Topic maps provide an approach that marries the best of several worlds, including those of traditional indexing, library science and knowledge representation, with advanced techniques of linking and addressing. It is our firm conviction that they will become as indispensable for tomorrow's information providers as maps for the traveller. And once topic maps have become ubiquitous, they will indeed constitute the GPS of the information universe.

The generality and expressive power of the topic map model bring with it other advantages that go far beyond those traditionally associated with indexes. The close similarity to semantic nets gives an idea of how topic maps, even without any occurrences connecting them to an information pool, can become valuable resources in their own right. This in turn opens up new business opportunities for creating and selling “portable topic maps” that can be overlaid on multiple information pools. For traditional commercial publishers, producing well-crafted topic maps could be a new way of leveraging their existing knowledge and experience and combating the threat to their existence posed by the vast amounts of information now available for free.

Exactly what one chooses to regard as topics in any particular application will vary according to the needs of the application, the nature of the information, and the uses to which the topic map will be put: In a thesaurus, topics would represent terms, meanings, and domains; in software documentation they might be functions, variables, objects, and methods; in legal publishing, laws, cases, courts, concepts, and commentators; in technical documentation, components, suppliers, procedures, error conditions, etc. Topic types are themselves defined as topics by the standard. You must explicitly declare “composer”, “opera”, “city”, etc. as topics in your topic map if you want to use them as types (in which case you will be able to say more about them using the topic map model itself).

A thesaurus, on the other hand, emphasizes other aspects of an index. It is basically a network of interrelated terms within a particular domain, and although it will often contain other information (such as definitions, examples of usage, etc.), the key feature of a thesaurus is the relationships, or associations, between terms. Given a particular term, a thesaurus will indicate which other terms mean the same, which terms denote a broader category of the same kind of thing, which denote a narrower category, and which are related in some other way.

An important point to note here is the separation into two layers of the topics and their occurrences. This separation is one of the clues to the power of topic maps and we shall return to it later. Occurrences, as we have already seen, may be of any number of different types (we gave the examples of “monograph”, “article”, “illustration”, “mention” and “commentary” above). Such distinctions are supported in the standard by the concepts of occurrence role and occurrence role type.

Any two topics that share one or more subject indicators (or that have the same subject address, in the case of addressable subjects) are considered to be semantically equivalent to a single topic that has the union of the characteristics (the names, occurrences and associations) of both topics. In a processed topic map a single topic node results from combining the characteristics of the two topics.[7] A subject indicator could be an official, publicly available document (for example, the ISO standard that defines 2- and 3-letter country codes), or it could simply be a definitional description within (or outside) one of the topic maps. A published subject indicator (PSI; originally called “public subject descriptor”) is a subject indicator that is published and maintained at an advertised address for the purpose of facilitating knowledge interchange and mergeability, either through topic maps or by other means. Published subjects are a necessary precondition for the widespread use of portable topic maps, since there is no point in offering a topic map to others if it is not guaranteed to “match up” with relevant occurrences in the receiver's pool of information resources.

A glossary is basically a list of terms and definitions. It can be thought of as a kind of index in which only one type of occurrence is of interest (the one that provides the “definition”), and which therefore includes the occurrence inline (instead of pointing to it via a locator).

A topic may be linked to one or more information resources that are deemed to be relevant to the topic in some way. Such resources are called occurrences of the topic. An occurrence could be a monograph devoted to a particular topic, for example, or an article about the topic in an encyclopaedia; it could be a picture or video depicting the topic, a simple mention of the topic in the context of something else, a commentary on the topic (if the topic were a law, say), or any of a host of other forms in which an information resource might have some relevance to the subject in question.

The goal with topic maps is to achieve a one-to-one relationship between topics and the subjects that they represent, in order to ensure that all knowledge about a particular subject can be accessed via a single topic. However, sometimes the same subject is represented by more than one topic, especially when two topic maps are being merged. In such a situation it is necessary to have some way of establishing the identity between seemingly disparate topics. For example, if reference works publishers from Norway, France and Germany were to merge their topic maps, there would be a need to be able to assert that the topics “Italia”, “l'Italie” and “Italien” all refer to the same subject. The concept that enables this is that of subject identity. When the subject is an addressable information resource (an “addressable subject”), its identity may be established directly through its address. However most subjects, such as Puccini, Italy, or the concept of opera, are not directly addressable. This problem is solved through the use of subject indicators (originally called “subject descriptors” in ISO 13250). A subject indicator is “a resource that is intended ... to provide a positive, unambiguous indication of the identity of a subject.” Because it is a resource, a subject indicator has an address (usually a URI) that can be used as a “subject identifier”.

Someone once said that “a book without an index is like a country without a map”. However interesting and worthwhile the experience of driving from A to B without a map might be in its own right, there can be no doubt that when the goal is to arrive at one's destination as quickly as possible (or at least without undue delay), some kind of a map is indispensable. Similarly, if you are looking for a particular piece of information in a book (as opposed to enjoying the experience of reading it from cover to cover), a good index is an immense asset. The traditional back-of-book index can be likened to a carefully researched and hand-crafted map, and the task of the indexer, as Larry Bonura puts it [Bonura 1994], “to chart[ing] the topics of the document and [presenting] a concise and accurate map for readers.”

