Topic Map technology - the state of the art   Table of contents   Indexes   Creating semantically valid topic maps
 ISO/IEC 13250 
 associations 
class hierarchies
consistency constraints
 indexing  
 knowledge representation 
 navigation 
relation properties
relation types
 semantic networks 
topic map template
topic map validation
 topic maps 
 

Making topic maps more colourful

Rath, Hans Holger
 
 Hans Holger  Rath
 Director Consulting
  Germany 
 Rimpar 
 STEP Electronic Publishing Solutions GmbH 
STEP Electronic Publishing Solutions GmbH,  Technologiepark Würzburg-Rimpar
Pavillon 7
 D-97222 Rimpar  Germany
Phone: +49.9365.8062.0 Fax: +49.9365.8062.66 email: consulting@step.de web site: www.topicmaps.com/
 Biography
 Hans Holger Rath — Hans Holger Rath is director of STEP's Consulting department since April 1998. He started at STEP in April 1996 as senior consultant/project manager. Before he joined STEP he was head of the Document Computing department at ZGDV (Computer Graphics Center, Darmstadt, Germany). Hans Holger studied computer science and graduated with the doctoral thesis "Literate Specifying of Hypermedia Documents" in 1996. He cooperates very closely with publishing houses, aircraft industry and telecommunication industry. All in all he has more than ten years experience in information architectures and related topics. Since May 1998 he represents Germany in ISO/JTC1/SC34 the ISO committee standardizing SGML, HyTime, DSSSL, Topic Maps etc.
 Abstract
 The new ISO standard ISO/IEC 13250 Topic Maps defines a model and architecture for the semantic structuring of link networks. Dubbed the “GPS of the information universe”, topic maps will become the solution for organizing and navigating large and continuously growing information pools, and provide a “bridge” between the domains of knowledge representation and information management. This paper presents several technical issues of which are of great interest when applying topic maps to real world applications. The main focus of the paper is the introduction of “topic map templates” – a semi-official term coined by the standards' committee for a concept that the authors argue is a necessary but as yet unstandardized addition to the basic model. Furthermore association taxonomies, class hierarchies, and consistency constraints of topic maps are presented and discussed.
 

Introduction

 ISO  
 		TheISO committee JTC 1/SC 34/WG 3Information Technology – Document Description and Processing Languages – Information Association standardized ISO/IEC 13250 Topic Maps in the autumn of 1999. Formally speaking, the ISO standard defines a model and interchange syntax for Topic Maps. The initial ideas – which date back to the early 1990's – related to the desire to model intelligent electronic indexes in order to be able to merge them automatically. But during several years of gestation, the topic map model has developed into something much more powerful that is no longer restricted to simply modelling indexes.
 GPS, Global Positioning System 
 		A topic map annotates and provides organising principles for large sets of information resources. It builds a structured semantic link network above those resources. The network allows easy and selective navigation to the requested information. Topic maps are the “GPS of the information universe”. Searching in a topic map can be compared to searching in knowledge structures. In fact, topic maps are a base technology for knowledge representation and knowledge management.
 The basic concepts of the standard are topics, occurrences of topics, and relationships (“associations”) between topics. Section gives a short overview.
 The editors of the standard, together with the other members of ISO JTC1/SC34/WG3 (the authors are among those “other members”), defined a well-considered and implementable set of concepts. But first prototypes of practical applications show that there are a number of issues that are not covered by the standard. This was only to be expected since the working group considered it more important to publish a base standard immediately than to delay publication in order to add further refinements. Section discusses some of the concepts that the standard does not cover explicitly and explains why they are important for practical applications.
 SGML and XML have DTDs defining classes of instances, but topic maps as currently specified do not have an equivalent construct. The standards working group has recognised this need and coined the termtopic map template for the “declarative part” of a map. Section explains what makes up a template.
 Three other additional concepts are also discussed:
 
  • a taxonomy of the basic properties of topic associations (section ),
    •  
  • class (or type) hierarchies and how they can be exploited in topic map software (section ), and
    •  
  • consistency checking and validity constraints for topic maps ( ).
    •  The section summarizes the paper and gives an outlook on further topic map developments.
     

    Topic maps in a nutshell

     SGML 
         		The standard defines an interchange representation of topic maps defined in terms of anSGML architecture . A topic map is basically an SGML (orXML ) document in which different element types, derived from a basic set of architectural forms, are used to represent topics, occurrences of topics, and relationships (associations) between topics. The key concepts are thetopic (andtopic type ), thetopic occurrence (andoccurrence role type ), and thetopic association (andassociation type as well asassociation role type ). Other concepts which extend the expressive power of the topic map model are those ofscope ,theme ,public subject andfacet .
     XML 
         		Note:
     This short overview about topic map concepts provides the basics only. Application examples can be found in .
     

