways, very large sets of data. Some systems designers [2] [7], aware of the inherent limits of the hierarchical file systems, provide the user with "link servers" or organisational tools for organizing all documents in databases. They all respond to a real need : the organization of documents so that any user may construct his own sets of links.

2.2. Hypertextual Databases

Databases systems, and, specially, relational databases, distinguish clearly between two levels for information retrieval : data and request. The link provided by a database, for the user, is a request in database language. The rules for the construction of relational databases and requests are constraining; they usually follow the Codd model. This model is constraining, but clearly formalised.
The databases query languages use to be very awkward : a request in SQL language may be formulated in several standard 80 column lines. But, on the other hand, one never has to pre-construct ones linking system : any data in a relational database system is virtually accessible, providing that the model is respected.
Actually, the querying system was quite boring. But some graphic database systems are now available, using different kinds of browsers, and widely distributed in professional graphic environments. These systems allow the user to "point and click" on the data they want. They are therefore able to construct almost any kind of request or links in a large amount of files.
Within a relational database system, the link is the request. In order to construct your link, you may use browsers into your data, and the program will automatically construct the request you want. What we could call a node is just the graphical interface or form in which you may collect data for a new request. But there is no difference between the data and the links : links are constructed with the data themselves.
A form may provide a browser on the document's keywords, and another one on the related quote's keywords. Quotes may be selected according to their keywords. You may have access either to the quotes, or to the keywords system. That kind of form intends, as much as possible, to be fast and easy for the user. One pay great attention to users if one is really one of them. That might be why successful Internet programs were usually designed for an internal use, responding to internal needs, before they became public.
Another form provides three browsers through documents, quotes and notes, all related through the same keyword.

2.3. Links and Concepts

Hypertext designers cannot avoid, one day or another, this question : "why this link ?" It actually appears that any link is a kind of concept.
Therefore, the use of concepts themselves as links may do a good part of the job.
So, what is the game ? To get as many links as possible. And, as a part of the game, to generate links between the concepts themselves.
How is that possible ? Very simply, by considering the texts, or what would be considered as nodes in a classical hypertextual system, like links between the concepts. We shall see further on examples of those procedure's results. A text, or an image, is not only a query's result, it is a link between concepts or sets of concepts. In an hypertextual database, concepts are not only links between nodes, they are inter-linked into the nodes themselves. And that specific inter-linking represents, through these particular nodes, a kind of thought environment.
Compared to a traditional hypertext or to an expert system, the game becomes totally different. It is no more : what kind of link organisation may I set for producing a complete and well-structured hypertext (avoiding user's disorientation, navigation problems, and so on). The question becomes : what are the internal structures in a database constructed step by step, link by link, any link reflecting a particular view on a specific problem ?
Or, what are these links ? The Bush's "associative trails"' problem, if it is one, is its richness. Hypertext systems may help us to know a little bit more about this associative thinking, not through the construction of artificial, intelligent, and coherent systems, but through exploration of the strange, or non-conscious, associations lying in databases.


3.1. Materials and tools

How will an anthropologist produce an hypertextual system ? He is aware at first of his own needs : he gets a large collection of bibliographic references, quotes and notes, plus the will of producing specific browsers on the data. The construction strategy is mainly bottom-up : the set of data is so wide that one cannot organize them in a pre-existing model. A model might be considered as belonging to a top- down strategy : you know how the data shall be processed or inter-linked, and each one gets its specific room in the system. Well, the anthropologist has a secret hope that he will someday come out with a fantastic model which will explain everything; his bottom-up strategy helping him to construct the top-down model. Unfortunately, the hypertext discipline will always demonstrate to him that human data are too rich for belonging to an unique model. *
* Corrector's note : I don't agree, I think that anthropologists think that they have to present things in a certain way in order in order to conform to "academic" standards and that they secretly know that human diversity and complexity cannot be reduced to any model - C.B.M.

