Merzscope map of Hypertext Gardens

Figure 6.29
Merzscope map of Hypertext Gardens

6.4 Web Sites

Machine-drawn webs: nodes and edges

A great deal of ingenuity and effort has gone into representing the structure of web sites in the last several years; robot programs that follow the links from pages and record the results are not hard to devise (though there are some delicate questions about javascript-assisted links), but displaying the results, even for moderately sized webs, has proved quite a challenge. The map at the left is made by a program called Merzscope of a site with 23 nodes (ignoring the links off site and to images,fonts and style sheet), namely Mark Bernstein's "Hypertext Gardens", which is, to be sure, about web site structure. Merzscope is/was a data-driven mapper which allows you to move, remove, and custom tune nodes (as has been done here). If the rest of the nodes were displayed by name, the screen would become very cluttered and the linking lines even harder to follow. And this is a small site. Clever devices such as fish-eye and hyperbolic displays allow high zooming over a part of the map, but the plain fact of the matter is that if you indicate all of the links (including back and cross links) and all of the files in the structure, it is hard to see very much or very clearly on a computer screen. So most data-driven site mappers provide filters to omit data deemed irrelevant to the purpose of the map. Many mappers do not even indicate cross or back links, thus reducing the number of edges in many cases by half.

Hypertext Gardens: Astra Map

Figure 4.30
Hypertext Gardens: Astra Map

This is the case with the most popular figure for mapping which is the star-burst (or, as someone said, the mature dandelion). Since these do not indicate cross links and returns, they represent web sites, which are cyclic graphs, as if they were mtrees. At the left is the same site (Hypertext Gardens) mapped by Astra (the file at the center of the largest burst is the one labeled "Hypertext Gardens"). It should be noted that Astra does have a second display window which for any node will indicate the links into and out of it; thus, a full account of the linkages is kept, though not displayed in one overall view. Other hierarchical tree-branching displays are fairly common, even (heaven help us) file-and-folder. This last really makes very little sense, since a hypertext page with link anchors is plainly a file but is diagrammed as a folder "containing" the files it links to.

Hypertext Gardens: Mapuccino View

Figure 4.31
Hypertext Gardens: Mapuccino View

A representation is only misleading, of course, in relation to certain questions and purposes, and a tree representation does serve certain purposes quite well. There seem to be a least three distinct purposes for mapping a site:

  • to maintain the site (find and fix broken links, moved files);
  • to show the types of information to be found on the site (i.e., to help people find the page they want);
  • to plan a site;
  • bed bookstore london
  • to chart possible paths through a site so that we can see what the consequences of particular choices are (which in many cases we would not want to make available to viewers);

For the first two purposes and possibly the third, an mtree representation is quite adequate. For the fourth, however, cross links and returns are crucial and cannot be omitted. In fact, the directionality of the links must be indicated (for web sites are directed cyclic graphs), and on this point, even the Merzscope network fails. One data-driven grapher that does/did do this is IBM's little Mapuccino program. The Mapuccino map of Hypertext Gardens is at the left, though it is simplified by tracing paths only to a depth of four (blue arrows are back links, which are defined as links to pages already linked). More back links would appear if the depth were raised to six or seven. Nonetheless Mapuccino misses links, in my experience, as do some of the others. Also, it does not allow trivial links to be pruned.

The great advantages of data-driven robot mappers are that they follow links that might be omitted manually, they do it automatically and very quickly, and they produce an orderly layout. The negatives, however, are that they do not discriminate between trivial links and substantial ones, they miss links, and their layout, while regular, is often not illuminating. Those that provide for hand editing, such as Merzscope and Mapuccino, allow one to respace and reconfigure the network to make it easier to grasp and so avoid this last problem. From there it is but a short step to positioning nodes for centrality and importance--to highlight the important nodes and sections and so to overlay a drawing of how meaning is distributed over the site. This use of space and location is not conventionalized and is rarely explicitly indicated, but may induce a certain view of the site all the same. However, none of these data-driven mappers (or the others I have looked at) is very good at representing a site as a set of possible paths.