As mentioned above, scope should not be confused with facets. The two mechanisms are different and complementary. Whereas scope can be seen as a filtering mechanism that is based on properties of the topics, facets provide for filtering based on properties of the information resources themselves.[10]

Topics can be categorized according to their kind. In a topic map, any given topic is an instance of zero or more topic types. This corresponds to the categorization inherent in the use of multiple indexes in a book (index of names, index of works, index of places, etc.), and to the use of typographic and other conventions to distinguish different types of topics.

Indexes, glossaries and thesauri are all ways of mapping the knowledge structures that exist implicitly in books and other sources of information. In the field of AI (Artificial Intelligence) there also exists the need to be able to represent knowledge (and meaning), in order to support communication between people and machines. One widely used knowledge representation formalism is that of conceptual graphs, whose building blocks are concepts and conceptual relations. In the following conceptual graph for the phrase “man biting dog” ([Sowa 1984]), square brackets denote concepts ('man', 'bite', 'dog'), and parentheses denote relations ('agent', 'object'): [man] <- (agent) <- [bite] -> (object) -> [dog] Similar graph structures have been implemented in various forms under names such as “semantic nets”, “associative nets”, “partioned nets” and “knowledge” (or “conceptual”) “maps” in many AI systems. The earliest forms, called existential graphs, were invented by the philosopher Charles Sanders Peirce at the end of the 19th century as a graphical notation for symbolic logic. One of the most completely worked out schemes, the conceptual graphs developed by John Sowa and his collaborators ([Sowa 2000]), is claimed to be completely isomorphic with first order logic.

Just as topics and occurrences can be grouped according to type (e.g., composer/opera/country and mention/article/commentary, respectively), so too can associations between topics be grouped according to their type. The association type for the relationships mentioned above are written_by, takes_place_in, born_in, is_in (or geographical containment), and influenced_by. As with most other constructs in the topic map standard, association types are themselves defined in terms of topics.

The special thing about associations in a thesaurus (as compared to associations found in a typical index or glossary) is that they are typed. This is important because it makes it possible not only to say that two terms are related, but also how or why they are related. It also makes it possible to group together terms that are associated in the same way, thus making navigation much easier. Commonly used association types like “broader term”, “narrower term”, “used for” and “related term” are defined in standards for thesauri such as [Z39.19], [ISO 5964] and [ISO 2788].

The really interesting thing, however, is to be able to describe relationships between topics, and for this the topic map standard provides a construct called the topic association. A topic association asserts a relationship between two or more topics. Examples might be as follows: “Tosca was written by Puccini” “Tosca takes place in Rome” “Puccini was born in Lucca” “Lucca is in Italy” “Puccini was influenced by Verdi”

The word “index” means many things, most of which relate to pointing in some way (“index”, plural “indicis”, is Latin for forefinger, informer, or sign). The sense we are interested in is that given in the Concise Oxford Dictionary as: 6. Alphabetical list, usu. at end of book, of names, subjects, etc., with references; A traditional index is in fact a map of the knowledge contained in a book; it lists the topics covered, by whatever name users might be expected to want to look them up, and includes salient (and only salient) references to those topics.

The standard therefore provides the facility to assign multiple base names to a single topic, and to provide variants of each base name for use in specific processing contexts. In the original ISO standard variants were limited to display name and sort name. XTM offers a more general variant name mechanism.

Each topic that participates in an association plays a role in that association called the association role. In the case of the relationship “Puccini was born in Lucca”, expressed by the association between Puccini and Lucca, those roles might be “person” and “place”; for “Tosca was composed by Puccini” they might be “opera” and “composer”. It will come as no surprise now to learn that association roles can also be typed and that the type of an association role is also a topic!

Topic maps are a new ISO standard for describing knowledge structures and associating them with information resources. As such they constitute an enabling technology for knowledge management. Dubbed “the GPS of the information universe”, topic maps are also destined to provide powerful new ways of navigating large and interconnected corpora.