Firstly, the main work is the collect of data themselves. Most of scientists in humanities have to spend years and years in National Libraries for the construction of their own collection of data and quotes that they consider to be important. Large hypertext systems in which one can get everything and anything may be useful for the first work of information retrieval, but are quite irrelevant for the construction of rich inter-linking. For that job, you need a set of very significant bibliographic references, quotes and your own notes. Why does a quote looks like significant ? Most of the time, demonstrations, analysis and so on are a kind of rhetorical game, sometimes brilliant, but they do not provide significant quotes. The significant quote is a node where concepts are inter-linked. When a researcher writes a draft note, he will most of the time link words or short phrases on a piece of paper, creating a visual representation of an inter-linking. Actually, he is creating a node.
The second work is to extract the keywords or concepts from the text. You may use full-text procedures for helping you in that job, but most of the time, concepts do not figure in the text as words, and this work may be done only by the researcher himself, due to his knowledge of the domain.
Third, the inter-linking procedures. Texts including the same concept are automatically inter-linked : they constitute a set. As a text is itself a node, it belongs to many different sets. Any concept may provide a view on a relevant set of texts.

3.2. Hierarchical concepts organisation

At this point of the work arises a very anguishing question : where is the model ? or, in other words, where is the organisation ?
As all the concepts or keywords were on the same level, a lot of my colleagues (but, significantly, not students) insisted for the reintroduction of hierarchical, correctly organized level. So, it gave a kind of semantic network, organized in multiple sets and subsets. But the problem remains always the same : any concept may belong to multiple sets.
This is a graphical representation of that kind of network :

By the way, I never really used it. In which set is it possible to include concepts like "love", "truth", "emotion" or "logic" ? Any classification looks like totally arbitrary, even if you may use multiple classifications.
Another question is the "levels" or "classes" problem. If it is obvious, in a semantic network, that a "dog" is a "mammal" and a "carnivore" (but they are "birds" who are "carnivore", too), it is most difficult to know what is a "Symbol" (almost everything), or what is "Money", and so on. In a bottom-up strategy, you cannot erase the first level; for instance, in this network, "Symbol" belongs to the set "Symbols", but, at a time in my research, I made it belong to "Exchanges" too (for my own reasons).
So, the organisational problem remains.

3.3. A non-hierarchical concepts inter-linking

It had to be some other solutions. One of them is this one : let's not consider anymore that the text is the dead-end in a concept's architecture, but fully let's consider it as a node. And, why not, as the only relevant possible node. As a node, the text assumes the inter-linking function. So, the result is : if text is a node, the inter-linking between concepts shall be done through the texts themselves.
In terms of nodes and links, it appears that the text, at first considered as a node or a final result in a request, becomes for that specific function a link between the concepts.
This provides this kind of network :

Unlike the previous kind of multi-hierarchical structure, I love to use this kind of inter-linking procedure. First, because it provides a view of the inherent inter- linkings in the database. Second, because these inter-linkings are not hand-made, but automatically computerised. Third, because the appearing inter-linkings may often show that a lot of texts you never consciously inter-linked, because they belong to very different domains or epochs, get some general conceptions in common.
I like this inter-linking about "happiness" because I am not fully responsible of it. In a demo, somebody asked me if I get anything in my database about the question of "Happiness". Actually, I never gave interest to that question (shame on me !). So, I tried to extract something about "happiness". As I only get a kind of common sense about "happiness", I would be very embarrassed if I had to classify it in an hierarchical, or even multi-hierarchical, structure. Would I include it in a "Feelings" set; itself related to a "Psychology" set, to "Philosophy", "Anthropology", or whatever ? It could, as well, be related to neuro-physiology, and so on. First, it's almost unfeasible; second, all these classifications are cultural.
So, I tried a first full-text research in the database. At this point, I have to apologise about the quality of translation : my original texts are mostly in French and I do not get at this point the corresponding English translation. And I only give here the relevant part of the quote.
This research gives :