Hypertext Gardens: Nestor Map

Figure 6.32
Hypertext Gardens: Nestor Map

The drawing at the left, however, is only slightly robot drawn, even in the initial sketch. It is a map of Hypertext Gardens made with Nestor, the free and ingenious mapper-browser by Romain Zeiliger of CNRS in France. Nestor keeps track of your moves as you browse, drawing a arrow from a node representing the page you are on to another node for the page you are going to. It thus maps your path as you browse, allowing you to visualize your immediate "history" and to return to earlier pages and check other hyperlinks you may have chosen to ignore. (You can do other things with Nestor too, like building knowledge structures of page sequences and sending your map to friends or posting it up on the Web.) I built the map by clicking my way through the "Hypertext Gardens" and then moving the nodes in the resulting Nestor map around for clarity. It is clearer in this map than in any of the previous, data-driven ones that there is a final page ("In conclusion"-- here, partly cut off--aka "Garden's End") and only one edge leading in to it. That means that though there are many possible paths through the gardens, they all must converge in Eastgate Lessons ("Seven Lessons") in order to get to the conclusion. It is also clear that the only way out of the Conclusion is to go back, but that there are links back to all the major nodes, not just the top. In this way we can see the consequences of various choices at each point along the way. Clearly, if we want the viewer to follow certain paths and make various choices, we would not want to make the map available as a navigational device (which it is, inside Nestor). Bernstein provides a map of the site drawn with a mapping tool "under development at Eastgate", but it has no hot spots and can't be used for navigation.

One may wonder whether "Hypertext Gardens" is an unusually webby site. Computing the edge-to-node ratio seems an obvious measure (as well as an indicator of which nodes are likely to be the centers of traffic). The edge/node ratio here is above 4. Since to be in a sequence takes two links, the ratio means that each page on the average offers two extra choices. There are lots of connecting paths in this little garden. There are a goodly number more, it should be noted, if we indicate (as the sitemappers do) all of the links off site. "Hypertext Gardens" links every page to the Eastgate Systems Home page and there are links to other pages under the parent directory as well as entirely off site to Netscape and Yahoo! Which links to count as part of a site is a matter of some delicacy.

Vera Frenkel: Detail Map for Body Missing

Figure 6.33
Vera Frenkel: Detail Map for Body Missing

Hand-drawn Webs

These machine- and machine-aided graphings of web site structure have their uses, but whatever one might say about them, they are not much to look at: they have no shape and there is no view to their overviews. By contrast, the hand-drawn sitemap of Figure 6.33, which may draw some inspiration from story boarding, is ideally adapted to showing flow from the top through the site. It is a detail map of part of the site, "Body Missing", which is a Web adaptation of an installation by Vera Frenkel and is quite large. It is an image map, the various handwritten titles serving as links. It indicates cross links but not returns. "Site Map" links to the top sitemap (click on this image to see the top sitemap) and "Pianists' pages" links to another detail map. The device of nested maps greatly relieves the congestion and clutter that would arise in the top sitemap if it were the only one for the entire site. It also avoids the "flattening effect" of sitemaps which are imagemaps; that is, such an imagemap where all the page names are hot in effect supercedes the web structure and makes all those pages one click away. This is fine for a menu or table of contents, but if you want the viewer to experience maze/garden structure, you would not give her an aerial view with direct access to all points. Frenkel is taking a certain risk that viewers will not follow her page-to-page sequences but use the map to dip in anywhere, but then again the pages are not on the whole strongly sequential (though they are so in certain series).

The next step away from the data-driven machine-drawn maps is to use an image that in some way epitomizes the site. Shelley Jackson's use of a white outline sketch of her own body as a contents page for her site "The Body" has been cited quite often (e.g. by me and by Martin Dodge and Rob Kitchin). Particular body parts and zones are labelled, and clicking on either the word or the part of the body takes you to a page providing her experiences and thoughts about the part. Since the entire site is simply a collection of these reflections upon parts of her body, the drawing does epitomize the site, though it is scarcely a map of it, since the site has many more pages than are linked from the drawing. This equivalence of drawn part, label, and linked text describing and reflecting on the part cause the normally accurate and sober Dodge and Kitchin to exclaim, "Here the map is the territory it seeks to represent." (83) That would make it a purely self-reflexive, poststructural artifact or one that treated the body as text. But viewers might with some justice assume that the normal conventions are still working here: the drawing represents the body, parts and whole, the texts describe the parts, and the territory these maps represent is Shelley Jackson's body and its parts.

Udi Aloni: Re-U-Man

Figure 6.34
Udi Aloni: Re-U-Man

Figure 6.34 performs a similar function at the top of Udi Aloni's "Re-U-Man" site. Here the male body (of Aloni, presumably) represents himself, which is indeed the topic of the site, but surrounded in this case with labels and symbols and glosses and various connecting lines. Various words and symbols cluster in hot areas which link to major divisions of the site, but there other things overlaid as well, namely the parts of the mystical body of Adam Kadmon with "crown" at the top and "kingdom" at the bottom from the Hebrew Kabbala and the crown diamond pattern of the Tree of Life, also from the Kabbala. In the Kabbala, the word, the symbol, the figure, and the physical are all overlaid and interpenetrate, so that indeed the body is the word is the mystical body and the map is the territory--not, as it would be in a poststructuralist view, a set of signifiers floating free of any physical or historical world but as a mystical body which incorporates the world. At this point we are not mapping a site, we are mapping the world and its most ancient and enduring structures.