    Topics

     Atopic , 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.
     With the words of the standard, the term “topic” refers to the element in the topic map instance (thetopic link ) that represents the subject being referred to.
     Examples of topics are : USA, Pennsylvania, Philadelphia, William Penn.
     A topic should have one or moretopic types . Topic types are a typical class-instance relation and they are themselves defined as topics by the standard. Having topic types as topics the expressive power of topic maps is used to say more about the type.
     Examples of topic types are : country, state, city, person.
     

    Topic characteristics

     Every topic has two characteristics (or at least one of them): atopic name and anoccurrence .
     The topic name consists of three parts: thebase name , thedisplay name , and thesort name . Only the base name is required.
     Examples of topic names (base / display / sort) are : U.S.A. / USA / United States of America.
     An occurrence is a link to an information resource that is somehow relevant to the topic. The linked resource is typically an information object outside the topic map.
     Examples of occurrences are : chart of the USA, article about Pennsylvania, video about Philadelphia, portrait of William Penn.
     Every occurrence belongs to oneoccurrence role type . Occurrence role types are – as topic types – themselves topics.
     Examples of occurrence role types are : chart, article, video, portrait.
     

    Associations

     The real power of topic maps results fromassociations between topics.
     Examples of associations are : Pennsylvania is in USA, Philadelphia is in Pennsylvania, Philadelphia was founded by William Penn.
     Each association has oneassociation type .
     Examples of association types are : is in, was founded by.
     Each topic that participates in an association plays a role. The role is described by anassociation role type .
     Examples of association role types are : state / country, city / state, city / person.
     Both association types and role types are again topics.
     

    Scopes

     The concept ofscope is important to avoid ambiguities between topics and their characteristics. Any assignment of a characteristic to a topic is considered 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. A scope is defined in terms ofthemes and themes are topics.
     Examples of scopes are : to distinguish between “Paris” in France and “Paris” in Texas assign the scopes “France” and “USA” to the two topics.
     

    Identity

     Merging of topic maps requires a way of establishing the identity between seemingly disparate topics from different maps. The specification ofidentity attributes on the topic elements that address the samepublic subject is the explicit solution the standard offers. The other solution is implicitly through thetopic naming constraint which states that any topics that have the same name in the same scope refer to the same subject.
     

    Facets

     Facets provide a mechanism for assigning property-value pairs to information resources without modifying them. A facet is a property; its values are calledfacet values .
     

    The missing pieces: an overview

     During the years of its gestation the topic map model changed many times – from an extremely high level of generality to much more specific models designed to be used solely for navigation. The final result is – as most standards – a compromise. The working group believes that it offers an optimal balance between extreme power and flexibility on the one hand, and sufficiently well-defined semantics on the other.
     W3C 
         		The members of the working group always had in mind that the standard has to be implementable, and they tended towards a more general model because of both implementability and applicability reasons. They knew that first practical applications might uncover concepts which are not explicitly described in the standard, but they felt it was more important to have a base standard approved and published than to delay publication any longer merely to add further refinements. Adapting the standard to the XPointer (or XPath) addressing format – as soon as it becomes aW3C recommendation – is already on the agenda of the working group.
     The STEP Group
     The STEP Group consists of STEP Electronic Publishing Solutions GmbH (Rimpar, Germany), STEP Infotek AS (Oslo, Norway), STEP Electronic Publishing Kft (Budapest, Hungary), STEP Poland Ltd. (Warsaw, Poland), and STEP-DPSL Ltd. (Swindon, UK).
    started investigating topic map applications in autumn 1998 in the context of reference works (especially encyclopedias and dictionaries). Applying topic maps to encyclopedias is quite natural: Topic maps model knowledge structures and lexicons represent large parts of the “knowledge” of society. Thus this application field is a perfect candidate for detecting shortcomings and finding improvements.
     

    Separating the declarative part

     Topic maps are a well-designed standard for modelling semantic information networks. It defines the basic concepts and almost everything in the map is itself a topic. Even the “objects” declaring a topic map are topics, namely themes, topic types, occurrence role types, association types, and association role types. Having such recursive declarations makes perfect sense when the goals are to limit the concepts to a sensible minimum and make topic maps self-contained and self-documenting.
     But the standard does not provide a name or definition for the list of declarative “objects” of a map and this can lead to some confusion: Users often mix up “declarative” topics and “regular” topics during discussions. In addition to that, the different tasks of topic map design, creation, and maintenance are hard to distinguish and to separate. The same is true for user access rights: As long there is no distinction, different rights cannot be assigned to the map.
     The separate declarative part could also be used for defining classes of topic maps that share a common set of topics for types with predefined semantics.
     The standard therefore stands in need of a formally defined construct that covers the declarative part of a topic map.
     