FREUD, Sigmund, Discontent in Civilisation
"Another process is more radical and more energetic; it sees in reality the unique enemy, the source of any suffering. As it makes life impossible to us, we shall break any relationship to it, if we want to be happy in some way. (...). But we may go further and try to transform this world, to edify another one where the most painful aspects will be erased and replaced with others in accordance with our own desires. The being who, racked by a hopeless revolt, follows this way for reaching happiness, will normally end to nothing." (....)
ROHEIM, Geza, Psychoanalysis and Anthropology
"They bring back to the village the man they killed. They don't eat him themselves but give his body's parts in return for kune objects. "The bagi is the pokara (initial gift) given in return for human flesh. Against a mwari, we give a breast, against a dona an arm, against a bagi a leg, a nasal bone buys a head, a stone axe a neck. If somebody does not get any kune object, he will say : "Please, cut a little piece for me"." "Pork is bad, man is better". "It's our own smell, so it is good". "Their happiness" means human flesh." (field notes)
SOCRATES (note from an encyclopaedia)
The essential of his philosophy consists in his faith in human reason by which man may reach self-knowledge and happiness ("Know yourself", "Nobody is willingly bad")
SOPHOCLES, Oedipus King
"The excess (hybris) gives birth to the tyrant. When the excess has foolishly stuffed itself, without considering the hour or it's interest, and when it climbed to the top, on the summit, it suddenly falls into a fatal rave, where it's broken feet refuse to serve it."
"He (Oedipus) had set his sights to the top. He had made himself the master of a perfect fortune and happiness"

SAINT-JUST (French Revolution), Discourses at the Convention
"Happiness is a new idea in Europe"
"We told you about happiness : the egoism took advantage of this idea to exasperate the aristocracy's cries and furores, the desires of this kind of happiness made of other's forgetting and superfluity's enjoyment were suddenly awakened."

At this point, we did nod build any philosophical theory about happiness, but we may notice that the "happiness" notion may be linked to a very large set of others : cannibalism, incest, excess, guilt, egoism, tyranny or aristocracy. It's not the place here for an analysis, but we cannot avoid to notice that in almost every text, excepting cannibals, "happiness" is linked to a reproved comportment. What is a bit surprising. Happiness, in common sense, is as everybody know linked to very positive values. On the other hand, a wide research on a large database would probably have given a lot of uninteresting considerations of common sense about "happiness". Maybe not, I did not try.
Feeling this "happiness" notion as interesting, I create it as a keyword and link it to the relevant texts. Then, I try to get all the related keywords through the texts.
This gives :

  • Anthropology : Roheim
  • Cannibalism : Roheim
  • Civilisation : Freud
  • Dream : Freud
  • Exchanges : Roheim
  • French Revolution : Saint-Just
  • Greece : Socrates
  • Greece : Sophocles
  • Idea : Saint-Just
  • Idea : Socrates
  • Ideology : Saint-Just
  • Incest : Sophocles
  • Measure /Moderation * : Sophocles
  • Melanesia : Roheim
  • Money : Roheim
  • Mother : Sophocles
  • Neolithic : Roheim
  • Philosophy : Socrates
  • Pleasure : Freud
  • Psychoanalysis : Freud
  • Psychoanalysis : Roheim
  • Sexuality : Freud
  • Society : Freud
  • Systems : Freud
  • Unconscious : Freud

* mesure cannot be translated in a single English word

This list gives the connections between the notion of happiness and others in the database, and the texts where the links are present. For instance, if I am interested in which way happiness is linked to exchanges, I will get Röheim's field notes on the specific link Melanesian people did between happiness, exchanges and human flesh. or, about incest, the link Greek culture did between happiness, incest, moderation and tyranny. Some of the list's keywords are related to other parts of the quotes, not reproduced here. But, usually, they are not much longer than a standard computer screen, because the number of keywords could increase considerably with the length of the quote.
It may be interesting, too, to look at the other common keywords on the texts. For instance, here, we may see that both Socrates and Saint-Just produce texts linking the "idea" notion to "happiness". We can, then, do the same kind of research from the keyword "Idea", and so on. It provides a view on the concept's environments. Any concept is a link and a node. Both structures of texts and concepts are fully intricate.