Screen capture of tendril crawling the Aesthetics and Computing Group site at MIT

Figure 6.35
Screen capture of tendril crawling
the Aesthetics and Computing Group site at MIT

Mature tendril diagram of ACG site with external links dark

Figure 6.36
Mature tendril diagram of ACG site with external links dark

Web site as life-form

Shedding any reliance on images of objects or even a metalanguage of nodes and edges, Tendril by Ben Fry at the Aesthetics and Computation Group at MIT's media lab uses the words of one hypertext document to trace the link to the target document. The words become tendrils that spiral over to the target documents, which in turn send out word tendrils to documents they target, and so on until one stops the mapping. The resulting structure apparently can be zoomed into and navigated around. Tendril makes us aware that the standard notations we have looked at represent documents as bounded objects (file icons, circles, boxes). The exception is Vera Frenkel's non-standard sketched sitemap. As we manipulate documents on a computer, opening and saving them, they do feel very discrete and object-like, but much has been said upon the softening of the borders of hypertext-linked documents, the dans/hors of texte, of the way a word which is a link-anchor both serves its purpose in the context of the document and yet yearns and harkens after another document, another context, and of the rhizome as the naturally occurring structure closest to a hypertext web. And indeed, tendril-diagrams, especially in their tan on white display option, resemble nothing so much as wildly ramifying ginger or Jerusalem artichoke. (The image at left is a rather mature tendril-diagram of the Aesthetics and Computing Group site with external links in indigo blue.)

The organic look is a part of Fry's larger project, which is the modelling of large, changing data systems as structures with many of the properties of living things such as growth, metabolism, homeostasis, responsiveness, and adaptation. These ideas are set forth on line in his Master's thesis, "Organic Information Design" (2000) ( people/fry/thesis). The general hope is to "augment intelligence" by making large-scale, emergent qualitative patterns in the data evident. So for example another of his creations is anemone, which represents patterns of visiting on a website. Pages are represented as at the ends of branches which grow bigger as the pages are visited. The resulting figures do grow by feeding on the data of visits and resemble anemones somewhat. It is not enough for Fry that an Organic Information Visualization (OIV) look like a plant--it must behave like one too. Thus screen captures just hint at the full realization of tendril and anemone. In the end, then, Fry's OIVs are not intended as art (that is, to be enjoyed for their shapes, colors, and design), but as a means of seeing what would otherwise appear jumbled, cluttered, and overrun with irrelevant data.

Screen cap of Anemone plotting traffic information for site; very early stage

Figure 6.37
Screen cap of Anemone plotting traffic information for site; very early stage.

The OIV framework is thus not a metaphor but a schema for a whole family of metaphors for abstract structures, specifically dynamic structures with large amounts of interrelated data. In a way, it is the computer-enhanced descendant ("outgrowth") of the trees, bodies, and, yes, even the webs that have been pressed into service to show the abstract relations of entities to each other and to a whole. The shapes that emerge from Fry's programs are not those of daily life and thus do not bear with them (for better and for worse) the abundent associations that trees, bodies, and webs do; rather, they are, somewhat vaguely, “biomorphs”, but they do turn our thinking about hypertext web sites away from points connected by straight lines to continuity and flow, strength and growth, atrophy and decline. We of course are not used to reading anemone diagrams, but Fry thinks that our normal "graphical sensibilities" (86) should be sufficient, though assisted in some cases by before-after comparisons and in some cases with further analysis of the data. "Brighter and thicker means more" is certainly easy enough, but I am less sure what the bulging, twisting spirals of tendril are supposed to mean. In general, spirals signify dynamism, and the overall image is of text connecting to other text, but unless we know something about how tendril selects the text for spiraling, it is hard to say much very specific about what a tendril map says.

translation of IP number to color

Figure 6.38
translation of IP number to color

Conclusion: Map or Territory?

Lisa Jevbratt has created quite a splash with "1:1", a data base of the 186,110 IP addresses that replied to her web-crawling mapper (i.e. that have active servers) along with several interfaces to the data base which display those addresses as single pixels in various ways. Sweeping claims have been made about this project to the effect that it represents the entire Internet, or that it is the entire Internet. To understand those claims, we need to look at the project somewhat closely.