    Applying theoretical background

     The most interesting constructs in topic maps as far as representing knowledge structures is concerned are associations. Because these are in fact relations it makes sense to take a look at mathematics and apply some of the theoretical background of relations. Furthermore the scientific fields of linguistics and philosophy may provide additional taxonomies.
     The concepts that we find could lead to predefined basic association types and association properties. Neither of these are covered by the standard today, but they could offer much more precise semantics in the maps. The topic map template will be the ideal place to define them.
     

    Class-instance relation is not enough

     All topics, occurrences, and associations can be seen as instances of classes (types). The classes themselves are expressed as topics.
     NB: The recursion “a topic has a type which is a topic which has a type” stops if no type is assigned. This is possible because the type is an optional attribute of the topic, occurrence, and association. If the attribute is not specified, the meaning is that the “object” has no more specific type (i.e. belongs to no more specific class) than that of the base class to which it belongs (“topic”, “occurrence”, or “association”, respectively).
     This class-instance relationship is in fact merely a syntactically privileged association type, as the standard makes clear:
     
     The class-instance relationship ... could alternatively be established by a topic association link whose semantic is the relationship between a class and an instance of that class.
     This means that the class-instance relation is an association type predefined by the standard. Any topic map software has to support it as a built-in function, e.g. by displaying the name of the referenced topic as the name of the type.
     If we are looking at the class-instance relation from an object oriented view, then there is a justifiable demand for a superclass-subclass relationship as well. However, the standard explicitly declares that such a relationship has to be user-defined. Here are the relevant quotes:
     
     The topic relationships established by the types attribute are not superclass-subclass relationships. They are only class-instance relationships.
     Superclass-subclass relationships between topics can be asserted by topic association links that have been user-defined for that purpose.
     STEP's experiences made with the encyclopedia applications show that the superclass-subclass relationship is a very powerful mechanism for performing inferencing, i.e. deriving implicit information about the current “object”. The implicit information can be used when querying the map or when declaring and/or checking consistency constraints. And because these features should be an integral part of a topic map software a user-defined and therefore application-specific solution is too weak.
     

    Questions of consistency

     The standard has almost nothing to say on the subject of validation and consistency. The “Conformance” section of the standard focuses on the understanding of the defined constructs, the interchange syntax, and import/export of topic maps. But nothing more, as this excerpt from the standard shows:
     
     This International Standard constrains neither the uses to which topic maps can be put, nor the character of the processing that may be applied by a conforming application.
     A topic map author (or authoring team) needs system support when developing a map with millions of topics and associations. The question of the consistency of the map becomes a key issue, because it is nearly impossible to check a map of that size manually.
     For that reason we need concepts to declare consistency constraints and to validate that those constraints have been obeyed.
     

    Topic map templates

     The ISO working group has already responded to the need to be able to separate the declarative part of a topic map. It coined the termtopic map template for a topic map that only consists of topics that are declared in order to be used as types in a class of topic maps. At the present time this term is only “semi-official”, since the concept has not yet been refined and added to the standard.
     
    Topic map template
     

    What is a topic map template?

     A topic map template consists of all constructs which have a declarative meaning for the map (see ). These are all the topics used as themes and as types for
     
  • other "regular" topics,
    •  
  • occurrence roles,
    •  
  • associations,
    •  
  • association roles,
    •  
  • facets, and
    •  
  • facet values.
    •  As we will see later, the class hierarchy information and consistency constraints will also become part of a topic map template.
     The topic map designer shall mark the topics in the template for which kind of type they could be used in the "real" map. This can be done by either grouping the topics (see below ) or by assigning attribute values. The latter approach provides more flexibility for marking topics for more than one kind of type.
     In any case it is clearly important that the topics of the template can be distinguished somehow from the topics of the topic map instance(s) belonging to the class of topic maps defined by the template, and that the template becomes a "manageable" object with its own (public) identifier, owner, version number, etc.
     

    Using templates in topic maps

     The topic map template – which is a topic map – can be copied into or referenced by another topic map.
     The copied template acts as a starting point for a new map containing all the themes and types which will be extended during the further development of the map.
     The referenced template provides the basic themes and types which are used by the referencing map. A referenced template makes use of the merging features of topic maps defined by the standard. Thus more than one template could be referenced. Though the precondition for merging is the existence of carefully worded subject identities.
     