4.1. Are we still thinking normally ?

I am not so sure that thought is per se associative, but I may afford that the use of hypertext systems widely develops associative thinking. The researchers who are used to follow links get a specific ability to connect their domains, even if they are distant, using links that are fully associative, and not rhetoric or argumentative. For instance, an ethnologist whose field is Amazonian Indians did a connection with my database through a link that was not theoretic at all, the "claw". In Amazonian Indian's culture, the Jaguar's claw is a very specific object, as the Jaguar belongs at the same time to the normal reality and to a kind of supra natural or infra natural world. A shaman may be sometimes considered as a kind of Jaguar, because he is able to travel into several worlds and manipulate them. So, the Jaguar's claw linked in the database to Egyptian initiations (where a lynx claws the initiate's eyes), to Sophocle's Oedipus (the Sphinx's claws) to the Chimera in Gérard de Nerval's work, and to a Vladimir Nabokov's text where an incestuous relationship between adolescent first cousins make them escape from "the claws of reality". A well- formatted spirit, convinced that a real thought could only come from his own spirit, would say it's just a coincidence. For me, I am convinced that I would never have a chance to link the Amazonian's conception of reality and Nabokov's conceptions without hypertextual methods.

4.2. And a little bit further...

As anthropologists, we are very interested with traditional, symbolic and associative thought. Robert Jaulin [3] demonstrated that divination systems combine a formal model and a symbolic representation. And the divination systems work in a fully associative way : the combination of the elements linked together by chance gives the answer. The number of possible combinations being very high (hundreds of thousands for a single question in the Tarot cards system), the answer depends on the non-logic associations between cards.
Some anthropologists consider that divination systems are hypertexts. I will not discuss it, but just notice that they pointed out the fact that divination systems and hypertext are both founded on formal models and associative methods. Sometimes, I worry about this question : are we creating a piece of a new way of looking at the reality ?
Anthropologists and ethnologists are used to move into different worlds, different perceptions of the reality. They are trained to consider that the formal and symbolic representation a culture uses is a view on reality, as well as a construction of this particular reality for the people who construct it. Ethnomethodologists go even further and assume that any individual gets a unique competence, characterised by his indexicality, his own set of concepts and symbols. The permanent contact with different cultures or realities discontructs the underlying cultural models we all use to construct our thought. By the way, they often consider that one gets particular views on the world, exactly as if the world was an hypertext.


Patrick Deshayes provided a lot of non-textual information through his movie "Nawa Huni" on dream, image and reality in an Amazonian Indian tribe, and through informal seminars. Yves Lecerf was a considerable help for the approach of formal models and the concept of "unique competence" in ethnomethodology. Harald Wertz, and many researchers involved in I.A. programming or in neural networks at the Paris 8 University Computer Sciences Department contributed to a wide approach of the organisational problems.


[1] Bush, V. "As We May Think", Atlantic Monthly, pp. 101-108, August 1945.

[2] Davis, H., Hall, W., Heath, I., Hill, G., Wilkins, R. "Towards an Integrated Information Environment With Open Hypermedia Systems", Proceedings of the ACM Conference on Hypertext, Milano, Italy, November 30 - December 4, 1992, 181-190

[3] Jaulin, R., Richard, P. Anthropologie et Calcul, Paris, U.G.E, 1971

[4] Moulthrop, S. "Towards a Rhetoric of Informating Texts", Proceedings of the ACM Conference on Hypertext, Milano, Italy, November 30 - December 4, 1992, 171-180

[5] Nanard, J., Nanard, M. "Conceptual Documents : a Mechanism for Specifying Active Views in Hypertexts", ACM Conference on Documents Processing Systems, Santa Fe, NM, Dec 1988, 37-42

[6] Nelson, T. Literary Machines. Sixth Edition, 1984.

[7] Rizk, A., Sauter, L. "Multicard : an open hypermedia System", Proceedings of the ACM Conference on Hypertext, Milano, Italy, November 30 - December 4, 1992, 4-10