Because my research includes the domain of knowledge representation, one of the things that I’m interested in is a better understanding of the role that concepts of subject play in wikis. Typically, a wiki page has a title that establishes its overriding subject. What is the relationship of that stated subject to the actual content? How does a single word or phrase come to represent what is likely a multitude of subjects?

Now, one of the things that kinda bothers me about wiki pages (just a bit, but not enough to lose any sleep over it) is that they have just one stated subject. When one looks into a book, the stated subject of a book uses a set of subjects. For example, Robert Brandom’s “Articulating Reasons” has this list of Library of Congress subjects:

1. Language and languages--Philosophy. 2. Semantics (Philosophy) 3. Inference. 4. Reasoning. 5. Language and logic. 6. Expression (Philosophy) I. Title.

This states that Brandom’s book is about six subjects (which of course only skims the surface. It’s likely about many more, as would be any book.) Point is, the technology behind most knowledge organization systems (KOS, the latest new buzzword) are built upon monolithic subjects. RDF has one “about” attribute. If someone puts something into a category, it’s often one category. In theory, one would classify things into every category they fit in. But this could get out of hand pretty quickly.

The world of library science hit this wall a long time ago. While their solution to this problem hasn’t been taken up entirely (mostly due to inertia and some rather nasty intellectual property battles), the core idea is called FacetedClassification (FC). In looking at FC, the idea of subject is not Platonic, it is divided like the surface of a gem into separate subjects (called “facets”), each of these facets a member in a facet hierarchy. That’s “classical” FC, which sadly isn’t formally defined anywhere, i.e., there are plenty of definitions. One might say the authority is the concept’s inventor, S.R. Ranganathan, but the world has moved on since he developed the system in the 1930s.

The idea that each facet has its own hierarchy comes from Ranganathan, who created a number of ancestor superclasses from which all facets (in his system) were derived. I disagree with this idea. It presupposes two layers: the layer of facets, and the hierarchy supporting those facets. It also presupposes his particular set of Platonic superclasses, which IMO is akin to religion or Ripley’s: you either believe it or not. My feeling is that the latter is an artificial and erroneous construct that ignores the complexities that each facet plays in its relation to other facets, and the recursive context (or role) that each facet plays in its relation with other facets. The classical view of FC has many of the same flaws as traditional knowledge representation systems that ignore context. And as pointed out by NicolasMontessuit, all of this also presupposes that both the subjects and the facets are somehow discrete, which is one of those a priori assumptions that FC doesn’t avoid, it just moves the atomicity down to the facet level. But I don’t mean here to criticise FC. I think that a hybridisation of FC and a context-based ontology comes close to representing the structure of knowledge, and this is the structure I’ve used in the design of my own system.

I’ll continue on with this later… – MurrayAltheim


See Also

See LinkLanguage, ReserveWiki


A title is not a subject. While Robert Brandom’s book may cover the subjects of Philosophy, Semantics, Inference, Reasoning,. Expression, Language and logic, it still only has one title: “Articulating Reasons”.

Wiki pages list relevant subjects by listing categories, and using a ReverseIndex to build up that subject (alternatively: making a ForwardIndex, using DublinCore, et cetera). The title is just an easy handle for reference.

Okay, fair enough. But I’m not clear on reading your references how the subjects (from either back or forward links) become explicitly available. Just by adding the WikiWord to the page so that the page is returned on a search? One of the situations that FC was designed to solve is a too-specific or too-general search. E.g., if you were searching on “dinosaur” you might not locate “stegosaurus”, or vice-versa. The other is that if any categorization becomes too full of entries, you can add additional facets to the entries to further differentiate them without altering the results of existing facet searches. Now, one of the things I’ve noticed is that while Dublin Core has been around for a fairly long time now, it’s been in the last year or so that its use has really caught on. I’m wondering if anyone has come up with any ways of allowing wiki writers to express subject hierarchies, which the LoC? makes a stab at. I didn’t find anything specific on how to use Dublin Core in linking. Putting subjects in invisible <meta> elements inaccessible to editing is probably not the solution, but perhaps some way of auto-generating metadata from the link targets (if they were explicitly marked as such) might work. I’ll have to check the FacetWiki to see what they’ve done. I see that you’re one of the contributors to that page, so obviously, I’ve got more reading to do… thanks!

DublinCore has no intrinsic way of creating hierarchies (or, indeed, lattices), but can still be used for this. If I tell my Wiki engine that <*, subject, stegosaurus> implies <*, subject, dinosaur>, that’s a hierarchy. (I might do that, say, by saying <stegosaurus, type, dinosaur>.)

Incidentally, the DublinCore page decribes a way of including DC in a Wiki page.

The title of a page is simply a semantic address; i.e. a word, part of vocabulary of the language of the wiki. – SunirShah

Hey Sunir. Yes, but what I’m interested in exploring is, without delving too deeply into the semiotics of words-as-symbols, the relationship between use of “normal” words within sentences of natural language, and the use of WikiWords, which seem to stand in as more profound identifiers for subjects. The reason I say that is that this paragraph is composed of words, but if I was to use a specific word, say FacetedClassification, I’ve now invoked a kind of “higher power” in my use of that word. In a sense, most every word we use can be looked up in a dictionary, where for most we find multiple definitions. The particular, appropriate definition for a given situation will be context-dependent, not Platonic (that’s all I meant by Platonic, i.e., universally or ideally true in all contexts).