    Template modules

     It might be meaningful that a template consists of sub-templates to modularize the design. Candidates for template modules are
     
  • clusters of all “typing” topics for the various “objects” as listed above, e.g. all topics which shall be used as topic types,
    •  
  • the class hierarchy information, or
    •  
  • the consistency constraints.
    •  But this is only one possibility. How the declarations will be clustered in modules depends to a large degree on the application specific requirements. The only important thing is that the template can easily identified and separated from the real map.
     

    Distributing the design and creation tasks

     The design and creation of topic maps can now be split up into subtasks because of the availability of templates and template modules. Furthermore, user access rights of user groups as well as roles can be assigned.
     The tasks of the designer might be:
     
  • declaration of themes,
    •  
  • declaration of all topics which are candidates for types,
    •  
  • marking the topics with the kind(s) of type it is intended for,
    •  
  • defining the consistency constraints.
    •  The tasks for the editor might be:
     
  • definition of the “real” topics,
    •  
  • definition of associations between them,
    •  
  • establishing the occurrence links to the relevant information objects,
    •  
  • checking the consistency of the map by applying the consistency constraints (this will be an automatic process).
    •  The assignment of facets can be seen as a completely separate task.
     

    Role of topic map templates for ISO/IEC 13250

     The concept of templates offers the ISO working group the possibility of defining various templates which are specific for different application areas. These templates would contain built-in types (i.e. topics) and association types with predefined semantics which could be supported by “template-conformant” applications.
     Such templates could be published as annexes to the standard or as separate standards, as has already been done with SGML DTDs (e.g. ISO 12083).
     

    Association taxonomy

     The investigation of the theoretical backgrounds of relations leads us to the domains of mathematics, linguistics, artificial intelligence, and philosophy. All these scientific fields deal with knowledge representation and knowledge structures in one way or another.
     We will concentrate on two issues from this broad research area: relations in mathematics (i.e. the abstract properties of associations) and relationship types in artificial intelligence and linguistics (i.e. specific classes of associations).
     

    Association properties

     The most important relations – in the mathematical sense – are thebinary relations .
     N-ary relations and “elementary associations” (in which the number of arguments cannot be further reduced) with more than two arguments are not covered in this paper, because they form a more complex class.
     Definition: Abinary relation between the sets A and B is: every subsetR ofA ×B (R ⊆A ×B ).
     The properties which are of interest for topic maps are only effective for a restricted kind of relations.
     Definition: Abinary relation in M is: a binary relationR withA =B =M , thusR ⊆M ×M .
     A binary relation is also a binary predicate.
     Definition: Apredicate (relation) R is fulfilled (true) forx ∊A andy ∊B  (x ,y ) ∊R .
     (x ,y ) ∊R can be abbreviated asxRy .
     Now we can define the properties for relations inM .
     
    Property ofR Definition
    reflexive ∀x ∊M :xRx
    symmetric ∀x ,y ∊M :xRy ⇒yRx
    transitive ∀x ,y ,z ∊M :xRy ∧yRz ⇒xRz
    anti-reflexive ∀x ∊M : ¬xRx
    anti-symmetric ∀x ,y ∊M ,x ≠y :xRy ⇒ ¬yRx
    connex ∀x ,y ∊M :xRy ∨yRx
     Certain combinations of these properties define special classes of relations, of which there are four:
     Definitions:
     
  • R is anequivalence relation :R is reflexive, symmetric, and transitive.
    •  
  • R is anpartial ordering relation :R is reflexive, anti-symmetric, and transitive.
    •  
  • R is atotal order relation :R is reflexive, anti-symmetric, transitive, and connex.
    •  
  • R is astrong order relation :R is anti-reflexive, anti-symmetric, and transitive.
    •  Some examples of specific relations will serve to illustrate the various properties and classes of relations (M = {0, 1, 2, 3, ...}).
     
    relation examples
    Property / class is denominator of is less than equal is less than
    reflexive yes yes no
    symmetric no no no
    transitive yes yes yes
    anti-reflexive no no yes
    anti-symmetric yes yes yes
    connex no yes no
    order rel. yes yes no
    total order rel. no yes no
    strong order rel. no no yes
     Why is all the theory relevant for topic maps? Let us analyze the association type “geographical_objectis in geographical_object”. It is transitive, anti-reflexive, and anti-symmetric; thus it is a strong order relation. Topic map software that was aware of these facts (i.e. the properties of this prarticular association type) would be capable of automatically deriving implicit knowledge from the map.
     An example: From the given associations
     
  • Pennsylvaniais in USA
    •  
  • Philadelphiais in Pennsylvania
    •  
  • Pittsburghis in Pennsylvania
    • the topic map software can derive that
     