Assignments of topic characteristics are always made within a specific context, which may or may not be explicit. For example, if I (yet again) mention “tosca”, I should expect my readers to think of the opera by Puccini (or its principle character), because of the context that has been set by the examples used so far in this paper. For an audience of bakers, however, the name “tosca” has quite other and sweeter connotations: it denotes another topic altogether. Although we seldom notice it in everyday life, the problem of context is with us all the time. According to [Sowa 1984] a sentence is derived from six different kinds of information, four of which ( tense and modality; presupposition; focus; and emotional connotations) are in one way or another related to context. Humans are remarkably good at dealing with context. It is that ability that enables them to make sense of two such similar statements as John Smith to marry Mary Jones on the one hand, and Retired priest to marry Bruce Springsteen on the other, or to parse and interpret the two sentences Time flies like an arrow and Fruit flies like an apple.[9] Computers, however, are not yet that smart. Given two such simple statements as Tosca takes place in Rome and Tosca kills Scarpia, most of today's computers would not be able to infer which of the topics named “Tosca” was involved. In order to avoid this kind of problem, topic maps consider any assignment of a characteristic to a topic, be it a name, an occurrence or a role, to be valid within certain limits, which may or may not be specified explicitly. The limit of validity of such an assignment is called its scope. Scope is defined in terms of themes, and a theme is defined as “a member of the set of topics used to specify a scope”. In other words, a theme is a topic that is used to limit the validity of a set of assignments. Thus, the name “tosca” might be assigned to three different topics in scopes defined by the themes “opera”, “opera”+“character”, and “baking” respectively, thereby removing any ambiguity and reducing the chance of errors, for example when merging topic maps. In fact, the well-designed, consistent and imaginative use of scope in topic maps does much more than simply remove ambiguity. It can also aid navigation, for example by dynamically altering the view on a topic map based on the user profile and the way in which the map is used. For example, any user that declares a specific interest in opera (or a specific lack of interest in baking!) can have the various toscas ranked accordingly.

What then is a topic? A topic, in its most generic sense, can be any “thing” whatsoever — a person, an entity, a concept, really anything — regardless of whether it exists or has any other specific characteristics, about which anything whatsoever may be asserted by any means whatsoever. You can't get much more general than that! In fact, this is almost word for word how the topic map standard defines subject, the term used for the real world “thing” that the topic itself stands in for. We might think of a “subject” as corresponding to what Plato called an idea. A topic, on the other hand, is like the shadow that the idea casts on the wall of Plato's cave: It is an object within a topic map that represents a subject. In the words of the standard: “The invisible heart of every topic link is the subject that its author had in mind when it was created. In some sense, a topic reifies a subject...” Strictly speaking, the term “topic” refers to the object or node in the topic map that represents the subject being referred to. However, there is (or should be) a one-to-one relationship between topics and subjects, with every topic representing a single subject and every subject being represented by just one topic. To a certain degree, therefore, the two terms can be used interchangeably.[3]

Sometimes it is convenient to be able to assign metadata to the information resources that constitute the occurrences of a topic from within the topic map. To provide this capability, the standard includes the concept of the facet. Facets basically provide a mechanism for assigning property-value pairs to information resources. A facet is simply a property; its values are called facet values. Facets are typically used for supplying the kind of metadata that might otherwise have been provided by SGML or XML attributes, or by a document management system. This could include properties such as “language”, “security”, “applicability”, “user level”, “online/offline”, etc. Once such properties have been assigned, they can be used to create query filters producing restricted subsets of resources, for example those whose language is “Italian” and user level is “secondary school student”.

Note: Once the distinction between addressable and non-addressable subjects had been clarified in XTM, it became clear that information resources could also be subjects (and hence topics). This rendered the concept of facets superfluous since metadata properties can now be assigned to a resource as characteristics of the topic that represents that resource. As a consequence, facets are not part of XML Topic Maps.

“Knowledge management” is of course one of today's buzzwords and a term that often involves not a little marketing hype. For the big consulting companies, knowledge management is essentially about new business management techniques designed to address the fact that people (and the expertise they possess) are the primary assets in an increasingly knowledge-based economy. Others equate knowledge management with information management (especially some vendors of information management tools, who are only too happy to slap a new label on their boxes). But knowledge is fundamentally different from information: the difference is that between knowing a thing versus simply having information about it. And if, as one writer claims ([Ruggles 1997]), “knowledge management covers three main knowledge activities: generation, codification, and transfer”, then topic maps can be regarded as the standard for codification that is the necessary prerequisite for the development of tools that assist in the generation and transfer of knowledge.

The key features of a typical index are thus: topics (identified by their names, of which there may be more than one); associations between topics; and occurrences of topics (pointed to via locators). For each of these constructs it is useful to be able to say something about the type, in order to convey more information to the user. Topics, Associations and Occurrences are also the key constructs in the topic map model

Normally topics have explicit names, since that makes them easier to talk about.[4] However, topics don't always have names: A simple cross reference, such as “see page 97”, is considered to be a link to a topic that has no (explicit) name.

It is also important to note that while both topic associations and normal cross references are hyperlinks, they are very different creatures: In a cross reference, the anchors (or end points) of the hyperlink occur within the information resources (although the link itself might be outside them); with topic associations, we are talking about links (between topics) that are completely independent of whatever information resources may or may not exist or be considered as occurrences of those topics. Why is this important? Because it means that topic maps are information assets in their own right, irrespective of whether they are actually connected to any information resources or not. The knowledge that Rome is in Italy, that Tosca was written by Puccini and is set in Rome, etc. etc. is useful and valuable, whether or not we have information resources that actually pertain to any of these topics. Also, because of the separation between the information resources and the topic map, the same topic map can be overlaid on different pools of information, just as different topic maps can be overlaid on the same pool of information to provide different “views” to different users. Furthermore, this separation provides the potential to be able to interchange topic maps among publishers and to merge one or more topic maps.[6]