WikiWords are special words, given their own web page. Topic Maps (and of course other knowledge representation schemes too) have a concept of subject identity, and I’m looking at the titles on wiki pages as more than simply titles, they seem to me to be really (coupled with the address of the WikiWord’s page, i.e., its URL) subject identifiers. As such, it seems very natural to create Topic Map Topics for them and treat them as subjects, by which I mean multi-faceted (FC) subjects. It becomes not a categorization problem, but a mapping problem; the categories are already existent in the content being mapped (there is no top-down). If all of this is applicable, we might have a very functional organizational scheme for wiki. (In reading FacetWiki I see some very related ideas, too.)

If I were to AttackTheDocumentMode, I’d start the page with something like:

PlatonicCategories are idealized categories…”

…and then continue with the discription you gave on (I think) LinkLanguage.

I’d talk about concern that people think that categories are real things, and the dangers in applying artificial categories too rigorously, as if they were “Platonic.”

Have you read CategoriesAndTopics? You might find that interesting. It’s something I’d like to try out locally, perhaps. (Seems like a lot of work. But, we could just StartInTheMiddle?.)

I wish there were some focus to what this page was saying.

“Platonic categories”, -or rather Platonic Ideas, for “category” is an aristotelician concept-, are sure one way of organizing this terrible mess that is called “universe”, but they are not the only one (this is obvious even from a purely philosophic viewpoint).

The use of hierarchies/data-trees/taxonomies is even the sole consequence of one very simple fact: that the data structures used to record the “semantic” values are discrete. For, if you represent any “thing” of the world in a discrete way, there is no mean to escape the basic logicist data structures (the most famous of which are the list, the chained list, the tree, the hash, etc.)

Now, I could develop this and explain plainly why, imho, discrete data structures are not sufficient to deal with all the semantic complexity of life, things, wikis, ”…women, fire and dangerous things” (this latter pun is the title of a Semantics book from the linguist Lakoff). Anyway, believe me for the moment, until I am more fluent on the question: there is a need to cope with continuous data, and even with hybrid data structures, that is data-structures that integrate both discrete and continuous components.

There is also a need to use special kinds of grammars that can handle continuous data (they exist, albeit they be very recent).

Suppose for example you describe a “semantic unit” of your wiki by a composite hybrid data structure which I assume completely continuous for clarity. Then this semantic unit is in fact some point in some (assumed topologic) space. Suppose too you have some grammar that can compute an open region of this semantic space for each phrase you give it. Then, you can define a mapping from the space of WikiWords to the space of semantic meanings (nearly, in fact to the set of subspaces of this space), and then your problem of naming the pages in a univoque way is solved (in a rather interesting way, because there’s no more need to retain a single title, and the search algorithm of your wiki has changed a lot).

For information, I am currently developing such a system on my personal wikiwiki. My provider has some troubles, but I will do the best to help this. More to be said next month. – esc

Nicolas, I believe we are in very strong agreement, certainly philosophically. There is a continuum of both time and matter, and the boundaries of both for what I call me are, as you say, continuous. Is my breath part of me while within the visible bounds of my body? What are the visible bounds of my body? If animal species don’t have firm boundaries, how do we reclassify zoological taxonomy rationally? My own project approaches this problem for authoring by using a hybrid of Faceted Classification, but as you say, it’s very difficult to get away from using discrete entities, especially in an object-oriented programming language. :-) I would be interested in any references you have available on grammars for representing continuous structures.

I looked at your home page on Meatball; we certainly share some interests, and look forward to further discussions, either here or elsewhere (email too), as well as seeing your wiki project as it goes online. Thanks.

Hi Murray. I’m sorry for this rather long period of mutism. Some non discrete trends of linguistics are found in neural networks and also a little in cognitive linguistics (see Langacker), even if this latter statement may seem a little strange. I have written some “continuous” code myself, for example this heuristic word-associator. It is a program that: 1) stock all the semantic data between words and their equivalents in several languages by using only two type of vertices in a non-oriented graph, and 2) represent a semantic unit by mean of a weighed subgraph (possibly weighed with real numbers). All the data are contained, not in the nodes of the graph, but in their connectivity, so there is nothing like a node indeed, each part of the graph must be interpreted socially and in the context of surrounding nodes. This is what I call “HolisticProgramming”. I have other programs which insist much more on the real values of semantic units.

((P.S.: the page tries to translate a word from french to english or from english to french, by considering 1°) a word possibly enriched with clues (ex: “organ”, “organ:music”, “organ:heart”, “organ:heart:lung”); 2°) a source language (either fr or en); 3°) a destination language (either en or fr)…

For instance, the french word “coeur” means either “core” (when akin to french idea of “centre”) either “heart” (when akin to french idea of “organe”), so if you enter “coeur:centre” and “fr->en” you will get “core”, and if you enter “coeur:organe” you will get “heart”.))

Please also note that the “clue notation” (which I invented) is a non-hierarchical way of getting rid with categories, whether they be platonic or not… – esc

NOTA: The script was in french, it is now in english too. – esc


Define external redirect: StartInTheMiddle LoC

EditNearLinks: SunirShah ReverseIndex NicolasMontessuit FacetWiki WikiWord WikiWords CategoriesAndTopics