  • Philadelphiais in USA
    •  
  • Pittsburghis in USA
    •  
  • USAis not in Pennsylvania
    •  
  • Philadelphiais not in Philadelphia
    •  
  • etc.
    •  It is obvious that the most informative statements of this example derive from the property of transitivity.
     Another example: Let us analyze the association type “streetis parallel to street”. It is reflexive, symmetric, and transitive; thus it is an equivalence relation.
     If we have the associations
     
  • Park Avenueis parallel to Madison Avenue
    •  
  • Madison Avenueis parallel to Fifth Avenue
    • then the associations
     
  • Park Avenueis parallel to Fifth Avenue
    •  
  • Fifth Avenueis parallel to Madison Avenue
    •  
  • etc.
    • can easily be derived. The relevant information comes from the symmetry and again from the transitivity property.
     The examples show that a simple set of association properties, i.e. the relation properties introduced above, would give more “knowledge” from the topic map than explicitly coded in it. This means that the map becomes smaller, that the effort creating a map will be minimized, that possible coding errors will be reduced tremendously, and that the inferencing capabilities of the topic map's query engine will be greatly enhanced. Furthermore the consistency checking can make use of the property information, which again improves the quality of the map.
     

    Basic association types

     The previous section introduced the basic association properties. This section investigates if also basic association types would make sense.
     Steve Pepper (STEP Infotek, Norway) provided substantial input to this section.
     A lot of research has been done in the area of knowledge structures
     See for an introduction and extensive bibliography.
    . Some of the research work covers relations in the lexicon . The results are a summary of relations that express the basics concepts of knowledge representation.
     A large class is comprised of thepart-whole orholonymy/meronymy relations. list six and seven subclasses of holonymy respectively:
     
  • component-object (e.g. branch/tree)
    •  
  • member-collection (e.g. tree/forest)
    •  
  • portion-mass (e.g. slice/cake)
    •  
  • stuff-object (e.g. aluminum/airplane)
    •  
  • feature-activity (e.g. paying/shopping)
    •  
  • place-area (e.g. Philadelphia/Pennsylvania)
    •  
  • phase-process (e.g. adolescence/growing up)
    •  Iris et al reduce this to four basic subclasses:
     
  • functional-part (← phase-process, feature-activity)
    •  
  • segmented-part (← component-object, place-area)
    •  
  • collection-member (← member-collection, stuff-object)
    •  
  • subset (← portion-mass)
    •  According to onlysegmented-part andsubset exhibit transitivity. Individualfunctional-part orcollection-member relations could be transitive, but the property does not apply to these classes as a whole.
     We can conclude that thepart-whole class with its subclassesfunctional-part ,segmented-part ,collection-member , andsubset shall be predefined association types – declared in a template.
     Some other relevant relationship types are
     
  • synonymy (e.g. equals, identical to),
    •  
  • similarity (e.g. similar to),
    •  
  • order (e.g. less than, older than, closer to),
    •  
  • result-agent (e.g. “object” is caused by “agent”, “artwork” created by “artist”, “painting” painted by “painter”),
    •  
  • tool-agent (e.g. “tool” is used by “agent”, “instrument” is played by “musician”), and
    •  
  • strict implication
     Definition ofstrict implication : A propositionP entails a propositionQ (P ⇒Q ) if and only if there is no conceivable state of affairs that could makeP true andQ false.
    (e.g. “activity 1” implies “activity 2”, ”snoring” implies “sleeping”).
    •  Thesynonymy ,order , andstrict implication are transitive relations.Synonymy andsimilarity are symmetric. For everyresult-agent andtool-agent relation exists an inverse one (“agent” causes “object”, “agent” uses “tool”). Strict implication is non-symmetrical: you can sleep without snoring, but you cannot snore without sleeping! All these relations are candidates to be predefined association types that are declared in a template.
     The contributions from linguistics introduce further subclasses forsynonymy relations (thesauri: ). Bothhyponymy andtroponymy represent the “is a” or “is a kind of” relation, which is already covered by the topic type construct. Thesynonymy subclasses seemed to be quite specific, thus there is no need to have them as predefined association types. They are in any case more appropriately handled through the use of multiple topic names.
     

    Class hierarchies

     The realisation of the need for class hierarchies stems from STEP's encyclopedia projects. A topic map for a lexicon contains a very large number of topics (typical orders of magnitude are hundreds of thousands or millions) and associations (even more). But most of the topic, association, and occurrence role types can be reduced to a small number of “super-types” – as we have already seen in the previous section.
     