In the most celebrated interface (every), the pixels, each of which are hot links to their site, are displayed in lines and ordered from lowest IP address to highest. Each pixel has a color based on the last three triples in its address. So in an address, the 50 would give the decimal value of R(ed) component, the 212 would give the G(reen) value, and the 110 would give the B(lue) values for the pixel (a kind of teal).

section of the every interface of "1:1"; click for full screen view

Figure 6.39
section of the "every" interface of 1:1
click for full screen view

The display area is divided into 74 zones (only the domains that have been probed); each zone has subzones corresponding to the second triplets of the IP numbers probed in that domain—e.g., the 128. domain has been probed for 48 sub-domains ranging from 2 to 253; each pixel representing a site in a subdomain will have the same value of red in its color mix. Thus within each domain, each site, potential or real, would have a pixel with a unique color. It is useful to see the sequencing as the inheritance of domains, as illustrated at the left. If every IP address in the 128 domain had a server responding at it, there would 256³ pixels, one for each color available on a 16 million colors video card. They would not fit on anyone's screen, I fear: a screen at 1024x756 resolution has fewer than a million pixels. If we could see this unseeable sight, we would see smooth sequences, each pixel changing by just one. If we had a bit more room, we could repeat this panorama for the other 255 domains, all of which, if completely full, would be identical. Perhaps tile one domain 256 times? But of course none of the domains are optimally full and in the case of the database, are rather lightly sampled, so each set of pixels for a domain will be different—with some matches with pixels in other domains. So you would not necessarily need 186,110 different colors for the 186,110 responding sites, but you would need 2515 (186,110/74) colors at the very least, and probably a very substantial number more. The highly-magnified section at the left of the upper corner of the interface-gif does not appear to conform to this description, since it has bars of identical color, whereas the fundamental element should be the individual pixels. These bars I suspect are a result of Jevbratt's decision to make the composed image a GIF file, which has a maximum of 256 colors (and this one has 241). Thus, a lot of the finest distinctions of color are lumped together. (This is the cause of "banding" when a full color gradient is saved as a GIF.) So not even a vision-enhanced cyborg could pick out the individual pixels on this map in many cases, except by approximation. So 1:1 database does not cover the entire Web, nor does the every interface represent the sampled Web 1:1, a pixel of one color for each site (in a domain).

We are now prepared to address the question: is Jevbratt's every array a map of the internet (or a part of it)? The title "1:1" suggests a scale for a map, and is identified as such by Jan Ekenberg in a Prologue attached to the site: " The title of this project: 1:1, as in scale 1:1 suggests that a map, or a model, has the same size as that which it refers to." This amounts, says Ekenberg, to a

a collapse between the map and the interface. But the postphotographic practice of the "1:1" project makes the implosion even more severe. The interface becomes not only the map, but the environment itself.

I confess I find this bewildering. The interface represents each responding IP address with a single pixel and arranges the pixels on a line from lowest to highest numerical value. Postphotographic practice is generally understood to mean dealing with images without thinking of them as depictions of parts of the physical world. So all of the maps and diagrams discussed in this chapter are "postphotographic." By environment, I think Ekenberg means what we have been calling territory. And so we are back at the by-now-familiar "the map becomes the territory" claim. Jevbratt herself refuses to call the interfaces of "1:1" maps:

The interfaces/visualizations are not maps of the web but are, in some sense, the web. They are super-realistic and yet function in ways images could not function in any other environment or time. They are a new kind of image of the web, and they are a new kind of image.

I suppose the pixels are called super-realistic because their colors are defined by the same values that define the IP address' position in the domain: each place in a domain is represented by a unique pixel (at least in principle). Beyond that, I am not sure what is meant by super-realistic, because the pixels are not exactly what you would call images, any more than the individual tiles of a mosaic are images. In any case, having suspended all standard ways of describing her new kind of image, she turns around and tantalizes us with geographical metaphors for the patterns in the every display:

The variations in the complexity of the striation patterns are indicative of the numerical distribution of web sites over the available spectrum. Larger gaps in the numerical space indicate an uneven and varied topography, while smoother color transitions and more consistent layers are indicative of "alluvial", or sedimentary, flat lands in the web's IP space.