    Superclass-subclass

     The superclass-subclass relationship of topic types, association types, and occurrence role types go hand in hand, as following examples shows:
     
  • Topic types: (person) → (artist, ...) → (painter, sculptor, writer, poet, composer, ...); (object) → (artwork, ...) → (painting, sculpture, novel, poem, opera, ...)
    •  
  • Association types and occurrence role types: (object “was caused by” person) → (artwork “was created by” artist) → (opera “was composed by” composer)
    •  

    Class hierarchy and association type properties

     The class hierarchies become even more important when the end-user navigates or queries the map. If someone would like to know “Which pieces of music were composed by Germans that were influenced by W.A. Mozart?”, it is very likely that this information is not exactly part of the map. But with just a few topics, transitive associations, and a class hierarchy the answer can be found very easily.
     The facts of the map:
     
  • The topic type (class) hierarchies: person → composer; piece of music → opera; geographical object → country; geographical object → city.
    •  
  • The transitive association type: “geographical object” is in “geographical object”.
    •  
  • Other association types: “composer” has composed “piece of music”; “person” was influenced by “person”; “person” was born in “geographical object”.
    •  
  • The topics: W.A. Mozart (composer); R. Wagner (composer); L. van Beethoven (composer); Bonn (city); Leipzig (city); Germany (country); Lohengrin (opera).
    •  
  • The associations: Bonn is in Germany; Leipzig is in Germany; L. van Beethoven was born in Bonn; R. Wagner was born in Leipzig; Lohengrin was composed by R. Wagner; R. Wagner was influenced by W.A. Mozart.
    •  The algorithm how the topic map software would find the solution with these facts could work as follows:
     
  • Extraction of the known topics from the query: Germany, W.A. Mozart.
    •  
  • Extraction of the types of the unknown topics: person (X ), piece of music (Y ).
    •  
  • Extraction of the association types: born in, influenced by, composed by.
    •  
  • Finding the missing topics using the associations and class hierarchies:
    X is born in Germany (country is also a geographical object) ⇒X is born in Bonn or Leipzig (both cities are in Germany) ⇒X is L. van Beethoven or R. Wagner (both composers are also persons);
    X was influenced by W.A. Mozart (composer is also a person) ⇒ R. Wagner was influenced by W.A. Mozart (both composers are also persons) ⇒X is R. Wagner;
    Y was composed byX ⇒Y was composed by R. Wagner ⇒ Lohengrin was composed by R. Wagner (opera is also piece of music) ⇒Y is Lohengrin.
    •  This very simple example shows the power of combining class hierarchies with properties of association types (here transitivity). As already stated above, both class hierarchies and association type properties are the basis for compact topic maps, minimized creation and maintenance efforts, and a reduction of coding errors.
     This supports our contention that the concept of class hierarchies should be a predefined association type of topic map template ensuring the correct built-in interpretation by the topic map software.
     

    Validation of consistency

     All the previously introduced concepts extend topic maps in ways that increase their expressive power and ease creation and maintenance efforts. In addition to this, the topic map developer wants to have something at hand to help ensure the quality of the map. The information provided by a topic map based on the standard architecture is not enough – the developer asks for validation concepts.
     Real life topic maps will consist of millions of topics and associations. Checking a map of such a size manually is clearly impossible, and yet checking is absolutely necessary for both proof-reading and quality assurance. It is obvious that both the designer and the editor need access to an automatic process that can validate a topic map against a set of consistency rules.
     The validation is the task of the topic map development environment (e.g. an editorial system). It should be performed permanently or on demand – like structure validation against the DTD in an SGML/XML editor.
     The standard has almost nothing to say on the subject of validation and consistency. The “Conformance” section of the standard focuses on the understanding of the defined constructs, the interchange syntax, and import/export of topic maps. But nothing more, as this excerpt from the standard shows:
     
     This International Standard constrains neither the uses to which topic maps can be put, nor the character of the processing that may be applied by a conforming application.
     This shows that we have to develop a schema language for the definition of the consistency constraints.
     

    Consistency constraints

     DTD, Document Type Definition 
         		The topic map standard provides the architectural element types which can be used in a derivedDTD . However, the degree to which semantics can be modelled in a DTD and through content models is rather limited. A topic map will consist of a large number of “independent” elements which are connected by links and not by element structures.
     Consequently a separate schema is needed which contains all the information necessary for the validation process. We call this constructconsistency constraints or justconstraints . The constraints are a set of predefined association types declared in the template.
     

    What should be validated?

     Constraints may be assigned to three potential layers:
     
  • topic map modeling,
    •  
  • user interface for topic maps, and
    •  
  • operations on the map.
    •  Here, we focus on the topic map modeling layer.
     

    Associations

     The most important candidates for validation are the associations. This is obvious because they are the key concept and carry a large number of parameters which might be "misused".
     The starting point is the association type. This controls which association role types can be combined. Beside the possible combination(s) the number of the various roles within these combinations might be of interest.
     The association role type in turn governs the set of topic types which may be referenced.
     It is necessary that the constraint schema brings the association type, the role type, and the topic type into a meaningful combination.
     An example:
     
    Association type is in (geographical containment)
    Valid association role types onecontainee: onecontainer
    Valid topic type combinations city: (country | state | county)
    county: (state | country)
    state: (country)
     

    Occurrences

     The assignment of the proper information resource types – if type information is provided by the editorial system – to the occurrence role types is also of interest as well as the meaningful combination of topic types and occurrence role types.
     An example:
     
    Topic type: person
    Valid occurrence role types: biography ,portrait
    Valid resource types forbiography: SGML/XML instance with public identifier "-//STEP//DTD biography//EN"
    Valid resource types forportrait: object types TIFF, GIF, JPEG
     

    Scopes

     Furthermore the correct use of scopes and especially the combination of different scopes might be checked.
     The topic type could restrict the possible scopes for the topics, their topic names, base name, display name, sort name, and their occurrences.
     Because assigning scopes to the topic or the topic name are just shortcuts for assignments to every name or occurrence, the set of scopes of the topic must be a superset of the scopes for the names and occurrences, and the set of scopes of the topic name must be a superset of the scopes for the individual names.
     The association types might restrict the meaningful scopes for the associations also. The combination of the meaningful scopes of the association and the referenced topics should be checked also because the association type is closely related to the possible types of the referenced topics.
     An example:
     
    Themes: before Einstein's theory of relativity ,after Einstein's theory of relativity
    Topic types: physical law ,mathematical axiom
    Occurrence role types: definition
    Constraints: The scopebefore Einstein's theory of relativity might be used for occurrences with roledefinition for topics of typephysical law ; but it must not be used fordefinitions ofmathematical axioms .
     

    Topic names

     For reasons of completeness checking of the topic names should also be possible. Topic names might be checked against text patterns or against database entries. The constraints will be governed by the topic type in question.
     An example:
     
    Topic types: component in assembly group ,chemical substance
    Constraints: Check base name of topic of typecomponent against pattern (regular expression) "P[0-9]+[A-D][E-G][0-5]"; check sort name ofchemical substance against table "substance names" in chemical database.
     All type combination constraints might limit the number of superclasses and/or subclasses of the affected types.
     

    Conclusions

     The new topic map standard ISO/IEC 13250 defines a model and architecture for the semantic structuring of link networks. It can be seen as a base technology for modeling knowledge structures. The standards working group defined topic maps in such a way that a limited but implementable set of core concepts express the necessary semantics.
     The STEP Group investigated how topic maps can be applied to reference works and uncovered some concepts which are not made explicit in the standard:
     
  • possibility to separate the declarative part from the “real” map,
    •  
  • predefined association types and association type properties,
    •  
  • class hierarchies for types, and
    •  
  • consistency constraints as input to map validation.
    •  The paper explained these concepts and presented meaningful solutions.
     First experiences have shown that the part of a topic map made up by all topics used as themes and types by other "objects" in the map should be clustered somehow. For this purpose the term "topic map template" was coined by the ISO working group. Templates can be used as starting points for new maps or can be used by reference in order to provide all the themes and types the map needs. Standardizing topic map templates will offer base topic maps for specific application areas and could form the basis of semantic application profiles.
     We looked at related academic fields like mathematics, linguistics, and philosophy to get some substantial input about relations. The results are a list of association type properties which give important hints to the topic map software and a list of basic association types which could act as built-in superclasses.
     The introduction of the superclass-subclass relationship was the logical consequence.
     Another technical issue covered by the paper is the validation problem. Topic maps might become rather big with millions of topics, occurrences, and associations. Manual consistency checking will be impossible. All the previously defined concepts open the possibility for sophisticated rule-based validation of topic maps. The proposed consistency constraints are those rules which declare the semantics not expressible with DTDs and which control the validation process.
     A couple of examples proved that standardizing the missing concepts as predefined topic map templates will help both the topic map developer and the topic map user. The improvements were presented on a level that they can be used as input to the ISO working group for further discussions.
     Acknowledgements
     The author would like to thank his colleagues Geir Ove Gr⊘nmo (STEP Infotek, Norway), Rafal Ksiezyk (STEP Poland), Graham Moore (STEP UK), and Steve Pepper (STEP Infotek, Norway), as well as all the members of STEP's Reference Works Module Club – leading European reference works publishers – for their input and open discussions about topic maps.
     Bibliography
     
    AiGiBa97 Aitchison, J.; Gilchrist, A.; Bawden, D.:Thesaurus construction and use – a practical manual , 3rd edition, London: Aslib, 1997.
     