(That is, when the various subdomains are well populated with active servers, each pixel is very close in color to the adjacent ones.) I can not see any serious purpose in this turning of IP space into geographical terrain. If anything, it is a demographic map of IP space. Smooth or imperceptible transitions indicate high density of population by responding servers. To speak of the putative map this way is like describing a topographic map in terms of pools of water and stretches of moss and dirt. At this point map and territory have collapsed, and the patterns of the putative map "indicate" a metaphorical topography of broken-up ground and alluvial flatlands. To be sure, Jevbratt likes to spin metaphors around her "new kind of image." She also speaks of it as interlaced, mainly because the robots continue to check whether the servers at the 186K IP addresses are online, and so refreshes segments of the master image from time to time. This has led Dodge and Kitchin to liken it to an image of interference on a TV set (249), though the point of the analogy remains unclear. Every is not the result of a malfunctioning device; it is a perfectly function device for some nonhuman eye. That is, the every array is a map, albeit an unusual one that requires more than human acuity of vision to be read. If there were a cyborg that could lock in on one pixel at a time and had "perfect color" so that it could read all but the top (domain) values of the IP address the pixel presented, then the map would tell it how populated the neighborhood of the pixel was and indeed what the addresses on the street are. To us humans, or at least to me, it looks like alien code, written for beings whose minds seem quite different than ours. The fact that it can not be read using our usual "visual sensibilities" may bring out the cryptographer in us, and it may be that some time in the future, should every arrays hold our attention that long, we may learn to read them, which is to say see significant patterns in them the way the cyborg might. But for now, every stands at the edge of the visible world, or slightly beyond it.

We began with the claim that visual representations (maps) of abstract things must be metaphorical, in the sense that the images are being pressed into service to represent things which cannot be seen. Each metaphor has connotations (or associations) which may or may not be relevent in the world of the abstrations. Reading a metaphor involves grasping the connoted as well as the said meanings and suppressing the irrelevant connotations. These connnotations may shape our thinking about the topic even when they are peripheral to the point at hand. All of this is very like what goes on with word-metaphors.

The images of structures are not the structures themselves, and it is very difficult to talk about those structures without using metaphors. We used terms from graph theory (nodes, edges, cyclic, directed) which are as colorless and connotatively weak as possible. Still, when we come to represent these structures visually even in the most austere mathematical fashion on the page, the visual elements we use bear connotations that could mislead if not suppressed and others that are relevant and should be activated. Does top-to-bottom or left-to-right orientation matter? Edge lengths? And when we use more metaphorical images (trees with trunks and leaves, columns, slabs, planets and galaxies, bioforms) the stream of connotations becomes a flood. This is one of the main reasons to use a visual metaphor, of course—to activate connotations which give a useful take on the abstract structures under discussion. We value a visual metaphor insofar as it provides multiple useful connotations and few misleading ones.

One can get carried away talking about connotations, for they are notoriously open-ended and variable from person to person, though they do have some impact on the way we may think about abstract things. Further, the words we use to describe the viusal connotations also add their own connotations to the soup. When we talk about relations in a tree and call them hierarchical, we summon a whole flock of political/social (verbal) metaphors of subordination, domination, and command. Indeed, Kress and van Leeuwen argue that maps of abstractions ("conceptual representations") are modelled on forms of social organization (87):

The taxonomy is modelled on a static, hierarchical organization in which everything has its pre-ordained place in a grand scheme unified by a single source of authority.

Taxonomy suggests also Linneus' great ordering of life forms, which doesn't seem very political, though it has kingdoms at the top. They contrast this view to that implicit in networks:

The network is modelled on a form of social organization which is a vast, labyrinthine network of intersecting local relations in which each node is related in many different ways to other nodes in its immediate encironment, but in which it is difficult, if not impossible, to form a coherent view of the whole.

This was written just as the Web was beginning to weave itself; they proceed to quote another writer who sees the Net as the 21st century cultural icon/emblem/archetype replacing the atom icon ("that law abiding solar system of energy") of the 20th century.

Lev Manovich is not so explicitly political in his recent The Language of New Media, but nonetheless sees software interfaces as privileging "particular models of the world and the human subject" (16). Beginning with the proper name of the Macintosh file system (HFS: Hierarchical File System), he also contrasts hierarchy and network:

A hierarchical file system assumes that the world can be reduced to a logical and hierarchical order, where every object has a distinct and well-defined place.

To this assumption he contrasts the Web model:

The World Wide Web model assumes that every object has the same importance as any other, and that everything is, or can be, connected to everything else.

Note that Kress and van Leeuwen, who clearly are thinking of social and political hierarchies, would never call them logical! Although these writers are officially talking about the power of models, one cannot help concluding that for grand world-encompassing generalizations (aka overview), there is nothing quite like words.