    ChHeWi88 Chaffin, R.; Hermann, D.J.; Winston, M.:An empirical taxonomy of part-whole relations: Effects of part-whole relation type on relation identification ,Language and Cognitive Process 3 , 1988.
     
    Fel98 Fellbaum, C. (ed):WordNet – An Electronic Lexical Database , MIT Press, 1998.
     
    ISO13250 International Organization for Standardization:ISO/IEC 13250:2000, Information technology – SGML Applications – Topic Maps , Geneva: ISO, 2000.
     
    ISO10744 International Organization for Standardization:ISO/IEC 10744:1999 Information technology – Hypermedia/Time-based Structuring Language (HyTime) , Geneva: ISO, 1997.
     
    ISO2788 International Organization for Standardization:ISO 2788:1986. Guidelines for the establishment and development of monolingual thesauri , Geneva: ISO, 1986.
     
    ISO5964 International Organization for Standardization:ISO 5964:1985. Guidelines for the establishment and development of multilingual thesauri , Geneva: ISO, 1985.
     
    IrLiEv88 Iris, M.; Litowitz. B.; Evens, M.:Problems of the part-whole relation , in: Evens, M. (ed):Relational models of the lexicon , Cambridge, 1988.
     
    Ksi99 Ksiezyk, R.:Trying not to get lost with a Topic Map , in:Proceedings of XML Europe 99 Conference , GCA, Alexandria, VA, 1999.
     
    Meg98 Megginson, D.:Structuring XML Documents , Prentice Hall, 1998.
     
    Pep99a Pepper, S.:Euler, Topic Maps, and Revolution , in:Proceedings of XML Europe 99 Conference , GCA, Alexandria, VA, 1999.
     
    Pep99b Pepper, S.:Navigating Haystacks, Discovering Needles , in: Sperberg-McQueen, C.M.; Usdin, B.T. (eds):Markup Languages , Vol 1 No 4, MIT Press, Cambridge (MA), 1999.
     
    Ran67 Ranganathan, S.R.:Prolegomena to Library Classification , Bombay: Asia Publishing House, 1967.
     
    RaPe99a Rath, H.H.; Pepper, S.:Topic maps: Knowldege navigation aids , in: Goldfarb, C.F., Prescod, P. (eds):XML Handbook , 2nd edition, Prentice Hall, 1999.
     
    RaPe99b Rath, H.H.; Pepper, S.:Topic Maps: Introduction and Allegro , in:Proceedings of Markup Technologies 99 Conference , GCA, Alexandria, VA, 1999.
     
    Rat99a Rath, H.H.:Mit Topic Maps intelligente Informationsnetze aufbauen – Mozart und Kugeln (German), in:iX Magazin , Dec. 1999, Heise Verlag, Hannover, Germany, 1999.
     
    Rat99b Rath, H.H.:Technical Issues on Topic Maps , in:Proceedings of Metastructures 99 Conference , GCA, Alexandria, VA, 1999.
     
    Rat00 Rath, H.H.:Topic Maps: Templates, topology, and type hierarchies , to appear in: Sperberg-McQueen, C.M.; Usdin, B.T. (eds):Markup Languages , MIT Press, Cambridge (MA), 2000.
     
    RiDu88 Ringland, G.A.; Duce, D.A.:Approaches to Knowledge Representation: An Introduction , Research Studies Press/John Wiley 1988.
     
    Rug97 Ruggles, R.L. (ed):Knowledge management tools , Boston: Butterworth-Heinemann, 1997.
     
    Str99 Streich, R.:Techniques for managing collections of interrelated text modules , in: Sperberg-McQueen, C.M.; Usdin, B.T. (eds):Markup Languages , Vol 1 No 2, MIT Press, Cambridge (MA), 1999.
     
    Vic60 Vickery, B.C.:Faceted classification: a guide to construction and use of special schemes , London: Aslib, 1960.
     
    Vic66 Vickery, B.C.:Faceted classification schemes , New Brunswick: Rutgers, 1966.
     
    WiChHe87 Winston, M.E.; Chaffin R.; Hermann, D.:A taxonomy of part-whole relations , Cognitive Science 11, 1987.
     
    Wordnet WordNet:A Lexical Database for English , Cognitive Science Laboratory, Princeton University, Princeton, NJ, http://www.cogsci.princeton.edu/~wn/.
     
    Z3919 ANSI/NISO:Z39.19. Guidelines for the construction, format and management of monolingual thesauri , Bethesda: ANSI/NISO 1993.
    Topic Map technology - the state of the art   Table of contents   Indexes   Creating semantically valid topic maps